2008-11-20 20:01:55 +00:00
|
|
|
/*
|
|
|
|
* CDDL HEADER START
|
|
|
|
*
|
|
|
|
* The contents of this file are subject to the terms of the
|
|
|
|
* Common Development and Distribution License (the "License").
|
|
|
|
* You may not use this file except in compliance with the License.
|
|
|
|
*
|
|
|
|
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
|
|
|
|
* or http://www.opensolaris.org/os/licensing.
|
|
|
|
* See the License for the specific language governing permissions
|
|
|
|
* and limitations under the License.
|
|
|
|
*
|
|
|
|
* When distributing Covered Code, include this CDDL HEADER in each
|
|
|
|
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
|
|
|
|
* If applicable, add the following below this CDDL HEADER, with the
|
|
|
|
* fields enclosed by brackets "[]" replaced with your own identifying
|
|
|
|
* information: Portions Copyright [yyyy] [name of copyright owner]
|
|
|
|
*
|
|
|
|
* CDDL HEADER END
|
|
|
|
*/
|
|
|
|
/*
|
2010-05-28 20:45:14 +00:00
|
|
|
* Copyright 2010 Sun Microsystems, Inc. All rights reserved.
|
2008-11-20 20:01:55 +00:00
|
|
|
* Use is subject to license terms.
|
|
|
|
*/
|
|
|
|
|
2012-01-24 02:43:32 +00:00
|
|
|
/*
|
2016-10-14 00:59:18 +00:00
|
|
|
* Copyright (c) 2012, 2015 by Delphix. All rights reserved.
|
2012-01-24 02:43:32 +00:00
|
|
|
*/
|
|
|
|
|
2008-11-20 20:01:55 +00:00
|
|
|
#include <sys/zfs_context.h>
|
|
|
|
#include <sys/spa.h>
|
|
|
|
#include <sys/vdev_impl.h>
|
|
|
|
#include <sys/zio.h>
|
|
|
|
#include <sys/fs/zfs.h>
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Virtual device vector for mirroring.
|
|
|
|
*/
|
|
|
|
|
|
|
|
typedef struct mirror_child {
|
|
|
|
vdev_t *mc_vd;
|
|
|
|
uint64_t mc_offset;
|
|
|
|
int mc_error;
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
int mc_load;
|
2008-12-03 20:09:06 +00:00
|
|
|
uint8_t mc_tried;
|
|
|
|
uint8_t mc_skipped;
|
|
|
|
uint8_t mc_speculative;
|
2008-11-20 20:01:55 +00:00
|
|
|
} mirror_child_t;
|
|
|
|
|
|
|
|
typedef struct mirror_map {
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
int *mm_preferred;
|
|
|
|
int mm_preferred_cnt;
|
2008-11-20 20:01:55 +00:00
|
|
|
int mm_children;
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
boolean_t mm_replacing;
|
|
|
|
boolean_t mm_root;
|
|
|
|
mirror_child_t mm_child[];
|
2008-11-20 20:01:55 +00:00
|
|
|
} mirror_map_t;
|
|
|
|
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
static int vdev_mirror_shift = 21;
|
|
|
|
|
2013-05-31 19:07:59 +00:00
|
|
|
/*
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
* The load configuration settings below are tuned by default for
|
|
|
|
* the case where all devices are of the same rotational type.
|
2013-05-31 19:07:59 +00:00
|
|
|
*
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
* If there is a mixture of rotating and non-rotating media, setting
|
|
|
|
* zfs_vdev_mirror_non_rotating_seek_inc to 0 may well provide better results
|
|
|
|
* as it will direct more reads to the non-rotating vdevs which are more likely
|
|
|
|
* to have a higher performance.
|
2013-05-31 19:07:59 +00:00
|
|
|
*/
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
|
|
|
|
/* Rotating media load calculation configuration. */
|
|
|
|
static int zfs_vdev_mirror_rotating_inc = 0;
|
|
|
|
static int zfs_vdev_mirror_rotating_seek_inc = 5;
|
|
|
|
static int zfs_vdev_mirror_rotating_seek_offset = 1 * 1024 * 1024;
|
|
|
|
|
|
|
|
/* Non-rotating media load calculation configuration. */
|
|
|
|
static int zfs_vdev_mirror_non_rotating_inc = 0;
|
|
|
|
static int zfs_vdev_mirror_non_rotating_seek_inc = 1;
|
|
|
|
|
|
|
|
static inline size_t
|
|
|
|
vdev_mirror_map_size(int children)
|
|
|
|
{
|
|
|
|
return (offsetof(mirror_map_t, mm_child[children]) +
|
|
|
|
sizeof (int) * children);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline mirror_map_t *
|
|
|
|
vdev_mirror_map_alloc(int children, boolean_t replacing, boolean_t root)
|
|
|
|
{
|
|
|
|
mirror_map_t *mm;
|
|
|
|
|
|
|
|
mm = kmem_zalloc(vdev_mirror_map_size(children), KM_SLEEP);
|
|
|
|
mm->mm_children = children;
|
|
|
|
mm->mm_replacing = replacing;
|
|
|
|
mm->mm_root = root;
|
|
|
|
mm->mm_preferred = (int *)((uintptr_t)mm +
|
|
|
|
offsetof(mirror_map_t, mm_child[children]));
|
|
|
|
|
|
|
|
return (mm);
|
|
|
|
}
|
2008-11-20 20:01:55 +00:00
|
|
|
|
2008-12-03 20:09:06 +00:00
|
|
|
static void
|
|
|
|
vdev_mirror_map_free(zio_t *zio)
|
|
|
|
{
|
|
|
|
mirror_map_t *mm = zio->io_vsd;
|
|
|
|
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
kmem_free(mm, vdev_mirror_map_size(mm->mm_children));
|
2008-12-03 20:09:06 +00:00
|
|
|
}
|
|
|
|
|
2010-05-28 20:45:14 +00:00
|
|
|
static const zio_vsd_ops_t vdev_mirror_vsd_ops = {
|
|
|
|
vdev_mirror_map_free,
|
|
|
|
zio_vsd_default_cksum_report
|
|
|
|
};
|
|
|
|
|
2013-05-31 19:07:59 +00:00
|
|
|
static int
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
vdev_mirror_load(mirror_map_t *mm, vdev_t *vd, uint64_t zio_offset)
|
2013-05-31 19:07:59 +00:00
|
|
|
{
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
uint64_t lastoffset;
|
|
|
|
int load;
|
|
|
|
|
|
|
|
/* All DVAs have equal weight at the root. */
|
|
|
|
if (mm->mm_root)
|
|
|
|
return (INT_MAX);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We don't return INT_MAX if the device is resilvering i.e.
|
|
|
|
* vdev_resilver_txg != 0 as when tested performance was slightly
|
|
|
|
* worse overall when resilvering with compared to without.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* Standard load based on pending queue length. */
|
|
|
|
load = vdev_queue_length(vd);
|
|
|
|
lastoffset = vdev_queue_lastoffset(vd);
|
|
|
|
|
|
|
|
if (vd->vdev_nonrot) {
|
|
|
|
/* Non-rotating media. */
|
|
|
|
if (lastoffset == zio_offset)
|
|
|
|
return (load + zfs_vdev_mirror_non_rotating_inc);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Apply a seek penalty even for non-rotating devices as
|
|
|
|
* sequential I/O's can be aggregated into fewer operations on
|
|
|
|
* the device, thus avoiding unnecessary per-command overhead
|
|
|
|
* and boosting performance.
|
|
|
|
*/
|
|
|
|
return (load + zfs_vdev_mirror_non_rotating_seek_inc);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Rotating media I/O's which directly follow the last I/O. */
|
|
|
|
if (lastoffset == zio_offset)
|
|
|
|
return (load + zfs_vdev_mirror_rotating_inc);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Apply half the seek increment to I/O's within seek offset
|
|
|
|
* of the last I/O queued to this vdev as they should incure less
|
|
|
|
* of a seek increment.
|
|
|
|
*/
|
|
|
|
if (ABS(lastoffset - zio_offset) <
|
|
|
|
zfs_vdev_mirror_rotating_seek_offset)
|
|
|
|
return (load + (zfs_vdev_mirror_rotating_seek_inc / 2));
|
|
|
|
|
|
|
|
/* Apply the full seek increment to all other I/O's. */
|
|
|
|
return (load + zfs_vdev_mirror_rotating_seek_inc);
|
2013-05-31 19:07:59 +00:00
|
|
|
}
|
|
|
|
|
2013-11-13 19:05:17 +00:00
|
|
|
/*
|
|
|
|
* Avoid inlining the function to keep vdev_mirror_io_start(), which
|
|
|
|
* is this functions only caller, as small as possible on the stack.
|
|
|
|
*/
|
|
|
|
noinline static mirror_map_t *
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
vdev_mirror_map_init(zio_t *zio)
|
2008-11-20 20:01:55 +00:00
|
|
|
{
|
|
|
|
mirror_map_t *mm = NULL;
|
|
|
|
mirror_child_t *mc;
|
|
|
|
vdev_t *vd = zio->io_vd;
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
int c;
|
2008-11-20 20:01:55 +00:00
|
|
|
|
|
|
|
if (vd == NULL) {
|
|
|
|
dva_t *dva = zio->io_bp->blk_dva;
|
|
|
|
spa_t *spa = zio->io_spa;
|
|
|
|
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
mm = vdev_mirror_map_alloc(BP_GET_NDVAS(zio->io_bp), B_FALSE,
|
|
|
|
B_TRUE);
|
2008-11-20 20:01:55 +00:00
|
|
|
for (c = 0; c < mm->mm_children; c++) {
|
|
|
|
mc = &mm->mm_child[c];
|
|
|
|
|
|
|
|
mc->mc_vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[c]));
|
|
|
|
mc->mc_offset = DVA_GET_OFFSET(&dva[c]);
|
|
|
|
}
|
|
|
|
} else {
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
mm = vdev_mirror_map_alloc(vd->vdev_children,
|
|
|
|
(vd->vdev_ops == &vdev_replacing_ops ||
|
|
|
|
vd->vdev_ops == &vdev_spare_ops), B_FALSE);
|
2008-11-20 20:01:55 +00:00
|
|
|
for (c = 0; c < mm->mm_children; c++) {
|
|
|
|
mc = &mm->mm_child[c];
|
|
|
|
mc->mc_vd = vd->vdev_child[c];
|
|
|
|
mc->mc_offset = zio->io_offset;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
zio->io_vsd = mm;
|
2010-05-28 20:45:14 +00:00
|
|
|
zio->io_vsd_ops = &vdev_mirror_vsd_ops;
|
2008-11-20 20:01:55 +00:00
|
|
|
return (mm);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
2012-01-24 02:43:32 +00:00
|
|
|
vdev_mirror_open(vdev_t *vd, uint64_t *asize, uint64_t *max_asize,
|
|
|
|
uint64_t *ashift)
|
2008-11-20 20:01:55 +00:00
|
|
|
{
|
|
|
|
int numerrors = 0;
|
2009-08-18 18:43:27 +00:00
|
|
|
int lasterror = 0;
|
2010-08-26 16:52:39 +00:00
|
|
|
int c;
|
2008-11-20 20:01:55 +00:00
|
|
|
|
|
|
|
if (vd->vdev_children == 0) {
|
|
|
|
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
|
2013-03-08 18:41:28 +00:00
|
|
|
return (SET_ERROR(EINVAL));
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
2009-08-18 18:43:27 +00:00
|
|
|
vdev_open_children(vd);
|
2008-11-20 20:01:55 +00:00
|
|
|
|
2010-08-26 16:52:39 +00:00
|
|
|
for (c = 0; c < vd->vdev_children; c++) {
|
2009-08-18 18:43:27 +00:00
|
|
|
vdev_t *cvd = vd->vdev_child[c];
|
|
|
|
|
|
|
|
if (cvd->vdev_open_error) {
|
|
|
|
lasterror = cvd->vdev_open_error;
|
2008-11-20 20:01:55 +00:00
|
|
|
numerrors++;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
*asize = MIN(*asize - 1, cvd->vdev_asize - 1) + 1;
|
2012-01-24 02:43:32 +00:00
|
|
|
*max_asize = MIN(*max_asize - 1, cvd->vdev_max_asize - 1) + 1;
|
2008-11-20 20:01:55 +00:00
|
|
|
*ashift = MAX(*ashift, cvd->vdev_ashift);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (numerrors == vd->vdev_children) {
|
|
|
|
vd->vdev_stat.vs_aux = VDEV_AUX_NO_REPLICAS;
|
|
|
|
return (lasterror);
|
|
|
|
}
|
|
|
|
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
vdev_mirror_close(vdev_t *vd)
|
|
|
|
{
|
2010-08-26 16:52:39 +00:00
|
|
|
int c;
|
|
|
|
|
|
|
|
for (c = 0; c < vd->vdev_children; c++)
|
2008-11-20 20:01:55 +00:00
|
|
|
vdev_close(vd->vdev_child[c]);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
vdev_mirror_child_done(zio_t *zio)
|
|
|
|
{
|
|
|
|
mirror_child_t *mc = zio->io_private;
|
|
|
|
|
|
|
|
mc->mc_error = zio->io_error;
|
|
|
|
mc->mc_tried = 1;
|
|
|
|
mc->mc_skipped = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
vdev_mirror_scrub_done(zio_t *zio)
|
|
|
|
{
|
|
|
|
mirror_child_t *mc = zio->io_private;
|
|
|
|
|
|
|
|
if (zio->io_error == 0) {
|
2009-02-18 20:51:31 +00:00
|
|
|
zio_t *pio;
|
2016-10-14 00:59:18 +00:00
|
|
|
zio_link_t *zl = NULL;
|
2009-02-18 20:51:31 +00:00
|
|
|
|
|
|
|
mutex_enter(&zio->io_lock);
|
2016-10-14 00:59:18 +00:00
|
|
|
while ((pio = zio_walk_parents(zio, &zl)) != NULL) {
|
2009-02-18 20:51:31 +00:00
|
|
|
mutex_enter(&pio->io_lock);
|
|
|
|
ASSERT3U(zio->io_size, >=, pio->io_size);
|
|
|
|
bcopy(zio->io_data, pio->io_data, pio->io_size);
|
|
|
|
mutex_exit(&pio->io_lock);
|
|
|
|
}
|
|
|
|
mutex_exit(&zio->io_lock);
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
zio_buf_free(zio->io_data, zio->io_size);
|
|
|
|
|
|
|
|
mc->mc_error = zio->io_error;
|
|
|
|
mc->mc_tried = 1;
|
|
|
|
mc->mc_skipped = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
* Check the other, lower-index DVAs to see if they're on the same
|
|
|
|
* vdev as the child we picked. If they are, use them since they
|
|
|
|
* are likely to have been allocated from the primary metaslab in
|
|
|
|
* use at the time, and hence are more likely to have locality with
|
|
|
|
* single-copy data.
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
vdev_mirror_dva_select(zio_t *zio, int p)
|
|
|
|
{
|
|
|
|
dva_t *dva = zio->io_bp->blk_dva;
|
|
|
|
mirror_map_t *mm = zio->io_vsd;
|
|
|
|
int preferred;
|
|
|
|
int c;
|
|
|
|
|
|
|
|
preferred = mm->mm_preferred[p];
|
|
|
|
for (p--; p >= 0; p--) {
|
|
|
|
c = mm->mm_preferred[p];
|
|
|
|
if (DVA_GET_VDEV(&dva[c]) == DVA_GET_VDEV(&dva[preferred]))
|
|
|
|
preferred = c;
|
|
|
|
}
|
|
|
|
return (preferred);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
vdev_mirror_preferred_child_randomize(zio_t *zio)
|
|
|
|
{
|
|
|
|
mirror_map_t *mm = zio->io_vsd;
|
|
|
|
int p;
|
|
|
|
|
|
|
|
if (mm->mm_root) {
|
|
|
|
p = spa_get_random(mm->mm_preferred_cnt);
|
|
|
|
return (vdev_mirror_dva_select(zio, p));
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* To ensure we don't always favour the first matching vdev,
|
|
|
|
* which could lead to wear leveling issues on SSD's, we
|
|
|
|
* use the I/O offset as a pseudo random seed into the vdevs
|
|
|
|
* which have the lowest load.
|
|
|
|
*/
|
|
|
|
p = (zio->io_offset >> vdev_mirror_shift) % mm->mm_preferred_cnt;
|
|
|
|
return (mm->mm_preferred[p]);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Try to find a vdev whose DTL doesn't contain the block we want to read
|
|
|
|
* prefering vdevs based on determined load.
|
|
|
|
*
|
2008-11-20 20:01:55 +00:00
|
|
|
* Try to find a child whose DTL doesn't contain the block we want to read.
|
|
|
|
* If we can't, try the read on any vdev we haven't already tried.
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
vdev_mirror_child_select(zio_t *zio)
|
|
|
|
{
|
|
|
|
mirror_map_t *mm = zio->io_vsd;
|
|
|
|
uint64_t txg = zio->io_txg;
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
int c, lowest_load;
|
2008-11-20 20:01:55 +00:00
|
|
|
|
2010-05-28 20:45:14 +00:00
|
|
|
ASSERT(zio->io_bp == NULL || BP_PHYSICAL_BIRTH(zio->io_bp) == txg);
|
2008-11-20 20:01:55 +00:00
|
|
|
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
lowest_load = INT_MAX;
|
|
|
|
mm->mm_preferred_cnt = 0;
|
|
|
|
for (c = 0; c < mm->mm_children; c++) {
|
|
|
|
mirror_child_t *mc;
|
|
|
|
|
2008-11-20 20:01:55 +00:00
|
|
|
mc = &mm->mm_child[c];
|
|
|
|
if (mc->mc_tried || mc->mc_skipped)
|
|
|
|
continue;
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
|
2014-10-11 01:12:47 +00:00
|
|
|
if (mc->mc_vd == NULL || !vdev_readable(mc->mc_vd)) {
|
2013-03-08 18:41:28 +00:00
|
|
|
mc->mc_error = SET_ERROR(ENXIO);
|
2008-11-20 20:01:55 +00:00
|
|
|
mc->mc_tried = 1; /* don't even try */
|
|
|
|
mc->mc_skipped = 1;
|
|
|
|
continue;
|
|
|
|
}
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
|
|
|
|
if (vdev_dtl_contains(mc->mc_vd, DTL_MISSING, txg, 1)) {
|
|
|
|
mc->mc_error = SET_ERROR(ESTALE);
|
|
|
|
mc->mc_skipped = 1;
|
|
|
|
mc->mc_speculative = 1;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
mc->mc_load = vdev_mirror_load(mm, mc->mc_vd, mc->mc_offset);
|
|
|
|
if (mc->mc_load > lowest_load)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (mc->mc_load < lowest_load) {
|
|
|
|
lowest_load = mc->mc_load;
|
|
|
|
mm->mm_preferred_cnt = 0;
|
|
|
|
}
|
|
|
|
mm->mm_preferred[mm->mm_preferred_cnt] = c;
|
|
|
|
mm->mm_preferred_cnt++;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (mm->mm_preferred_cnt == 1) {
|
|
|
|
vdev_queue_register_lastoffset(
|
|
|
|
mm->mm_child[mm->mm_preferred[0]].mc_vd, zio);
|
|
|
|
return (mm->mm_preferred[0]);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (mm->mm_preferred_cnt > 1) {
|
|
|
|
int c = vdev_mirror_preferred_child_randomize(zio);
|
|
|
|
|
|
|
|
vdev_queue_register_lastoffset(mm->mm_child[c].mc_vd, zio);
|
|
|
|
return (c);
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Every device is either missing or has this txg in its DTL.
|
|
|
|
* Look for any child we haven't already tried before giving up.
|
|
|
|
*/
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
for (c = 0; c < mm->mm_children; c++) {
|
|
|
|
if (!mm->mm_child[c].mc_tried) {
|
|
|
|
vdev_queue_register_lastoffset(mm->mm_child[c].mc_vd,
|
|
|
|
zio);
|
2008-11-20 20:01:55 +00:00
|
|
|
return (c);
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
}
|
|
|
|
}
|
2008-11-20 20:01:55 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Every child failed. There's no place left to look.
|
|
|
|
*/
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
|
2014-10-20 22:07:45 +00:00
|
|
|
static void
|
2008-11-20 20:01:55 +00:00
|
|
|
vdev_mirror_io_start(zio_t *zio)
|
|
|
|
{
|
|
|
|
mirror_map_t *mm;
|
|
|
|
mirror_child_t *mc;
|
|
|
|
int c, children;
|
|
|
|
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
mm = vdev_mirror_map_init(zio);
|
2008-11-20 20:01:55 +00:00
|
|
|
|
|
|
|
if (zio->io_type == ZIO_TYPE_READ) {
|
|
|
|
if ((zio->io_flags & ZIO_FLAG_SCRUB) && !mm->mm_replacing) {
|
|
|
|
/*
|
|
|
|
* For scrubbing reads we need to allocate a read
|
|
|
|
* buffer for each child and issue reads to all
|
|
|
|
* children. If any child succeeds, it will copy its
|
|
|
|
* data into zio->io_data in vdev_mirror_scrub_done.
|
|
|
|
*/
|
|
|
|
for (c = 0; c < mm->mm_children; c++) {
|
|
|
|
mc = &mm->mm_child[c];
|
|
|
|
zio_nowait(zio_vdev_child_io(zio, zio->io_bp,
|
|
|
|
mc->mc_vd, mc->mc_offset,
|
|
|
|
zio_buf_alloc(zio->io_size), zio->io_size,
|
2008-12-03 20:09:06 +00:00
|
|
|
zio->io_type, zio->io_priority, 0,
|
2008-11-20 20:01:55 +00:00
|
|
|
vdev_mirror_scrub_done, mc));
|
|
|
|
}
|
2014-10-20 22:07:45 +00:00
|
|
|
zio_execute(zio);
|
|
|
|
return;
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
/*
|
|
|
|
* For normal reads just pick one child.
|
|
|
|
*/
|
|
|
|
c = vdev_mirror_child_select(zio);
|
|
|
|
children = (c >= 0);
|
|
|
|
} else {
|
|
|
|
ASSERT(zio->io_type == ZIO_TYPE_WRITE);
|
|
|
|
|
|
|
|
/*
|
2009-01-15 21:59:39 +00:00
|
|
|
* Writes go to all children.
|
2008-11-20 20:01:55 +00:00
|
|
|
*/
|
2009-01-15 21:59:39 +00:00
|
|
|
c = 0;
|
|
|
|
children = mm->mm_children;
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
while (children--) {
|
|
|
|
mc = &mm->mm_child[c];
|
|
|
|
zio_nowait(zio_vdev_child_io(zio, zio->io_bp,
|
2008-12-03 20:09:06 +00:00
|
|
|
mc->mc_vd, mc->mc_offset, zio->io_data, zio->io_size,
|
|
|
|
zio->io_type, zio->io_priority, 0,
|
|
|
|
vdev_mirror_child_done, mc));
|
2008-11-20 20:01:55 +00:00
|
|
|
c++;
|
|
|
|
}
|
|
|
|
|
2014-10-20 22:07:45 +00:00
|
|
|
zio_execute(zio);
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
2008-12-03 20:09:06 +00:00
|
|
|
vdev_mirror_worst_error(mirror_map_t *mm)
|
|
|
|
{
|
2010-08-26 16:52:39 +00:00
|
|
|
int c, error[2] = { 0, 0 };
|
2008-12-03 20:09:06 +00:00
|
|
|
|
2010-08-26 16:52:39 +00:00
|
|
|
for (c = 0; c < mm->mm_children; c++) {
|
2008-12-03 20:09:06 +00:00
|
|
|
mirror_child_t *mc = &mm->mm_child[c];
|
|
|
|
int s = mc->mc_speculative;
|
|
|
|
error[s] = zio_worst_error(error[s], mc->mc_error);
|
|
|
|
}
|
|
|
|
|
|
|
|
return (error[0] ? error[0] : error[1]);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
2008-11-20 20:01:55 +00:00
|
|
|
vdev_mirror_io_done(zio_t *zio)
|
|
|
|
{
|
|
|
|
mirror_map_t *mm = zio->io_vsd;
|
|
|
|
mirror_child_t *mc;
|
|
|
|
int c;
|
|
|
|
int good_copies = 0;
|
|
|
|
int unexpected_errors = 0;
|
|
|
|
|
|
|
|
for (c = 0; c < mm->mm_children; c++) {
|
|
|
|
mc = &mm->mm_child[c];
|
|
|
|
|
|
|
|
if (mc->mc_error) {
|
|
|
|
if (!mc->mc_skipped)
|
|
|
|
unexpected_errors++;
|
2008-12-03 20:09:06 +00:00
|
|
|
} else if (mc->mc_tried) {
|
|
|
|
good_copies++;
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (zio->io_type == ZIO_TYPE_WRITE) {
|
|
|
|
/*
|
|
|
|
* XXX -- for now, treat partial writes as success.
|
2008-12-03 20:09:06 +00:00
|
|
|
*
|
|
|
|
* Now that we support write reallocation, it would be better
|
|
|
|
* to treat partial failure as real failure unless there are
|
|
|
|
* no non-degraded top-level vdevs left, and not update DTLs
|
|
|
|
* if we intend to reallocate.
|
2008-11-20 20:01:55 +00:00
|
|
|
*/
|
|
|
|
/* XXPOLICY */
|
2008-12-03 20:09:06 +00:00
|
|
|
if (good_copies != mm->mm_children) {
|
|
|
|
/*
|
|
|
|
* Always require at least one good copy.
|
|
|
|
*
|
|
|
|
* For ditto blocks (io_vd == NULL), require
|
|
|
|
* all copies to be good.
|
|
|
|
*
|
|
|
|
* XXX -- for replacing vdevs, there's no great answer.
|
|
|
|
* If the old device is really dead, we may not even
|
|
|
|
* be able to access it -- so we only want to
|
|
|
|
* require good writes to the new device. But if
|
|
|
|
* the new device turns out to be flaky, we want
|
|
|
|
* to be able to detach it -- which requires all
|
|
|
|
* writes to the old device to have succeeded.
|
|
|
|
*/
|
|
|
|
if (good_copies == 0 || zio->io_vd == NULL)
|
|
|
|
zio->io_error = vdev_mirror_worst_error(mm);
|
|
|
|
}
|
|
|
|
return;
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
ASSERT(zio->io_type == ZIO_TYPE_READ);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If we don't have a good copy yet, keep trying other children.
|
|
|
|
*/
|
|
|
|
/* XXPOLICY */
|
|
|
|
if (good_copies == 0 && (c = vdev_mirror_child_select(zio)) != -1) {
|
|
|
|
ASSERT(c >= 0 && c < mm->mm_children);
|
|
|
|
mc = &mm->mm_child[c];
|
|
|
|
zio_vdev_io_redone(zio);
|
|
|
|
zio_nowait(zio_vdev_child_io(zio, zio->io_bp,
|
|
|
|
mc->mc_vd, mc->mc_offset, zio->io_data, zio->io_size,
|
2008-12-03 20:09:06 +00:00
|
|
|
ZIO_TYPE_READ, zio->io_priority, 0,
|
2008-11-20 20:01:55 +00:00
|
|
|
vdev_mirror_child_done, mc));
|
2008-12-03 20:09:06 +00:00
|
|
|
return;
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* XXPOLICY */
|
2008-12-03 20:09:06 +00:00
|
|
|
if (good_copies == 0) {
|
|
|
|
zio->io_error = vdev_mirror_worst_error(mm);
|
2008-11-20 20:01:55 +00:00
|
|
|
ASSERT(zio->io_error != 0);
|
2008-12-03 20:09:06 +00:00
|
|
|
}
|
2008-11-20 20:01:55 +00:00
|
|
|
|
2009-01-15 21:59:39 +00:00
|
|
|
if (good_copies && spa_writeable(zio->io_spa) &&
|
2008-11-20 20:01:55 +00:00
|
|
|
(unexpected_errors ||
|
|
|
|
(zio->io_flags & ZIO_FLAG_RESILVER) ||
|
|
|
|
((zio->io_flags & ZIO_FLAG_SCRUB) && mm->mm_replacing))) {
|
|
|
|
/*
|
|
|
|
* Use the good data we have in hand to repair damaged children.
|
|
|
|
*/
|
|
|
|
for (c = 0; c < mm->mm_children; c++) {
|
|
|
|
/*
|
|
|
|
* Don't rewrite known good children.
|
|
|
|
* Not only is it unnecessary, it could
|
|
|
|
* actually be harmful: if the system lost
|
|
|
|
* power while rewriting the only good copy,
|
|
|
|
* there would be no good copies left!
|
|
|
|
*/
|
|
|
|
mc = &mm->mm_child[c];
|
|
|
|
|
|
|
|
if (mc->mc_error == 0) {
|
|
|
|
if (mc->mc_tried)
|
|
|
|
continue;
|
|
|
|
if (!(zio->io_flags & ZIO_FLAG_SCRUB) &&
|
2009-01-15 21:59:39 +00:00
|
|
|
!vdev_dtl_contains(mc->mc_vd, DTL_PARTIAL,
|
2008-11-20 20:01:55 +00:00
|
|
|
zio->io_txg, 1))
|
|
|
|
continue;
|
2013-03-08 18:41:28 +00:00
|
|
|
mc->mc_error = SET_ERROR(ESTALE);
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
2008-12-03 20:09:06 +00:00
|
|
|
zio_nowait(zio_vdev_child_io(zio, zio->io_bp,
|
|
|
|
mc->mc_vd, mc->mc_offset,
|
|
|
|
zio->io_data, zio->io_size,
|
Illumos #4045 write throttle & i/o scheduler performance work
4045 zfs write throttle & i/o scheduler performance work
1. The ZFS i/o scheduler (vdev_queue.c) now divides i/os into 5 classes: sync
read, sync write, async read, async write, and scrub/resilver. The scheduler
issues a number of concurrent i/os from each class to the device. Once a class
has been selected, an i/o is selected from this class using either an elevator
algorithem (async, scrub classes) or FIFO (sync classes). The number of
concurrent async write i/os is tuned dynamically based on i/o load, to achieve
good sync i/o latency when there is not a high load of writes, and good write
throughput when there is. See the block comment in vdev_queue.c (reproduced
below) for more details.
2. The write throttle (dsl_pool_tempreserve_space() and
txg_constrain_throughput()) is rewritten to produce much more consistent delays
when under constant load. The new write throttle is based on the amount of
dirty data, rather than guesses about future performance of the system. When
there is a lot of dirty data, each transaction (e.g. write() syscall) will be
delayed by the same small amount. This eliminates the "brick wall of wait"
that the old write throttle could hit, causing all transactions to wait several
seconds until the next txg opens. One of the keys to the new write throttle is
decrementing the amount of dirty data as i/o completes, rather than at the end
of spa_sync(). Note that the write throttle is only applied once the i/o
scheduler is issuing the maximum number of outstanding async writes. See the
block comments in dsl_pool.c and above dmu_tx_delay() (reproduced below) for
more details.
This diff has several other effects, including:
* the commonly-tuned global variable zfs_vdev_max_pending has been removed;
use per-class zfs_vdev_*_max_active values or zfs_vdev_max_active instead.
* the size of each txg (meaning the amount of dirty data written, and thus the
time it takes to write out) is now controlled differently. There is no longer
an explicit time goal; the primary determinant is amount of dirty data.
Systems that are under light or medium load will now often see that a txg is
always syncing, but the impact to performance (e.g. read latency) is minimal.
Tune zfs_dirty_data_max and zfs_dirty_data_sync to control this.
* zio_taskq_batch_pct = 75 -- Only use 75% of all CPUs for compression,
checksum, etc. This improves latency by not allowing these CPU-intensive tasks
to consume all CPU (on machines with at least 4 CPU's; the percentage is
rounded up).
--matt
APPENDIX: problems with the current i/o scheduler
The current ZFS i/o scheduler (vdev_queue.c) is deadline based. The problem
with this is that if there are always i/os pending, then certain classes of
i/os can see very long delays.
For example, if there are always synchronous reads outstanding, then no async
writes will be serviced until they become "past due". One symptom of this
situation is that each pass of the txg sync takes at least several seconds
(typically 3 seconds).
If many i/os become "past due" (their deadline is in the past), then we must
service all of these overdue i/os before any new i/os. This happens when we
enqueue a batch of async writes for the txg sync, with deadlines 2.5 seconds in
the future. If we can't complete all the i/os in 2.5 seconds (e.g. because
there were always reads pending), then these i/os will become past due. Now we
must service all the "async" writes (which could be hundreds of megabytes)
before we service any reads, introducing considerable latency to synchronous
i/os (reads or ZIL writes).
Notes on porting to ZFS on Linux:
- zio_t gained new members io_physdone and io_phys_children. Because
object caches in the Linux port call the constructor only once at
allocation time, objects may contain residual data when retrieved
from the cache. Therefore zio_create() was updated to zero out the two
new fields.
- vdev_mirror_pending() relied on the depth of the per-vdev pending queue
(vq->vq_pending_tree) to select the least-busy leaf vdev to read from.
This tree has been replaced by vq->vq_active_tree which is now used
for the same purpose.
- vdev_queue_init() used the value of zfs_vdev_max_pending to determine
the number of vdev I/O buffers to pre-allocate. That global no longer
exists, so we instead use the sum of the *_max_active values for each of
the five I/O classes described above.
- The Illumos implementation of dmu_tx_delay() delays a transaction by
sleeping in condition variable embedded in the thread
(curthread->t_delay_cv). We do not have an equivalent CV to use in
Linux, so this change replaced the delay logic with a wrapper called
zfs_sleep_until(). This wrapper could be adopted upstream and in other
downstream ports to abstract away operating system-specific delay logic.
- These tunables are added as module parameters, and descriptions added
to the zfs-module-parameters.5 man page.
spa_asize_inflation
zfs_deadman_synctime_ms
zfs_vdev_max_active
zfs_vdev_async_write_active_min_dirty_percent
zfs_vdev_async_write_active_max_dirty_percent
zfs_vdev_async_read_max_active
zfs_vdev_async_read_min_active
zfs_vdev_async_write_max_active
zfs_vdev_async_write_min_active
zfs_vdev_scrub_max_active
zfs_vdev_scrub_min_active
zfs_vdev_sync_read_max_active
zfs_vdev_sync_read_min_active
zfs_vdev_sync_write_max_active
zfs_vdev_sync_write_min_active
zfs_dirty_data_max_percent
zfs_delay_min_dirty_percent
zfs_dirty_data_max_max_percent
zfs_dirty_data_max
zfs_dirty_data_max_max
zfs_dirty_data_sync
zfs_delay_scale
The latter four have type unsigned long, whereas they are uint64_t in
Illumos. This accommodates Linux's module_param() supported types, but
means they may overflow on 32-bit architectures.
The values zfs_dirty_data_max and zfs_dirty_data_max_max are the most
likely to overflow on 32-bit systems, since they express physical RAM
sizes in bytes. In fact, Illumos initializes zfs_dirty_data_max_max to
2^32 which does overflow. To resolve that, this port instead initializes
it in arc_init() to 25% of physical RAM, and adds the tunable
zfs_dirty_data_max_max_percent to override that percentage. While this
solution doesn't completely avoid the overflow issue, it should be a
reasonable default for most systems, and the minority of affected
systems can work around the issue by overriding the defaults.
- Fixed reversed logic in comment above zfs_delay_scale declaration.
- Clarified comments in vdev_queue.c regarding when per-queue minimums take
effect.
- Replaced dmu_tx_write_limit in the dmu_tx kstat file
with dmu_tx_dirty_delay and dmu_tx_dirty_over_max. The first counts
how many times a transaction has been delayed because the pool dirty
data has exceeded zfs_delay_min_dirty_percent. The latter counts how
many times the pool dirty data has exceeded zfs_dirty_data_max (which
we expect to never happen).
- The original patch would have regressed the bug fixed in
zfsonlinux/zfs@c418410, which prevented users from setting the
zfs_vdev_aggregation_limit tuning larger than SPA_MAXBLOCKSIZE.
A similar fix is added to vdev_queue_aggregate().
- In vdev_queue_io_to_issue(), dynamically allocate 'zio_t search' on the
heap instead of the stack. In Linux we can't afford such large
structures on the stack.
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Ned Bass <bass6@llnl.gov>
Reviewed by: Brendan Gregg <brendan.gregg@joyent.com>
Approved by: Robert Mustacchi <rm@joyent.com>
References:
http://www.illumos.org/issues/4045
illumos/illumos-gate@69962b5647e4a8b9b14998733b765925381b727e
Ported-by: Ned Bass <bass6@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #1913
2013-08-29 03:01:20 +00:00
|
|
|
ZIO_TYPE_WRITE, ZIO_PRIORITY_ASYNC_WRITE,
|
2009-01-15 21:59:39 +00:00
|
|
|
ZIO_FLAG_IO_REPAIR | (unexpected_errors ?
|
|
|
|
ZIO_FLAG_SELF_HEAL : 0), NULL, NULL));
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
vdev_mirror_state_change(vdev_t *vd, int faulted, int degraded)
|
|
|
|
{
|
|
|
|
if (faulted == vd->vdev_children)
|
|
|
|
vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
|
|
|
|
VDEV_AUX_NO_REPLICAS);
|
|
|
|
else if (degraded + faulted != 0)
|
|
|
|
vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, VDEV_AUX_NONE);
|
|
|
|
else
|
|
|
|
vdev_set_state(vd, B_FALSE, VDEV_STATE_HEALTHY, VDEV_AUX_NONE);
|
|
|
|
}
|
|
|
|
|
|
|
|
vdev_ops_t vdev_mirror_ops = {
|
|
|
|
vdev_mirror_open,
|
|
|
|
vdev_mirror_close,
|
|
|
|
vdev_default_asize,
|
|
|
|
vdev_mirror_io_start,
|
|
|
|
vdev_mirror_io_done,
|
|
|
|
vdev_mirror_state_change,
|
2010-05-28 20:45:14 +00:00
|
|
|
NULL,
|
|
|
|
NULL,
|
2008-11-20 20:01:55 +00:00
|
|
|
VDEV_TYPE_MIRROR, /* name of this vdev type */
|
|
|
|
B_FALSE /* not a leaf vdev */
|
|
|
|
};
|
|
|
|
|
|
|
|
vdev_ops_t vdev_replacing_ops = {
|
|
|
|
vdev_mirror_open,
|
|
|
|
vdev_mirror_close,
|
|
|
|
vdev_default_asize,
|
|
|
|
vdev_mirror_io_start,
|
|
|
|
vdev_mirror_io_done,
|
|
|
|
vdev_mirror_state_change,
|
2010-05-28 20:45:14 +00:00
|
|
|
NULL,
|
|
|
|
NULL,
|
2008-11-20 20:01:55 +00:00
|
|
|
VDEV_TYPE_REPLACING, /* name of this vdev type */
|
|
|
|
B_FALSE /* not a leaf vdev */
|
|
|
|
};
|
|
|
|
|
|
|
|
vdev_ops_t vdev_spare_ops = {
|
|
|
|
vdev_mirror_open,
|
|
|
|
vdev_mirror_close,
|
|
|
|
vdev_default_asize,
|
|
|
|
vdev_mirror_io_start,
|
|
|
|
vdev_mirror_io_done,
|
|
|
|
vdev_mirror_state_change,
|
2010-05-28 20:45:14 +00:00
|
|
|
NULL,
|
|
|
|
NULL,
|
2008-11-20 20:01:55 +00:00
|
|
|
VDEV_TYPE_SPARE, /* name of this vdev type */
|
|
|
|
B_FALSE /* not a leaf vdev */
|
|
|
|
};
|
2013-05-31 19:07:59 +00:00
|
|
|
|
|
|
|
#if defined(_KERNEL) && defined(HAVE_SPL)
|
FreeBSD r256956: Improve ZFS N-way mirror read performance by using load and locality information.
The existing algorithm selects a preferred leaf vdev based on offset of the zio
request modulo the number of members in the mirror. It assumes the devices are
of equal performance and that spreading the requests randomly over both drives
will be sufficient to saturate them. In practice this results in the leaf vdevs
being under utilized.
The new algorithm takes into the following additional factors:
* Load of the vdevs (number outstanding I/O requests)
* The locality of last queued I/O vs the new I/O request.
Within the locality calculation additional knowledge about the underlying vdev
is considered such as; is the device backing the vdev a rotating media device.
This results in performance increases across the board as well as significant
increases for predominantly streaming loads and for configurations which don't
have evenly performing devices.
The following are results from a setup with 3 Way Mirror with 2 x HD's and
1 x SSD from a basic test running multiple parrallel dd's.
With pre-fetch disabled (vfs.zfs.prefetch_disable=1):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 161 seconds @ 95 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 297 seconds @ 51 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 54 seconds @ 284 MB/s
With pre-fetch enabled (vfs.zfs.prefetch_disable=0):
== Stripe Balanced (default) ==
Read 15360MB using bs: 1048576, readers: 3, took 91 seconds @ 168 MB/s
== Load Balanced (zfslinux) ==
Read 15360MB using bs: 1048576, readers: 3, took 108 seconds @ 142 MB/s
== Load Balanced (locality freebsd) ==
Read 15360MB using bs: 1048576, readers: 3, took 48 seconds @ 320 MB/s
In addition to the performance changes the code was also restructured, with
the help of Justin Gibbs, to provide a more logical flow which also ensures
vdevs loads are only calculated from the set of valid candidates.
The following additional sysctls where added to allow the administrator
to tune the behaviour of the load algorithm:
* vfs.zfs.vdev.mirror.rotating_inc
* vfs.zfs.vdev.mirror.rotating_seek_inc
* vfs.zfs.vdev.mirror.rotating_seek_offset
* vfs.zfs.vdev.mirror.non_rotating_inc
* vfs.zfs.vdev.mirror.non_rotating_seek_inc
These changes where based on work started by the zfsonlinux developers:
https://github.com/zfsonlinux/zfs/pull/1487
Reviewed by: gibbs, mav, will
MFC after: 2 weeks
Sponsored by: Multiplay
References:
https://github.com/freebsd/freebsd@5c7a6f5d
https://github.com/freebsd/freebsd@31b7f68d
https://github.com/freebsd/freebsd@e186f564
Performance Testing:
https://github.com/zfsonlinux/zfs/pull/4334#issuecomment-189057141
Porting notes:
- The tunables were adjusted to have ZoL-style names.
- The code was modified to use ZoL's vd_nonrot.
- Fixes were done to make cstyle.pl happy
- Merge conflicts were handled manually
- freebsd/freebsd@e186f564bc946f82c76e0b34c2f0370ed9aea022 by my
collegue Andriy Gapon has been included. It applied perfectly, but
added a cstyle regression.
- This replaces 556011dbec2d10579819078559a77630fc559112 entirely.
- A typo "IO'a" has been corrected to say "IO's"
- Descriptions of new tunables were added to man/man5/zfs-module-parameters.5.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4334
2016-02-13 01:47:22 +00:00
|
|
|
module_param(zfs_vdev_mirror_rotating_inc, int, 0644);
|
|
|
|
MODULE_PARM_DESC(zfs_vdev_mirror_rotating_inc,
|
|
|
|
"Rotating media load increment for non-seeking I/O's");
|
|
|
|
|
|
|
|
module_param(zfs_vdev_mirror_rotating_seek_inc, int, 0644);
|
|
|
|
MODULE_PARM_DESC(zfs_vdev_mirror_rotating_seek_inc,
|
|
|
|
"Rotating media load increment for seeking I/O's");
|
|
|
|
|
|
|
|
module_param(zfs_vdev_mirror_rotating_seek_offset, int, 0644);
|
|
|
|
MODULE_PARM_DESC(zfs_vdev_mirror_rotating_seek_offset,
|
|
|
|
"Offset in bytes from the last I/O which "
|
|
|
|
"triggers a reduced rotating media seek increment");
|
|
|
|
|
|
|
|
module_param(zfs_vdev_mirror_non_rotating_inc, int, 0644);
|
|
|
|
MODULE_PARM_DESC(zfs_vdev_mirror_non_rotating_inc,
|
|
|
|
"Non-rotating media load increment for non-seeking I/O's");
|
|
|
|
|
|
|
|
module_param(zfs_vdev_mirror_non_rotating_seek_inc, int, 0644);
|
|
|
|
MODULE_PARM_DESC(zfs_vdev_mirror_non_rotating_seek_inc,
|
|
|
|
"Non-rotating media load increment for seeking I/O's");
|
|
|
|
|
2013-05-31 19:07:59 +00:00
|
|
|
#endif
|