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
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
would resize a partition, but label providers - e.g. /dev/gptid/XXX - would
stay the same size.
Reviewed by: mav
MFC after: 1 month
Sponsored by: FreeBSD Foundation
When parent provider has been resized, the scheme specific G_PART_RESIZE
method does an update of scheme's metadata. But all changes are not saved
to disk, until `gpart commit` will be called.
Discussed with: trasz
MFC after: 1 month
Introduce new function devstat_end_transaction_bio_bt(), adding new argument
to specify present time. Use this function to move binuptime() out of lock,
substantially reducing lock congestion when slow timecounter is used.
disable GEOM tasting to avoid the "bouncing GEOM" problem where, when
you shut down the consumer of a provider which can be viewed in multiple
ways (typically a mirror whose members are labeled partitions), GEOM
will immediately taste that provider's alter ego and reattach the
consumer.
Approved by: re (glebius)
always wait for provider close. Old algorithm was reported to cause NULL
dereference panic on attempt to close provider after softc destruction.
If not global workaroung in GEOM, that could even cause destruction with
requests still in flight.
The previous method was to set the D_UNMAPPED_IO flag in the cdevsw
for the driver. The problem with this is that in many cases (e.g.
sa(4)) there may be some instances of the driver that can handle
unmapped I/O and some that can't. The isp(4) driver can handle
unmapped I/O, but the esp(4) driver currently cannot. The cdevsw
is shared among all driver instances.
So instead of setting a flag on the cdevsw, set a flag on the cdev.
This allows drivers to indicate support for unmapped I/O on a
per-instance basis.
sys/conf.h: Remove the D_UNMAPPED_IO cdevsw flag and replace it
with an SI_UNMAPPED cdev flag.
kern_physio.c: Look at the cdev SI_UNMAPPED flag to determine
whether or not a particular driver can handle
unmapped I/O.
geom_dev.c: Set the SI_UNMAPPED flag for all GEOM cdevs.
Since GEOM will create a temporary mapping when
needed, setting SI_UNMAPPED unconditionally will
work.
Remove the D_UNMAPPED_IO flag.
nvme_ns.c: Set the SI_UNMAPPED flag on cdevs created here
if NVME_UNMAPPED_BIO_SUPPORT is enabled.
vfs_aio.c: In aio_qphysio(), check the SI_UNMAPPED flag on a
cdev instead of the D_UNMAPPED_IO flag on the cdevsw.
sys/param.h: Bump __FreeBSD_version to 1000045 for the switch from
setting the D_UNMAPPED_IO flag in the cdevsw to setting
SI_UNMAPPED in the cdev.
Reviewed by: kib, jimharris
MFC after: 1 week
Sponsored by: Spectra Logic
the GEOM_PART. Instead of just using number of entries from the GPT
header, calculate this limit based on the reserved space between
GPT header and first available LBA.
MFC after: 2 weeks
being defined in <sys/diskmbr.h>. Instead give the symbols here a
"PC98_" prefix. This way, both <sys/diskmbr.h> and <sys/diskpc98.h>
can be included in the same C source file.
The renaming is trivial. The only gotcha is that DOSBBSECTOR is
also redefined from 0 to 1. This because DOSBBSECTOR was always
used in conjunction with an addition of 1. The PC98_BBSECTOR symbol
is defined as 1 and the expression is simplified.
Note: it is not believed that ports are seriously impacted; or at
all for that matter.
Approved by: nyan@
vfs_busy(mp);
vfs_write_suspend(mp);
which are problematic if other thread starts unmount between two
calls. The unmount starts a write, while vfs_write_suspend() drain
writers. On the other hand, unmount drains busy references, causing
the deadlock.
Add a flag argument to vfs_write_suspend and require the callers of it
to specify VS_SKIP_UNMOUNT flag, when the call is performed not in the
mount path, i.e. the covered vnode is not locked. The suspension is
not attempted if VS_SKIP_UNMOUNT is specified and unmount is in
progress.
Reported and tested by: Andreas Longwitz <longwitz@incore.de>
Sponsored by: The FreeBSD Foundation
MFC after: 3 weeks
Ensure that d_delmaxsize is always set, removing init to 0 which could cause
future issues if use cases change.
Allow kern.cam.da.X.delete_max (which maps to d_delmaxsize) to be increased
up to the calculated max after being reduced.
MFC after: 1 day
X-MFC-With: r249940
originally inspired by the Solaris vmem detailed in the proceedings
of usenix 2001. The NetBSD version was heavily refactored for bugs
and simplicity.
- Use this resource allocator to allocate the buffer and transient maps.
Buffer cache defrags are reduced by 25% when used by filesystems with
mixed block sizes. Ultimately this may permit dynamic buffer cache
sizing on low KVA machines.
Discussed with: alc, kib, attilio
Tested by: pho
Sponsored by: EMC / Isilon Storage Division
SPC-4 specification states that serial number may be property of device,
but not a specific logical unit. People reported about FC storages using
serial number in that way, making it unusable for purposes of LUN multipath
detection. SPC-4 states that designators associated with logical unit from
the VPD page 83h "Device Identification" should be used for that purpose.
Report first of them in the new attribute in such preference order: NAA,
EUI-64, T10 and SCSI name string.
While there, make GEOM DISK properly report GEOM::ident in XML output also
using d_getattr() method, if available. This fixes serial numbers reporting
for SCSI disks in `geom disk list` output and confxml.
Discussed with: gibbs, ken
Sponsored by: iXsystems, Inc.
MFC after: 2 weeks
This allows setting attributes on tables. One simply does not provide
an index in that case. Otherwise the entry corresponding the index has
the attribute set or unset.
Use this change to fix a relatively longstanding bug in our GPT scheme
that's the result of rev 198097 (relatively harmless) followed by rev
237057 (damaging). The damaging part being that our GPT scheme always
has the active flag set on the PMBR slice. This is in violation with
EFI. Existing EFI implementions for both x86 and ia64 reject the GPT.
As such, GPT disks created by us aren't usable under EFI because of
that.
After this change, GPT disks never have the active flag set on the PMBR
slice. In order to make the GPT disk bootable under some x86 BIOSes,
the reason of rev 198097, one must now set the active attribute on the
gpt table. The kernel will apply this to the PMBR slice For (S)ATA:
gpart set -a active ada0
To fix an existing GPT disk that has the active flag set in the PMBR,
and that does not need the flag, use (again for (S)ATA):
gpart unset -a active ada0
The EBR, MBR & PC98 schemes, which also impement at least 1 attribute,
now check to make sure the entry passed is valid. They do not have
attributes that apply to the table.
used previously caused probe failure on platforms where char is unsigned
(e.g. ARM), as mftrecsz can be negative.
Submitted by: Ilya Bakulin <ilya@bakulin.de>
MFC after: 2 weeks
requests.
sys/geom/geom_disk.h:
- Added d_delmaxsize which represents the maximum size of individual
device delete requests in bytes. This can be used by devices to
inform geom of their size limitations regarding delete operations
which are generally different from the read / write limits as data
is not usually transferred from the host to physical device.
sys/geom/geom_disk.c:
- Use new d_delmaxsize to calculate the size of chunks passed through to
the underlying strategy during deletes instead of using read / write
optimised values. This defaults to d_maxsize if unset (0).
- Moved d_maxsize default up so it can be used to default d_delmaxsize
sys/cam/ata/ata_da.c:
- Added d_delmaxsize calculations for TRIM and CFA
sys/cam/scsi/scsi_da.c:
- Added re-calculation of d_delmaxsize whenever delete_method is set.
- Added kern.cam.da.X.delete_max sysctl which allows the max size for
delete requests to be limited. This is useful in preventing timeouts
on devices who's delete methods are slow. It should be noted that
this limit is reset then the device delete method is changed and
that it can only be lowered not increased from the device max.
Reviewed by: mav
Approved by: pjd (mentor)
size of a delete request sent to the providing device performed by g_dev_ioctl.
This allows the kernel and apps via ioctl e.g. newfs -E to request large LBA
deletes which siginificantly improves performance.
Previously this was hard coded to 65536 sectors, the new default is 262144
which doubles the throughput of deletes on commonly available SSD's.
In tests on a Intel 520 120GB FW: 400i disk it improved the delete throughput
from 1.6GB/s to over 2.6GB/s on a full disk delete such as that done via
newfs -E
For some SSD's where delete time is pretty much constant, no matter what
the request, setting this to 0 will provide significantly better throughput
e.g. Samsung 840 240GB FW DXT07B0Q @ 262144 = 79G/s, @ 0 = 2259G/s
Reviewed by: mav
Approved by: pjd (mentor)
MFC after: 2 weeks
Pointy-hat to: me, for not realizing snprintf() is available in kernel.
Thanks to: jh, for bringing me the good news of snprintf(), Pawel Worach, for
noting that the panic can be provoked in i386 and not in amd64
Only look for FDT partitions if our potential parent is a DISK device.
Excluding direct recursion on the flashmap geoms was insufficient
because it did not prevent the underlying device from being retrieved if
flashmap geoms were further partitioned.
Reviewed by: imp
Sponsored by: DARPA, AFRL
implementation, error on the side of conservatism and only create labels
for GEOMs of classes DISK and MULTIPATH.
Discussed with: trasz
Approved by: silence from freebsd-geom@
