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
do not map the b_pages pages into buffer_map KVA. The use of the
unmapped buffers eliminate the need to perform TLB shootdown for
mapping on the buffer creation and reuse, greatly reducing the amount
of IPIs for shootdown on big-SMP machines and eliminating up to 25-30%
of the system time on i/o intensive workloads.
The unmapped buffer should be explicitely requested by the GB_UNMAPPED
flag by the consumer. For unmapped buffer, no KVA reservation is
performed at all. The consumer might request unmapped buffer which
does have a KVA reserve, to manually map it without recursing into
buffer cache and blocking, with the GB_KVAALLOC flag.
When the mapped buffer is requested and unmapped buffer already
exists, the cache performs an upgrade, possibly reusing the KVA
reservation.
Unmapped buffer is translated into unmapped bio in g_vfs_strategy().
Unmapped bio carry a pointer to the vm_page_t array, offset and length
instead of the data pointer. The provider which processes the bio
should explicitely specify a readiness to accept unmapped bio,
otherwise g_down geom thread performs the transient upgrade of the bio
request by mapping the pages into the new bio_transient_map KVA
submap.
The bio_transient_map submap claims up to 10% of the buffer map, and
the total buffer_map + bio_transient_map KVA usage stays the
same. Still, it could be manually tuned by kern.bio_transient_maxcnt
tunable, in the units of the transient mappings. Eventually, the
bio_transient_map could be removed after all geom classes and drivers
can accept unmapped i/o requests.
Unmapped support can be turned off by the vfs.unmapped_buf_allowed
tunable, disabling which makes the buffer (or cluster) creation
requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped
buffers are only enabled by default on the architectures where
pmap_copy_page() was implemented and tested.
In the rework, filesystem metadata is not the subject to maxbufspace
limit anymore. Since the metadata buffers are always mapped, the
buffers still have to fit into the buffer map, which provides a
reasonable (but practically unreachable) upper bound on it. The
non-metadata buffer allocations, both mapped and unmapped, is
accounted against maxbufspace, as before. Effectively, this means that
the maxbufspace is forced on mapped and unmapped buffers separately.
The pre-patch bufspace limiting code did not worked, because
buffer_map fragmentation does not allow the limit to be reached.
By Jeff Roberson request, the getnewbuf() function was split into
smaller single-purpose functions.
Sponsored by: The FreeBSD Foundation
Discussed with: jeff (previous version)
Tested by: pho, scottl (previous version), jhb, bf
MFC after: 2 weeks
It includes three parts:
1) Modifications to CAM to detect media media changes and report them to
disk(9) layer. For modern SATA (and potentially UAS) devices it utilizes
Asynchronous Notification mechanism to receive events from hardware.
Active polling with TEST UNIT READY commands with 3 seconds period is used
for incapable hardware. After that both CD and DA drivers work the same way,
detecting two conditions: "NOT READY: Medium not present" after medium was
detected previously, and "UNIT ATTENTION: Not ready to ready change, medium
may have changed". First one reported to disk(9) as media removal, second
as media insert/change. To reliably receive second event new
AC_UNIT_ATTENTION async added to make UAs broadcasted to all periphs by
generic error handling code in cam_periph_error().
2) Modifications to GEOM core to handle media remove and change events.
Media removal handled by spoiling all consumers attached to the provider.
Media change event also schedules provider retaste after spoiling to probe
new media. New flag G_CF_ORPHAN was added to consumers to reflect that
consumer is in process of destruction. It allows retaste to create new
geom instance of the same class, while previous one is still dying.
3) Modifications to some GEOM classes: DEV -- to report media change
events to devd; VFS -- to handle spoiling same as orphan to prevent
accessing replaced media. PART class already handles spoiling alike to
orphan.
Reviewed by: silence on geom@ and scsi@
Tested by: avg
Sponsored by: iXsystems, Inc. / PC-BSD
MFC after: 2 months
a da(4) instance going away while GEOM is still probing it.
In this case, the GEOM disk class instance has been created by
disk_create(), and the taste of the disk is queued in the GEOM
event queue.
While that event is queued, the da(4) instance goes away. When the
open call comes into the da(4) driver, it dereferences the freed
(but non-NULL) peripheral pointer provided by GEOM, which results
in a panic.
The solution is to add a callback to the GEOM disk code that is
called when all of its resources are cleaned up. This is
implemented inside GEOM by adding an optional callback that is
called when all consumers have detached from a provider, and the
provider is about to be deleted.
scsi_cd.c,
scsi_da.c: In the register routine for the cd(4) and da(4)
routines, acquire a reference to the CAM peripheral
instance just before we call disk_create().
Use the new GEOM disk d_gone() callback to register
a callback (dadiskgonecb()/cddiskgonecb()) that
decrements the peripheral reference count once GEOM
has finished cleaning up its resources.
In the cd(4) driver, clean up open and close
behavior slightly. GEOM makes sure we only get one
open() and one close call, so there is no need to
set an open flag and decrement the reference count
if we are not the first open.
In the cd(4) driver, use cam_periph_release_locked()
in a couple of error scenarios to avoid extra mutex
calls.
geom.h: Add a new, optional, providergone callback that
is called when a provider is about to be deleted.
geom_disk.h: Add a new d_gone() callback to the GEOM disk
interface.
Bump the DISK_VERSION to version 2. This probably
should have been done after a couple of previous
changes, especially the addition of the d_getattr()
callback.
geom_disk.c: Add a providergone callback for the disk class,
g_disk_providergone(), that calls the user's
d_gone() callback if it exists.
Bump the DISK_VERSION to 2.
geom_subr.c: In g_destroy_provider(), call the providergone
callback if it has been provided.
In g_new_geomf(), propagate the class's
providergone callback to the new geom instance.
blkfront.c: Callers of disk_create() are supposed to pass in
DISK_VERSION, not an explicit disk API version
number. Update the blkfront driver to do that.
disk.9: Update the disk(9) man page to include information
on the new d_gone() callback, as well as the
previously added d_getattr() callback, d_descr
field, and HBA PCI ID fields.
MFC after: 5 days
DEVFS, and make it accessible via the diskinfo utility.
Extend GEOM's generic attribute query mechanism into generic disk consumers.
sys/geom/geom_disk.c:
sys/geom/geom_disk.h:
sys/cam/scsi/scsi_da.c:
sys/cam/ata/ata_da.c:
- Allow disk providers to implement a new method which can override
the default BIO_GETATTR response, d_getattr(struct bio *). This
function returns -1 if not handled, otherwise it returns 0 or an
errno to be passed to g_io_deliver().
sys/cam/scsi/scsi_da.c:
sys/cam/ata/ata_da.c:
- Don't copy the serial number to dp->d_ident anymore, as the CAM XPT
is now responsible for returning this information via
d_getattr()->(a)dagetattr()->xpt_getatr().
sys/geom/geom_dev.c:
- Implement a new ioctl, DIOCGPHYSPATH, which returns the GEOM
attribute "GEOM::physpath", if possible. If the attribute request
returns a zero-length string, ENOENT is returned.
usr.sbin/diskinfo/diskinfo.c:
- If the DIOCGPHYSPATH ioctl is successful, report physical path
data when diskinfo is executed with the '-v' option.
Submitted by: will
Reviewed by: gibbs
Sponsored by: Spectra Logic Corporation
Add generic attribute change notification support to GEOM.
sys/sys/geom/geom.h:
Add a new attrchanged method field to both g_class
and g_geom.
sys/sys/geom/geom.h:
sys/geom/geom_event.c:
- Provide the g_attr_changed() function that providers
can use to advertise attribute changes.
- Perform delivery of attribute change notifications
from a thread context via the standard GEOM event
mechanism.
sys/geom/geom_subr.c:
Inherit the attrchanged method from class to geom (class instance).
sys/geom/geom_disk.c:
Provide disk_attr_changed() to provide g_attr_changed() access
to consumers of the disk API.
sys/cam/scsi/scsi_pass.c:
sys/cam/scsi/scsi_da.c:
sys/geom/geom_dev.c:
sys/geom/geom_disk.c:
Use attribute changed events to track updates to physical path
information.
sys/cam/scsi/scsi_da.c:
Add AC_ADVINFO_CHANGED to the registered asynchronous CAM
events for this driver. When this event occurs, and
the updated buffer type references our physical path
attribute, emit a GEOM attribute changed event via the
disk_attr_changed() API.
sys/cam/scsi/scsi_pass.c:
Add AC_ADVINFO_CHANGED to the registered asynchronous CAM
events for this driver. When this event occurs, update
the physical patch devfs alias for this pass instance.
Submitted by: gibbs
Sponsored by: Spectra Logic Corporation
Introduce new type of BIO_GETATTR -- GEOM::setstate, used to inform lower
GEOM about state of it's providers from the point of upper layers.
Make geom_disk use led(4) subsystem to illuminate states in such fashion:
FAILED - "1" (on), REBUILD - "f5" (slow blink), RESYNC - "f1" (fast blink),
ACTIVE - "0" (off).
LED name should be set for each disk via kern.geom.disk.%s.led sysctl.
Later disk API could be extended to allow disk driver to report this info
in custom way via it's own facilities.
Change BIO_GETATTR("GEOM::kerneldump") API to make set_dumper() called by
consumer (geom_dev) instead of provider (geom_disk). This allows any geom
insert it's code into the dump call chain, implementing more sophisticated
functionality then just disk partitioning.
proceed while g_unload_class() blocks the event thread. Fix this by not
running g_unload_class() as a GEOM event and dropping the topology lock
when withering needs to proceed.
PR: kern/139847
Silence on: freebsd-geom
struct bio to store classification information, and a hook
for classifier functions that can be called by g_io_request().
This code is from Fabio Checconi as part of his GSOC work.
to it for tasting. This is useful when the class, through means outside
the scope of GEOM, can claim providers previously unclaimed.
The g_retaste() function posts an event which is handled by the
g_retaste_event().
Event suggested by: phk
particular provider. Use this function where g_orphan_provider()
is being called so that the flags are updated correctly and
g_orphan_provider() is called only when allowed.
error on the request. Add a wrapper, gctl_set_param_err(), that
sets the error on the request from the error returned by
gctl_set_param() and update current callers of gctl_set_param()
to call gctl_set_param_err() instead.
This makes gctl_set_param() much more usable in situations where
the caller knows better what to do with certain (apparent) error
conditions and setting an error on the request is not one of the
things that need to be done.
understood. This makes room for additional binary compatibility in the
future.
Put fields in the class for the geom's methods and initialize the methods
of a new geom from these fields. This saves some code in all classes.
Retire g_sanity() and corresponding debugflag (0x8)
Retire g_{stall,release}_events().
Under #ifdef DIAGNOSTIC:
Make g_valid_obj() an official function and have it return an an
non-zero integer which indicates the kind of object when found.
Implement G_VALID_{CLASS,GEOM,CONSUMER,PROVIDER}() macros based
on g_valid_obj().
Sprinkle calls to these macros liberally over the infrastructure.
Always check that we do not free a live object.
shown that it is not useful.
Rename the relative count g_access_rel() function to g_access(), only
the name has changed.
Change all g_access_rel() calls in our CVS tree to call g_access() instead.
Add an #ifndef BURN_BRIDGES #define of g_access_rel() for source
code compatibility.
