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 2009 Sun Microsystems, Inc. All rights reserved.
|
2008-11-20 20:01:55 +00:00
|
|
|
* Use is subject to license terms.
|
|
|
|
*/
|
|
|
|
|
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
|
|
|
/*
|
|
|
|
* Copyright (c) 2013 by Delphix. All rights reserved.
|
|
|
|
*/
|
|
|
|
|
2008-11-20 20:01:55 +00:00
|
|
|
#include <sys/zfs_context.h>
|
|
|
|
#include <sys/dnode.h>
|
|
|
|
#include <sys/dmu_objset.h>
|
|
|
|
#include <sys/dmu_zfetch.h>
|
|
|
|
#include <sys/dmu.h>
|
|
|
|
#include <sys/dbuf.h>
|
2010-05-28 20:45:14 +00:00
|
|
|
#include <sys/kstat.h>
|
2008-11-20 20:01:55 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* I'm against tune-ables, but these should probably exist as tweakable globals
|
|
|
|
* until we can get this working the way we want it to.
|
|
|
|
*/
|
|
|
|
|
|
|
|
int zfs_prefetch_disable = 0;
|
|
|
|
|
|
|
|
/* max # of streams per zfetch */
|
2011-05-03 22:09:28 +00:00
|
|
|
unsigned int zfetch_max_streams = 8;
|
2008-11-20 20:01:55 +00:00
|
|
|
/* min time before stream reclaim */
|
2011-05-03 22:09:28 +00:00
|
|
|
unsigned int zfetch_min_sec_reap = 2;
|
2008-11-20 20:01:55 +00:00
|
|
|
/* max number of blocks to fetch at a time */
|
2011-05-03 22:09:28 +00:00
|
|
|
unsigned int zfetch_block_cap = 256;
|
2008-11-20 20:01:55 +00:00
|
|
|
/* number of bytes in a array_read at which we stop prefetching (1Mb) */
|
2011-05-03 22:09:28 +00:00
|
|
|
unsigned long zfetch_array_rd_sz = 1024 * 1024;
|
2008-11-20 20:01:55 +00:00
|
|
|
|
|
|
|
/* forward decls for static routines */
|
2013-06-11 17:12:34 +00:00
|
|
|
static boolean_t dmu_zfetch_colinear(zfetch_t *, zstream_t *);
|
2008-11-20 20:01:55 +00:00
|
|
|
static void dmu_zfetch_dofetch(zfetch_t *, zstream_t *);
|
|
|
|
static uint64_t dmu_zfetch_fetch(dnode_t *, uint64_t, uint64_t);
|
|
|
|
static uint64_t dmu_zfetch_fetchsz(dnode_t *, uint64_t, uint64_t);
|
2013-06-11 17:12:34 +00:00
|
|
|
static boolean_t dmu_zfetch_find(zfetch_t *, zstream_t *, int);
|
2008-11-20 20:01:55 +00:00
|
|
|
static int dmu_zfetch_stream_insert(zfetch_t *, zstream_t *);
|
|
|
|
static zstream_t *dmu_zfetch_stream_reclaim(zfetch_t *);
|
|
|
|
static void dmu_zfetch_stream_remove(zfetch_t *, zstream_t *);
|
|
|
|
static int dmu_zfetch_streams_equal(zstream_t *, zstream_t *);
|
|
|
|
|
2010-05-28 20:45:14 +00:00
|
|
|
typedef struct zfetch_stats {
|
|
|
|
kstat_named_t zfetchstat_hits;
|
|
|
|
kstat_named_t zfetchstat_misses;
|
|
|
|
kstat_named_t zfetchstat_colinear_hits;
|
|
|
|
kstat_named_t zfetchstat_colinear_misses;
|
|
|
|
kstat_named_t zfetchstat_stride_hits;
|
|
|
|
kstat_named_t zfetchstat_stride_misses;
|
|
|
|
kstat_named_t zfetchstat_reclaim_successes;
|
|
|
|
kstat_named_t zfetchstat_reclaim_failures;
|
|
|
|
kstat_named_t zfetchstat_stream_resets;
|
|
|
|
kstat_named_t zfetchstat_stream_noresets;
|
|
|
|
kstat_named_t zfetchstat_bogus_streams;
|
|
|
|
} zfetch_stats_t;
|
|
|
|
|
|
|
|
static zfetch_stats_t zfetch_stats = {
|
|
|
|
{ "hits", KSTAT_DATA_UINT64 },
|
|
|
|
{ "misses", KSTAT_DATA_UINT64 },
|
|
|
|
{ "colinear_hits", KSTAT_DATA_UINT64 },
|
|
|
|
{ "colinear_misses", KSTAT_DATA_UINT64 },
|
|
|
|
{ "stride_hits", KSTAT_DATA_UINT64 },
|
|
|
|
{ "stride_misses", KSTAT_DATA_UINT64 },
|
|
|
|
{ "reclaim_successes", KSTAT_DATA_UINT64 },
|
|
|
|
{ "reclaim_failures", KSTAT_DATA_UINT64 },
|
|
|
|
{ "streams_resets", KSTAT_DATA_UINT64 },
|
|
|
|
{ "streams_noresets", KSTAT_DATA_UINT64 },
|
|
|
|
{ "bogus_streams", KSTAT_DATA_UINT64 },
|
|
|
|
};
|
|
|
|
|
|
|
|
#define ZFETCHSTAT_INCR(stat, val) \
|
|
|
|
atomic_add_64(&zfetch_stats.stat.value.ui64, (val));
|
|
|
|
|
|
|
|
#define ZFETCHSTAT_BUMP(stat) ZFETCHSTAT_INCR(stat, 1);
|
|
|
|
|
|
|
|
kstat_t *zfetch_ksp;
|
|
|
|
|
2008-11-20 20:01:55 +00:00
|
|
|
/*
|
|
|
|
* Given a zfetch structure and a zstream structure, determine whether the
|
|
|
|
* blocks to be read are part of a co-linear pair of existing prefetch
|
|
|
|
* streams. If a set is found, coalesce the streams, removing one, and
|
|
|
|
* configure the prefetch so it looks for a strided access pattern.
|
|
|
|
*
|
|
|
|
* In other words: if we find two sequential access streams that are
|
|
|
|
* the same length and distance N appart, and this read is N from the
|
|
|
|
* last stream, then we are probably in a strided access pattern. So
|
|
|
|
* combine the two sequential streams into a single strided stream.
|
|
|
|
*
|
2013-06-11 17:12:34 +00:00
|
|
|
* Returns whether co-linear streams were found.
|
2008-11-20 20:01:55 +00:00
|
|
|
*/
|
2013-06-11 17:12:34 +00:00
|
|
|
static boolean_t
|
2008-11-20 20:01:55 +00:00
|
|
|
dmu_zfetch_colinear(zfetch_t *zf, zstream_t *zh)
|
|
|
|
{
|
|
|
|
zstream_t *z_walk;
|
|
|
|
zstream_t *z_comp;
|
|
|
|
|
|
|
|
if (! rw_tryenter(&zf->zf_rwlock, RW_WRITER))
|
|
|
|
return (0);
|
|
|
|
|
|
|
|
if (zh == NULL) {
|
|
|
|
rw_exit(&zf->zf_rwlock);
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
for (z_walk = list_head(&zf->zf_stream); z_walk;
|
|
|
|
z_walk = list_next(&zf->zf_stream, z_walk)) {
|
|
|
|
for (z_comp = list_next(&zf->zf_stream, z_walk); z_comp;
|
|
|
|
z_comp = list_next(&zf->zf_stream, z_comp)) {
|
|
|
|
int64_t diff;
|
|
|
|
|
|
|
|
if (z_walk->zst_len != z_walk->zst_stride ||
|
|
|
|
z_comp->zst_len != z_comp->zst_stride) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
diff = z_comp->zst_offset - z_walk->zst_offset;
|
|
|
|
if (z_comp->zst_offset + diff == zh->zst_offset) {
|
|
|
|
z_walk->zst_offset = zh->zst_offset;
|
2014-01-22 04:44:35 +00:00
|
|
|
z_walk->zst_direction = diff < 0 ?
|
|
|
|
ZFETCH_BACKWARD : ZFETCH_FORWARD;
|
2008-11-20 20:01:55 +00:00
|
|
|
z_walk->zst_stride =
|
|
|
|
diff * z_walk->zst_direction;
|
|
|
|
z_walk->zst_ph_offset =
|
|
|
|
zh->zst_offset + z_walk->zst_stride;
|
|
|
|
dmu_zfetch_stream_remove(zf, z_comp);
|
|
|
|
mutex_destroy(&z_comp->zst_lock);
|
|
|
|
kmem_free(z_comp, sizeof (zstream_t));
|
|
|
|
|
|
|
|
dmu_zfetch_dofetch(zf, z_walk);
|
|
|
|
|
|
|
|
rw_exit(&zf->zf_rwlock);
|
|
|
|
return (1);
|
|
|
|
}
|
|
|
|
|
|
|
|
diff = z_walk->zst_offset - z_comp->zst_offset;
|
|
|
|
if (z_walk->zst_offset + diff == zh->zst_offset) {
|
|
|
|
z_walk->zst_offset = zh->zst_offset;
|
2014-01-22 04:44:35 +00:00
|
|
|
z_walk->zst_direction = diff < 0 ?
|
|
|
|
ZFETCH_BACKWARD : ZFETCH_FORWARD;
|
2008-11-20 20:01:55 +00:00
|
|
|
z_walk->zst_stride =
|
|
|
|
diff * z_walk->zst_direction;
|
|
|
|
z_walk->zst_ph_offset =
|
|
|
|
zh->zst_offset + z_walk->zst_stride;
|
|
|
|
dmu_zfetch_stream_remove(zf, z_comp);
|
|
|
|
mutex_destroy(&z_comp->zst_lock);
|
|
|
|
kmem_free(z_comp, sizeof (zstream_t));
|
|
|
|
|
|
|
|
dmu_zfetch_dofetch(zf, z_walk);
|
|
|
|
|
|
|
|
rw_exit(&zf->zf_rwlock);
|
|
|
|
return (1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
rw_exit(&zf->zf_rwlock);
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Given a zstream_t, determine the bounds of the prefetch. Then call the
|
|
|
|
* routine that actually prefetches the individual blocks.
|
|
|
|
*/
|
|
|
|
static void
|
|
|
|
dmu_zfetch_dofetch(zfetch_t *zf, zstream_t *zs)
|
|
|
|
{
|
|
|
|
uint64_t prefetch_tail;
|
|
|
|
uint64_t prefetch_limit;
|
|
|
|
uint64_t prefetch_ofst;
|
|
|
|
uint64_t prefetch_len;
|
|
|
|
uint64_t blocks_fetched;
|
|
|
|
|
|
|
|
zs->zst_stride = MAX((int64_t)zs->zst_stride, zs->zst_len);
|
|
|
|
zs->zst_cap = MIN(zfetch_block_cap, 2 * zs->zst_cap);
|
|
|
|
|
|
|
|
prefetch_tail = MAX((int64_t)zs->zst_ph_offset,
|
|
|
|
(int64_t)(zs->zst_offset + zs->zst_stride));
|
|
|
|
/*
|
|
|
|
* XXX: use a faster division method?
|
|
|
|
*/
|
|
|
|
prefetch_limit = zs->zst_offset + zs->zst_len +
|
|
|
|
(zs->zst_cap * zs->zst_stride) / zs->zst_len;
|
|
|
|
|
|
|
|
while (prefetch_tail < prefetch_limit) {
|
|
|
|
prefetch_ofst = zs->zst_offset + zs->zst_direction *
|
|
|
|
(prefetch_tail - zs->zst_offset);
|
|
|
|
|
|
|
|
prefetch_len = zs->zst_len;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Don't prefetch beyond the end of the file, if working
|
|
|
|
* backwards.
|
|
|
|
*/
|
|
|
|
if ((zs->zst_direction == ZFETCH_BACKWARD) &&
|
|
|
|
(prefetch_ofst > prefetch_tail)) {
|
|
|
|
prefetch_len += prefetch_ofst;
|
|
|
|
prefetch_ofst = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* don't prefetch more than we're supposed to */
|
|
|
|
if (prefetch_len > zs->zst_len)
|
|
|
|
break;
|
|
|
|
|
|
|
|
blocks_fetched = dmu_zfetch_fetch(zf->zf_dnode,
|
|
|
|
prefetch_ofst, zs->zst_len);
|
|
|
|
|
|
|
|
prefetch_tail += zs->zst_stride;
|
|
|
|
/* stop if we've run out of stuff to prefetch */
|
|
|
|
if (blocks_fetched < zs->zst_len)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
zs->zst_ph_offset = prefetch_tail;
|
2010-05-28 20:45:14 +00:00
|
|
|
zs->zst_last = ddi_get_lbolt();
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
zfetch_init(void)
|
|
|
|
{
|
|
|
|
|
|
|
|
zfetch_ksp = kstat_create("zfs", 0, "zfetchstats", "misc",
|
|
|
|
KSTAT_TYPE_NAMED, sizeof (zfetch_stats) / sizeof (kstat_named_t),
|
|
|
|
KSTAT_FLAG_VIRTUAL);
|
|
|
|
|
|
|
|
if (zfetch_ksp != NULL) {
|
|
|
|
zfetch_ksp->ks_data = &zfetch_stats;
|
|
|
|
kstat_install(zfetch_ksp);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
zfetch_fini(void)
|
|
|
|
{
|
|
|
|
if (zfetch_ksp != NULL) {
|
|
|
|
kstat_delete(zfetch_ksp);
|
|
|
|
zfetch_ksp = NULL;
|
|
|
|
}
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This takes a pointer to a zfetch structure and a dnode. It performs the
|
|
|
|
* necessary setup for the zfetch structure, grokking data from the
|
|
|
|
* associated dnode.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
dmu_zfetch_init(zfetch_t *zf, dnode_t *dno)
|
|
|
|
{
|
|
|
|
if (zf == NULL) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
zf->zf_dnode = dno;
|
|
|
|
zf->zf_stream_cnt = 0;
|
|
|
|
zf->zf_alloc_fail = 0;
|
|
|
|
|
|
|
|
list_create(&zf->zf_stream, sizeof (zstream_t),
|
|
|
|
offsetof(zstream_t, zst_node));
|
|
|
|
|
|
|
|
rw_init(&zf->zf_rwlock, NULL, RW_DEFAULT, NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This function computes the actual size, in blocks, that can be prefetched,
|
|
|
|
* and fetches it.
|
|
|
|
*/
|
|
|
|
static uint64_t
|
|
|
|
dmu_zfetch_fetch(dnode_t *dn, uint64_t blkid, uint64_t nblks)
|
|
|
|
{
|
|
|
|
uint64_t fetchsz;
|
|
|
|
uint64_t i;
|
|
|
|
|
|
|
|
fetchsz = dmu_zfetch_fetchsz(dn, blkid, nblks);
|
|
|
|
|
|
|
|
for (i = 0; i < fetchsz; i++) {
|
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
|
|
|
dbuf_prefetch(dn, blkid + i, ZIO_PRIORITY_ASYNC_READ);
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
return (fetchsz);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* this function returns the number of blocks that would be prefetched, based
|
|
|
|
* upon the supplied dnode, blockid, and nblks. This is used so that we can
|
|
|
|
* update streams in place, and then prefetch with their old value after the
|
|
|
|
* fact. This way, we can delay the prefetch, but subsequent accesses to the
|
|
|
|
* stream won't result in the same data being prefetched multiple times.
|
|
|
|
*/
|
|
|
|
static uint64_t
|
|
|
|
dmu_zfetch_fetchsz(dnode_t *dn, uint64_t blkid, uint64_t nblks)
|
|
|
|
{
|
|
|
|
uint64_t fetchsz;
|
|
|
|
|
|
|
|
if (blkid > dn->dn_maxblkid) {
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* compute fetch size */
|
|
|
|
if (blkid + nblks + 1 > dn->dn_maxblkid) {
|
|
|
|
fetchsz = (dn->dn_maxblkid - blkid) + 1;
|
|
|
|
ASSERT(blkid + fetchsz - 1 <= dn->dn_maxblkid);
|
|
|
|
} else {
|
|
|
|
fetchsz = nblks;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
return (fetchsz);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2010-05-28 20:45:14 +00:00
|
|
|
* given a zfetch and a zstream structure, see if there is an associated zstream
|
2008-11-20 20:01:55 +00:00
|
|
|
* for this block read. If so, it starts a prefetch for the stream it
|
|
|
|
* located and returns true, otherwise it returns false
|
|
|
|
*/
|
2013-06-11 17:12:34 +00:00
|
|
|
static boolean_t
|
2008-11-20 20:01:55 +00:00
|
|
|
dmu_zfetch_find(zfetch_t *zf, zstream_t *zh, int prefetched)
|
|
|
|
{
|
|
|
|
zstream_t *zs;
|
|
|
|
int64_t diff;
|
|
|
|
int reset = !prefetched;
|
|
|
|
int rc = 0;
|
|
|
|
|
|
|
|
if (zh == NULL)
|
|
|
|
return (0);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* XXX: This locking strategy is a bit coarse; however, it's impact has
|
|
|
|
* yet to be tested. If this turns out to be an issue, it can be
|
|
|
|
* modified in a number of different ways.
|
|
|
|
*/
|
|
|
|
|
|
|
|
rw_enter(&zf->zf_rwlock, RW_READER);
|
|
|
|
top:
|
|
|
|
|
|
|
|
for (zs = list_head(&zf->zf_stream); zs;
|
|
|
|
zs = list_next(&zf->zf_stream, zs)) {
|
|
|
|
|
|
|
|
/*
|
|
|
|
* XXX - should this be an assert?
|
|
|
|
*/
|
|
|
|
if (zs->zst_len == 0) {
|
|
|
|
/* bogus stream */
|
2010-05-28 20:45:14 +00:00
|
|
|
ZFETCHSTAT_BUMP(zfetchstat_bogus_streams);
|
2008-11-20 20:01:55 +00:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We hit this case when we are in a strided prefetch stream:
|
|
|
|
* we will read "len" blocks before "striding".
|
|
|
|
*/
|
|
|
|
if (zh->zst_offset >= zs->zst_offset &&
|
|
|
|
zh->zst_offset < zs->zst_offset + zs->zst_len) {
|
2010-05-28 20:45:14 +00:00
|
|
|
if (prefetched) {
|
|
|
|
/* already fetched */
|
|
|
|
ZFETCHSTAT_BUMP(zfetchstat_stride_hits);
|
|
|
|
rc = 1;
|
|
|
|
goto out;
|
|
|
|
} else {
|
|
|
|
ZFETCHSTAT_BUMP(zfetchstat_stride_misses);
|
|
|
|
}
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is the forward sequential read case: we increment
|
|
|
|
* len by one each time we hit here, so we will enter this
|
|
|
|
* case on every read.
|
|
|
|
*/
|
|
|
|
if (zh->zst_offset == zs->zst_offset + zs->zst_len) {
|
|
|
|
|
|
|
|
reset = !prefetched && zs->zst_len > 1;
|
|
|
|
|
|
|
|
mutex_enter(&zs->zst_lock);
|
|
|
|
|
|
|
|
if (zh->zst_offset != zs->zst_offset + zs->zst_len) {
|
|
|
|
mutex_exit(&zs->zst_lock);
|
|
|
|
goto top;
|
|
|
|
}
|
|
|
|
zs->zst_len += zh->zst_len;
|
|
|
|
diff = zs->zst_len - zfetch_block_cap;
|
|
|
|
if (diff > 0) {
|
|
|
|
zs->zst_offset += diff;
|
|
|
|
zs->zst_len = zs->zst_len > diff ?
|
|
|
|
zs->zst_len - diff : 0;
|
|
|
|
}
|
|
|
|
zs->zst_direction = ZFETCH_FORWARD;
|
|
|
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Same as above, but reading backwards through the file.
|
|
|
|
*/
|
|
|
|
} else if (zh->zst_offset == zs->zst_offset - zh->zst_len) {
|
|
|
|
/* backwards sequential access */
|
|
|
|
|
|
|
|
reset = !prefetched && zs->zst_len > 1;
|
|
|
|
|
|
|
|
mutex_enter(&zs->zst_lock);
|
|
|
|
|
|
|
|
if (zh->zst_offset != zs->zst_offset - zh->zst_len) {
|
|
|
|
mutex_exit(&zs->zst_lock);
|
|
|
|
goto top;
|
|
|
|
}
|
|
|
|
|
|
|
|
zs->zst_offset = zs->zst_offset > zh->zst_len ?
|
|
|
|
zs->zst_offset - zh->zst_len : 0;
|
|
|
|
zs->zst_ph_offset = zs->zst_ph_offset > zh->zst_len ?
|
|
|
|
zs->zst_ph_offset - zh->zst_len : 0;
|
|
|
|
zs->zst_len += zh->zst_len;
|
|
|
|
|
|
|
|
diff = zs->zst_len - zfetch_block_cap;
|
|
|
|
if (diff > 0) {
|
|
|
|
zs->zst_ph_offset = zs->zst_ph_offset > diff ?
|
|
|
|
zs->zst_ph_offset - diff : 0;
|
|
|
|
zs->zst_len = zs->zst_len > diff ?
|
|
|
|
zs->zst_len - diff : zs->zst_len;
|
|
|
|
}
|
|
|
|
zs->zst_direction = ZFETCH_BACKWARD;
|
|
|
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
} else if ((zh->zst_offset - zs->zst_offset - zs->zst_stride <
|
|
|
|
zs->zst_len) && (zs->zst_len != zs->zst_stride)) {
|
|
|
|
/* strided forward access */
|
|
|
|
|
|
|
|
mutex_enter(&zs->zst_lock);
|
|
|
|
|
|
|
|
if ((zh->zst_offset - zs->zst_offset - zs->zst_stride >=
|
|
|
|
zs->zst_len) || (zs->zst_len == zs->zst_stride)) {
|
|
|
|
mutex_exit(&zs->zst_lock);
|
|
|
|
goto top;
|
|
|
|
}
|
|
|
|
|
|
|
|
zs->zst_offset += zs->zst_stride;
|
|
|
|
zs->zst_direction = ZFETCH_FORWARD;
|
|
|
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
} else if ((zh->zst_offset - zs->zst_offset + zs->zst_stride <
|
|
|
|
zs->zst_len) && (zs->zst_len != zs->zst_stride)) {
|
|
|
|
/* strided reverse access */
|
|
|
|
|
|
|
|
mutex_enter(&zs->zst_lock);
|
|
|
|
|
|
|
|
if ((zh->zst_offset - zs->zst_offset + zs->zst_stride >=
|
|
|
|
zs->zst_len) || (zs->zst_len == zs->zst_stride)) {
|
|
|
|
mutex_exit(&zs->zst_lock);
|
|
|
|
goto top;
|
|
|
|
}
|
|
|
|
|
|
|
|
zs->zst_offset = zs->zst_offset > zs->zst_stride ?
|
|
|
|
zs->zst_offset - zs->zst_stride : 0;
|
|
|
|
zs->zst_ph_offset = (zs->zst_ph_offset >
|
|
|
|
(2 * zs->zst_stride)) ?
|
|
|
|
(zs->zst_ph_offset - (2 * zs->zst_stride)) : 0;
|
|
|
|
zs->zst_direction = ZFETCH_BACKWARD;
|
|
|
|
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (zs) {
|
|
|
|
if (reset) {
|
|
|
|
zstream_t *remove = zs;
|
|
|
|
|
2010-05-28 20:45:14 +00:00
|
|
|
ZFETCHSTAT_BUMP(zfetchstat_stream_resets);
|
2008-11-20 20:01:55 +00:00
|
|
|
rc = 0;
|
|
|
|
mutex_exit(&zs->zst_lock);
|
|
|
|
rw_exit(&zf->zf_rwlock);
|
|
|
|
rw_enter(&zf->zf_rwlock, RW_WRITER);
|
|
|
|
/*
|
|
|
|
* Relocate the stream, in case someone removes
|
|
|
|
* it while we were acquiring the WRITER lock.
|
|
|
|
*/
|
|
|
|
for (zs = list_head(&zf->zf_stream); zs;
|
|
|
|
zs = list_next(&zf->zf_stream, zs)) {
|
|
|
|
if (zs == remove) {
|
|
|
|
dmu_zfetch_stream_remove(zf, zs);
|
|
|
|
mutex_destroy(&zs->zst_lock);
|
|
|
|
kmem_free(zs, sizeof (zstream_t));
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else {
|
2010-05-28 20:45:14 +00:00
|
|
|
ZFETCHSTAT_BUMP(zfetchstat_stream_noresets);
|
2008-11-20 20:01:55 +00:00
|
|
|
rc = 1;
|
|
|
|
dmu_zfetch_dofetch(zf, zs);
|
|
|
|
mutex_exit(&zs->zst_lock);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
out:
|
|
|
|
rw_exit(&zf->zf_rwlock);
|
|
|
|
return (rc);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Clean-up state associated with a zfetch structure. This frees allocated
|
|
|
|
* structure members, empties the zf_stream tree, and generally makes things
|
|
|
|
* nice. This doesn't free the zfetch_t itself, that's left to the caller.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
dmu_zfetch_rele(zfetch_t *zf)
|
|
|
|
{
|
|
|
|
zstream_t *zs;
|
|
|
|
zstream_t *zs_next;
|
|
|
|
|
|
|
|
ASSERT(!RW_LOCK_HELD(&zf->zf_rwlock));
|
|
|
|
|
|
|
|
for (zs = list_head(&zf->zf_stream); zs; zs = zs_next) {
|
|
|
|
zs_next = list_next(&zf->zf_stream, zs);
|
|
|
|
|
|
|
|
list_remove(&zf->zf_stream, zs);
|
|
|
|
mutex_destroy(&zs->zst_lock);
|
|
|
|
kmem_free(zs, sizeof (zstream_t));
|
|
|
|
}
|
|
|
|
list_destroy(&zf->zf_stream);
|
|
|
|
rw_destroy(&zf->zf_rwlock);
|
|
|
|
|
|
|
|
zf->zf_dnode = NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Given a zfetch and zstream structure, insert the zstream structure into the
|
|
|
|
* AVL tree contained within the zfetch structure. Peform the appropriate
|
|
|
|
* book-keeping. It is possible that another thread has inserted a stream which
|
|
|
|
* matches one that we are about to insert, so we must be sure to check for this
|
|
|
|
* case. If one is found, return failure, and let the caller cleanup the
|
|
|
|
* duplicates.
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
dmu_zfetch_stream_insert(zfetch_t *zf, zstream_t *zs)
|
|
|
|
{
|
|
|
|
zstream_t *zs_walk;
|
|
|
|
zstream_t *zs_next;
|
|
|
|
|
|
|
|
ASSERT(RW_WRITE_HELD(&zf->zf_rwlock));
|
|
|
|
|
|
|
|
for (zs_walk = list_head(&zf->zf_stream); zs_walk; zs_walk = zs_next) {
|
|
|
|
zs_next = list_next(&zf->zf_stream, zs_walk);
|
|
|
|
|
|
|
|
if (dmu_zfetch_streams_equal(zs_walk, zs)) {
|
2010-05-28 20:45:14 +00:00
|
|
|
return (0);
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
list_insert_head(&zf->zf_stream, zs);
|
|
|
|
zf->zf_stream_cnt++;
|
|
|
|
return (1);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Walk the list of zstreams in the given zfetch, find an old one (by time), and
|
|
|
|
* reclaim it for use by the caller.
|
|
|
|
*/
|
|
|
|
static zstream_t *
|
|
|
|
dmu_zfetch_stream_reclaim(zfetch_t *zf)
|
|
|
|
{
|
|
|
|
zstream_t *zs;
|
|
|
|
|
|
|
|
if (! rw_tryenter(&zf->zf_rwlock, RW_WRITER))
|
|
|
|
return (0);
|
|
|
|
|
|
|
|
for (zs = list_head(&zf->zf_stream); zs;
|
|
|
|
zs = list_next(&zf->zf_stream, zs)) {
|
|
|
|
|
2010-05-28 20:45:14 +00:00
|
|
|
if (((ddi_get_lbolt() - zs->zst_last)/hz) > zfetch_min_sec_reap)
|
2008-11-20 20:01:55 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (zs) {
|
|
|
|
dmu_zfetch_stream_remove(zf, zs);
|
|
|
|
mutex_destroy(&zs->zst_lock);
|
|
|
|
bzero(zs, sizeof (zstream_t));
|
|
|
|
} else {
|
|
|
|
zf->zf_alloc_fail++;
|
|
|
|
}
|
|
|
|
rw_exit(&zf->zf_rwlock);
|
|
|
|
|
|
|
|
return (zs);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Given a zfetch and zstream structure, remove the zstream structure from its
|
|
|
|
* container in the zfetch structure. Perform the appropriate book-keeping.
|
|
|
|
*/
|
|
|
|
static void
|
|
|
|
dmu_zfetch_stream_remove(zfetch_t *zf, zstream_t *zs)
|
|
|
|
{
|
|
|
|
ASSERT(RW_WRITE_HELD(&zf->zf_rwlock));
|
|
|
|
|
|
|
|
list_remove(&zf->zf_stream, zs);
|
|
|
|
zf->zf_stream_cnt--;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
dmu_zfetch_streams_equal(zstream_t *zs1, zstream_t *zs2)
|
|
|
|
{
|
|
|
|
if (zs1->zst_offset != zs2->zst_offset)
|
|
|
|
return (0);
|
|
|
|
|
|
|
|
if (zs1->zst_len != zs2->zst_len)
|
|
|
|
return (0);
|
|
|
|
|
|
|
|
if (zs1->zst_stride != zs2->zst_stride)
|
|
|
|
return (0);
|
|
|
|
|
|
|
|
if (zs1->zst_ph_offset != zs2->zst_ph_offset)
|
|
|
|
return (0);
|
|
|
|
|
|
|
|
if (zs1->zst_cap != zs2->zst_cap)
|
|
|
|
return (0);
|
|
|
|
|
|
|
|
if (zs1->zst_direction != zs2->zst_direction)
|
|
|
|
return (0);
|
|
|
|
|
|
|
|
return (1);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is the prefetch entry point. It calls all of the other dmu_zfetch
|
|
|
|
* routines to create, delete, find, or operate upon prefetch streams.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
dmu_zfetch(zfetch_t *zf, uint64_t offset, uint64_t size, int prefetched)
|
|
|
|
{
|
|
|
|
zstream_t zst;
|
|
|
|
zstream_t *newstream;
|
2013-06-11 17:12:34 +00:00
|
|
|
boolean_t fetched;
|
2008-11-20 20:01:55 +00:00
|
|
|
int inserted;
|
|
|
|
unsigned int blkshft;
|
|
|
|
uint64_t blksz;
|
|
|
|
|
|
|
|
if (zfs_prefetch_disable)
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* files that aren't ln2 blocksz are only one block -- nothing to do */
|
|
|
|
if (!zf->zf_dnode->dn_datablkshift)
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* convert offset and size, into blockid and nblocks */
|
|
|
|
blkshft = zf->zf_dnode->dn_datablkshift;
|
|
|
|
blksz = (1 << blkshft);
|
|
|
|
|
|
|
|
bzero(&zst, sizeof (zstream_t));
|
|
|
|
zst.zst_offset = offset >> blkshft;
|
|
|
|
zst.zst_len = (P2ROUNDUP(offset + size, blksz) -
|
|
|
|
P2ALIGN(offset, blksz)) >> blkshft;
|
|
|
|
|
|
|
|
fetched = dmu_zfetch_find(zf, &zst, prefetched);
|
2010-05-28 20:45:14 +00:00
|
|
|
if (fetched) {
|
|
|
|
ZFETCHSTAT_BUMP(zfetchstat_hits);
|
|
|
|
} else {
|
|
|
|
ZFETCHSTAT_BUMP(zfetchstat_misses);
|
2010-08-26 16:52:42 +00:00
|
|
|
if ((fetched = dmu_zfetch_colinear(zf, &zst))) {
|
2010-05-28 20:45:14 +00:00
|
|
|
ZFETCHSTAT_BUMP(zfetchstat_colinear_hits);
|
|
|
|
} else {
|
|
|
|
ZFETCHSTAT_BUMP(zfetchstat_colinear_misses);
|
|
|
|
}
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
if (!fetched) {
|
|
|
|
newstream = dmu_zfetch_stream_reclaim(zf);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* we still couldn't find a stream, drop the lock, and allocate
|
|
|
|
* one if possible. Otherwise, give up and go home.
|
|
|
|
*/
|
2010-05-28 20:45:14 +00:00
|
|
|
if (newstream) {
|
|
|
|
ZFETCHSTAT_BUMP(zfetchstat_reclaim_successes);
|
|
|
|
} else {
|
2008-11-20 20:01:55 +00:00
|
|
|
uint64_t maxblocks;
|
|
|
|
uint32_t max_streams;
|
|
|
|
uint32_t cur_streams;
|
|
|
|
|
2010-05-28 20:45:14 +00:00
|
|
|
ZFETCHSTAT_BUMP(zfetchstat_reclaim_failures);
|
2008-11-20 20:01:55 +00:00
|
|
|
cur_streams = zf->zf_stream_cnt;
|
|
|
|
maxblocks = zf->zf_dnode->dn_maxblkid;
|
|
|
|
|
|
|
|
max_streams = MIN(zfetch_max_streams,
|
|
|
|
(maxblocks / zfetch_block_cap));
|
|
|
|
if (max_streams == 0) {
|
|
|
|
max_streams++;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (cur_streams >= max_streams) {
|
|
|
|
return;
|
|
|
|
}
|
2013-11-01 19:26:11 +00:00
|
|
|
newstream =
|
2014-11-21 00:09:39 +00:00
|
|
|
kmem_zalloc(sizeof (zstream_t), KM_SLEEP);
|
2008-11-20 20:01:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
newstream->zst_offset = zst.zst_offset;
|
|
|
|
newstream->zst_len = zst.zst_len;
|
|
|
|
newstream->zst_stride = zst.zst_len;
|
|
|
|
newstream->zst_ph_offset = zst.zst_len + zst.zst_offset;
|
|
|
|
newstream->zst_cap = zst.zst_len;
|
|
|
|
newstream->zst_direction = ZFETCH_FORWARD;
|
2010-05-28 20:45:14 +00:00
|
|
|
newstream->zst_last = ddi_get_lbolt();
|
2008-11-20 20:01:55 +00:00
|
|
|
|
|
|
|
mutex_init(&newstream->zst_lock, NULL, MUTEX_DEFAULT, NULL);
|
|
|
|
|
|
|
|
rw_enter(&zf->zf_rwlock, RW_WRITER);
|
|
|
|
inserted = dmu_zfetch_stream_insert(zf, newstream);
|
|
|
|
rw_exit(&zf->zf_rwlock);
|
|
|
|
|
|
|
|
if (!inserted) {
|
|
|
|
mutex_destroy(&newstream->zst_lock);
|
|
|
|
kmem_free(newstream, sizeof (zstream_t));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2010-08-26 18:49:16 +00:00
|
|
|
|
|
|
|
#if defined(_KERNEL) && defined(HAVE_SPL)
|
|
|
|
module_param(zfs_prefetch_disable, int, 0644);
|
|
|
|
MODULE_PARM_DESC(zfs_prefetch_disable, "Disable all ZFS prefetching");
|
2011-05-03 22:09:28 +00:00
|
|
|
|
|
|
|
module_param(zfetch_max_streams, uint, 0644);
|
|
|
|
MODULE_PARM_DESC(zfetch_max_streams, "Max number of streams per zfetch");
|
|
|
|
|
|
|
|
module_param(zfetch_min_sec_reap, uint, 0644);
|
|
|
|
MODULE_PARM_DESC(zfetch_min_sec_reap, "Min time before stream reclaim");
|
|
|
|
|
|
|
|
module_param(zfetch_block_cap, uint, 0644);
|
|
|
|
MODULE_PARM_DESC(zfetch_block_cap, "Max number of blocks to fetch at a time");
|
|
|
|
|
|
|
|
module_param(zfetch_array_rd_sz, ulong, 0644);
|
|
|
|
MODULE_PARM_DESC(zfetch_array_rd_sz, "Number of bytes in a array_read");
|
2010-08-26 18:49:16 +00:00
|
|
|
#endif
|