2 .\" Copyright (c) 2013 by Turbo Fredriksson <turbo@bayour.com>. All rights reserved.
3 .\" Copyright (c) 2019 by Delphix. All rights reserved.
4 .\" Copyright (c) 2019 Datto Inc.
5 .\" The contents of this file are subject to the terms of the Common Development
6 .\" and Distribution License (the "License"). You may not use this file except
7 .\" in compliance with the License. You can obtain a copy of the license at
8 .\" usr/src/OPENSOLARIS.LICENSE or http://www.opensolaris.org/os/licensing.
10 .\" See the License for the specific language governing permissions and
11 .\" limitations under the License. When distributing Covered Code, include this
12 .\" CDDL HEADER in each file and include the License file at
13 .\" usr/src/OPENSOLARIS.LICENSE. If applicable, add the following below this
14 .\" CDDL HEADER, with the fields enclosed by brackets "[]" replaced with your
15 .\" own identifying information:
16 .\" Portions Copyright [yyyy] [name of copyright owner]
17 .TH ZFS-MODULE-PARAMETERS 5 "Feb 15, 2019"
19 zfs\-module\-parameters \- ZFS module parameters
23 Description of the different parameters to the ZFS module.
25 .SS "Module parameters"
32 \fBdbuf_cache_max_bytes\fR (ulong)
35 Maximum size in bytes of the dbuf cache. When \fB0\fR this value will default
36 to \fB1/2^dbuf_cache_shift\fR (1/32) of the target ARC size, otherwise the
37 provided value in bytes will be used. The behavior of the dbuf cache and its
38 associated settings can be observed via the \fB/proc/spl/kstat/zfs/dbufstats\fR
41 Default value: \fB0\fR.
47 \fBdbuf_metadata_cache_max_bytes\fR (ulong)
50 Maximum size in bytes of the metadata dbuf cache. When \fB0\fR this value will
51 default to \fB1/2^dbuf_cache_shift\fR (1/16) of the target ARC size, otherwise
52 the provided value in bytes will be used. The behavior of the metadata dbuf
53 cache and its associated settings can be observed via the
54 \fB/proc/spl/kstat/zfs/dbufstats\fR kstat.
56 Default value: \fB0\fR.
62 \fBdbuf_cache_hiwater_pct\fR (uint)
65 The percentage over \fBdbuf_cache_max_bytes\fR when dbufs must be evicted
68 Default value: \fB10\fR%.
74 \fBdbuf_cache_lowater_pct\fR (uint)
77 The percentage below \fBdbuf_cache_max_bytes\fR when the evict thread stops
80 Default value: \fB10\fR%.
86 \fBdbuf_cache_shift\fR (int)
89 Set the size of the dbuf cache, \fBdbuf_cache_max_bytes\fR, to a log2 fraction
90 of the target arc size.
92 Default value: \fB5\fR.
98 \fBdbuf_metadata_cache_shift\fR (int)
101 Set the size of the dbuf metadata cache, \fBdbuf_metadata_cache_max_bytes\fR,
102 to a log2 fraction of the target arc size.
104 Default value: \fB6\fR.
110 \fBignore_hole_birth\fR (int)
113 This is an alias for \fBsend_holes_without_birth_time\fR.
119 \fBl2arc_feed_again\fR (int)
122 Turbo L2ARC warm-up. When the L2ARC is cold the fill interval will be set as
125 Use \fB1\fR for yes (default) and \fB0\fR to disable.
131 \fBl2arc_feed_min_ms\fR (ulong)
134 Min feed interval in milliseconds. Requires \fBl2arc_feed_again=1\fR and only
135 applicable in related situations.
137 Default value: \fB200\fR.
143 \fBl2arc_feed_secs\fR (ulong)
146 Seconds between L2ARC writing
148 Default value: \fB1\fR.
154 \fBl2arc_headroom\fR (ulong)
157 How far through the ARC lists to search for L2ARC cacheable content, expressed
158 as a multiplier of \fBl2arc_write_max\fR
160 Default value: \fB2\fR.
166 \fBl2arc_headroom_boost\fR (ulong)
169 Scales \fBl2arc_headroom\fR by this percentage when L2ARC contents are being
170 successfully compressed before writing. A value of 100 disables this feature.
172 Default value: \fB200\fR%.
178 \fBl2arc_noprefetch\fR (int)
181 Do not write buffers to L2ARC if they were prefetched but not used by
184 Use \fB1\fR for yes (default) and \fB0\fR to disable.
190 \fBl2arc_norw\fR (int)
193 No reads during writes
195 Use \fB1\fR for yes and \fB0\fR for no (default).
201 \fBl2arc_write_boost\fR (ulong)
204 Cold L2ARC devices will have \fBl2arc_write_max\fR increased by this amount
205 while they remain cold.
207 Default value: \fB8,388,608\fR.
213 \fBl2arc_write_max\fR (ulong)
216 Max write bytes per interval
218 Default value: \fB8,388,608\fR.
224 \fBmetaslab_aliquot\fR (ulong)
227 Metaslab granularity, in bytes. This is roughly similar to what would be
228 referred to as the "stripe size" in traditional RAID arrays. In normal
229 operation, ZFS will try to write this amount of data to a top-level vdev
230 before moving on to the next one.
232 Default value: \fB524,288\fR.
238 \fBmetaslab_bias_enabled\fR (int)
241 Enable metaslab group biasing based on its vdev's over- or under-utilization
242 relative to the pool.
244 Use \fB1\fR for yes (default) and \fB0\fR for no.
250 \fBmetaslab_force_ganging\fR (ulong)
253 Make some blocks above a certain size be gang blocks. This option is used
254 by the test suite to facilitate testing.
256 Default value: \fB16,777,217\fR.
262 \fBzfs_metaslab_segment_weight_enabled\fR (int)
265 Enable/disable segment-based metaslab selection.
267 Use \fB1\fR for yes (default) and \fB0\fR for no.
273 \fBzfs_metaslab_switch_threshold\fR (int)
276 When using segment-based metaslab selection, continue allocating
277 from the active metaslab until \fBzfs_metaslab_switch_threshold\fR
278 worth of buckets have been exhausted.
280 Default value: \fB2\fR.
286 \fBmetaslab_debug_load\fR (int)
289 Load all metaslabs during pool import.
291 Use \fB1\fR for yes and \fB0\fR for no (default).
297 \fBmetaslab_debug_unload\fR (int)
300 Prevent metaslabs from being unloaded.
302 Use \fB1\fR for yes and \fB0\fR for no (default).
308 \fBmetaslab_fragmentation_factor_enabled\fR (int)
311 Enable use of the fragmentation metric in computing metaslab weights.
313 Use \fB1\fR for yes (default) and \fB0\fR for no.
319 \fBzfs_vdev_default_ms_count\fR (int)
322 When a vdev is added target this number of metaslabs per top-level vdev.
324 Default value: \fB200\fR.
330 \fBzfs_vdev_min_ms_count\fR (int)
333 Minimum number of metaslabs to create in a top-level vdev.
335 Default value: \fB16\fR.
341 \fBvdev_ms_count_limit\fR (int)
344 Practical upper limit of total metaslabs per top-level vdev.
346 Default value: \fB131,072\fR.
352 \fBmetaslab_preload_enabled\fR (int)
355 Enable metaslab group preloading.
357 Use \fB1\fR for yes (default) and \fB0\fR for no.
363 \fBmetaslab_lba_weighting_enabled\fR (int)
366 Give more weight to metaslabs with lower LBAs, assuming they have
367 greater bandwidth as is typically the case on a modern constant
368 angular velocity disk drive.
370 Use \fB1\fR for yes (default) and \fB0\fR for no.
376 \fBsend_holes_without_birth_time\fR (int)
379 When set, the hole_birth optimization will not be used, and all holes will
380 always be sent on zfs send. This is useful if you suspect your datasets are
381 affected by a bug in hole_birth.
383 Use \fB1\fR for on (default) and \fB0\fR for off.
389 \fBspa_config_path\fR (charp)
394 Default value: \fB/etc/zfs/zpool.cache\fR.
400 \fBspa_asize_inflation\fR (int)
403 Multiplication factor used to estimate actual disk consumption from the
404 size of data being written. The default value is a worst case estimate,
405 but lower values may be valid for a given pool depending on its
406 configuration. Pool administrators who understand the factors involved
407 may wish to specify a more realistic inflation factor, particularly if
408 they operate close to quota or capacity limits.
410 Default value: \fB24\fR.
416 \fBspa_load_print_vdev_tree\fR (int)
419 Whether to print the vdev tree in the debugging message buffer during pool import.
420 Use 0 to disable and 1 to enable.
422 Default value: \fB0\fR.
428 \fBspa_load_verify_data\fR (int)
431 Whether to traverse data blocks during an "extreme rewind" (\fB-X\fR)
432 import. Use 0 to disable and 1 to enable.
434 An extreme rewind import normally performs a full traversal of all
435 blocks in the pool for verification. If this parameter is set to 0,
436 the traversal skips non-metadata blocks. It can be toggled once the
437 import has started to stop or start the traversal of non-metadata blocks.
439 Default value: \fB1\fR.
445 \fBspa_load_verify_metadata\fR (int)
448 Whether to traverse blocks during an "extreme rewind" (\fB-X\fR)
449 pool import. Use 0 to disable and 1 to enable.
451 An extreme rewind import normally performs a full traversal of all
452 blocks in the pool for verification. If this parameter is set to 0,
453 the traversal is not performed. It can be toggled once the import has
454 started to stop or start the traversal.
456 Default value: \fB1\fR.
462 \fBspa_load_verify_maxinflight\fR (int)
465 Maximum concurrent I/Os during the traversal performed during an "extreme
466 rewind" (\fB-X\fR) pool import.
468 Default value: \fB10000\fR.
474 \fBspa_slop_shift\fR (int)
477 Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space
478 in the pool to be consumed. This ensures that we don't run the pool
479 completely out of space, due to unaccounted changes (e.g. to the MOS).
480 It also limits the worst-case time to allocate space. If we have
481 less than this amount of free space, most ZPL operations (e.g. write,
482 create) will return ENOSPC.
484 Default value: \fB5\fR.
490 \fBvdev_removal_max_span\fR (int)
493 During top-level vdev removal, chunks of data are copied from the vdev
494 which may include free space in order to trade bandwidth for IOPS.
495 This parameter determines the maximum span of free space (in bytes)
496 which will be included as "unnecessary" data in a chunk of copied data.
498 The default value here was chosen to align with
499 \fBzfs_vdev_read_gap_limit\fR, which is a similar concept when doing
500 regular reads (but there's no reason it has to be the same).
502 Default value: \fB32,768\fR.
508 \fBzfetch_array_rd_sz\fR (ulong)
511 If prefetching is enabled, disable prefetching for reads larger than this size.
513 Default value: \fB1,048,576\fR.
519 \fBzfetch_max_distance\fR (uint)
522 Max bytes to prefetch per stream (default 8MB).
524 Default value: \fB8,388,608\fR.
530 \fBzfetch_max_streams\fR (uint)
533 Max number of streams per zfetch (prefetch streams per file).
535 Default value: \fB8\fR.
541 \fBzfetch_min_sec_reap\fR (uint)
544 Min time before an active prefetch stream can be reclaimed
546 Default value: \fB2\fR.
552 \fBzfs_abd_scatter_min_size\fR (uint)
555 This is the minimum allocation size that will use scatter (page-based)
556 ABD's. Smaller allocations will use linear ABD's.
558 Default value: \fB1536\fR (512B and 1KB allocations will be linear).
564 \fBzfs_arc_dnode_limit\fR (ulong)
567 When the number of bytes consumed by dnodes in the ARC exceeds this number of
568 bytes, try to unpin some of it in response to demand for non-metadata. This
569 value acts as a ceiling to the amount of dnode metadata, and defaults to 0 which
570 indicates that a percent which is based on \fBzfs_arc_dnode_limit_percent\fR of
571 the ARC meta buffers that may be used for dnodes.
573 See also \fBzfs_arc_meta_prune\fR which serves a similar purpose but is used
574 when the amount of metadata in the ARC exceeds \fBzfs_arc_meta_limit\fR rather
575 than in response to overall demand for non-metadata.
578 Default value: \fB0\fR.
584 \fBzfs_arc_dnode_limit_percent\fR (ulong)
587 Percentage that can be consumed by dnodes of ARC meta buffers.
589 See also \fBzfs_arc_dnode_limit\fR which serves a similar purpose but has a
590 higher priority if set to nonzero value.
592 Default value: \fB10\fR%.
598 \fBzfs_arc_dnode_reduce_percent\fR (ulong)
601 Percentage of ARC dnodes to try to scan in response to demand for non-metadata
602 when the number of bytes consumed by dnodes exceeds \fBzfs_arc_dnode_limit\fR.
605 Default value: \fB10\fR% of the number of dnodes in the ARC.
611 \fBzfs_arc_average_blocksize\fR (int)
614 The ARC's buffer hash table is sized based on the assumption of an average
615 block size of \fBzfs_arc_average_blocksize\fR (default 8K). This works out
616 to roughly 1MB of hash table per 1GB of physical memory with 8-byte pointers.
617 For configurations with a known larger average block size this value can be
618 increased to reduce the memory footprint.
621 Default value: \fB8192\fR.
627 \fBzfs_arc_evict_batch_limit\fR (int)
630 Number ARC headers to evict per sub-list before proceeding to another sub-list.
631 This batch-style operation prevents entire sub-lists from being evicted at once
632 but comes at a cost of additional unlocking and locking.
634 Default value: \fB10\fR.
640 \fBzfs_arc_grow_retry\fR (int)
643 If set to a non zero value, it will replace the arc_grow_retry value with this value.
644 The arc_grow_retry value (default 5) is the number of seconds the ARC will wait before
645 trying to resume growth after a memory pressure event.
647 Default value: \fB0\fR.
653 \fBzfs_arc_lotsfree_percent\fR (int)
656 Throttle I/O when free system memory drops below this percentage of total
657 system memory. Setting this value to 0 will disable the throttle.
659 Default value: \fB10\fR%.
665 \fBzfs_arc_max\fR (ulong)
668 Max arc size of ARC in bytes. If set to 0 then it will consume 1/2 of system
669 RAM. This value must be at least 67108864 (64 megabytes).
671 This value can be changed dynamically with some caveats. It cannot be set back
672 to 0 while running and reducing it below the current ARC size will not cause
673 the ARC to shrink without memory pressure to induce shrinking.
675 Default value: \fB0\fR.
681 \fBzfs_arc_meta_adjust_restarts\fR (ulong)
684 The number of restart passes to make while scanning the ARC attempting
685 the free buffers in order to stay below the \fBzfs_arc_meta_limit\fR.
686 This value should not need to be tuned but is available to facilitate
687 performance analysis.
689 Default value: \fB4096\fR.
695 \fBzfs_arc_meta_limit\fR (ulong)
698 The maximum allowed size in bytes that meta data buffers are allowed to
699 consume in the ARC. When this limit is reached meta data buffers will
700 be reclaimed even if the overall arc_c_max has not been reached. This
701 value defaults to 0 which indicates that a percent which is based on
702 \fBzfs_arc_meta_limit_percent\fR of the ARC may be used for meta data.
704 This value my be changed dynamically except that it cannot be set back to 0
705 for a specific percent of the ARC; it must be set to an explicit value.
707 Default value: \fB0\fR.
713 \fBzfs_arc_meta_limit_percent\fR (ulong)
716 Percentage of ARC buffers that can be used for meta data.
718 See also \fBzfs_arc_meta_limit\fR which serves a similar purpose but has a
719 higher priority if set to nonzero value.
722 Default value: \fB75\fR%.
728 \fBzfs_arc_meta_min\fR (ulong)
731 The minimum allowed size in bytes that meta data buffers may consume in
732 the ARC. This value defaults to 0 which disables a floor on the amount
733 of the ARC devoted meta data.
735 Default value: \fB0\fR.
741 \fBzfs_arc_meta_prune\fR (int)
744 The number of dentries and inodes to be scanned looking for entries
745 which can be dropped. This may be required when the ARC reaches the
746 \fBzfs_arc_meta_limit\fR because dentries and inodes can pin buffers
747 in the ARC. Increasing this value will cause to dentry and inode caches
748 to be pruned more aggressively. Setting this value to 0 will disable
749 pruning the inode and dentry caches.
751 Default value: \fB10,000\fR.
757 \fBzfs_arc_meta_strategy\fR (int)
760 Define the strategy for ARC meta data buffer eviction (meta reclaim strategy).
761 A value of 0 (META_ONLY) will evict only the ARC meta data buffers.
762 A value of 1 (BALANCED) indicates that additional data buffers may be evicted if
763 that is required to in order to evict the required number of meta data buffers.
765 Default value: \fB1\fR.
771 \fBzfs_arc_min\fR (ulong)
774 Min arc size of ARC in bytes. If set to 0 then arc_c_min will default to
775 consuming the larger of 32M or 1/32 of total system memory.
777 Default value: \fB0\fR.
783 \fBzfs_arc_min_prefetch_ms\fR (int)
786 Minimum time prefetched blocks are locked in the ARC, specified in ms.
787 A value of \fB0\fR will default to 1000 ms.
789 Default value: \fB0\fR.
795 \fBzfs_arc_min_prescient_prefetch_ms\fR (int)
798 Minimum time "prescient prefetched" blocks are locked in the ARC, specified
799 in ms. These blocks are meant to be prefetched fairly aggresively ahead of
800 the code that may use them. A value of \fB0\fR will default to 6000 ms.
802 Default value: \fB0\fR.
808 \fBzfs_max_missing_tvds\fR (int)
811 Number of missing top-level vdevs which will be allowed during
812 pool import (only in read-only mode).
814 Default value: \fB0\fR
820 \fBzfs_multilist_num_sublists\fR (int)
823 To allow more fine-grained locking, each ARC state contains a series
824 of lists for both data and meta data objects. Locking is performed at
825 the level of these "sub-lists". This parameters controls the number of
826 sub-lists per ARC state, and also applies to other uses of the
827 multilist data structure.
829 Default value: \fB4\fR or the number of online CPUs, whichever is greater
835 \fBzfs_arc_overflow_shift\fR (int)
838 The ARC size is considered to be overflowing if it exceeds the current
839 ARC target size (arc_c) by a threshold determined by this parameter.
840 The threshold is calculated as a fraction of arc_c using the formula
841 "arc_c >> \fBzfs_arc_overflow_shift\fR".
843 The default value of 8 causes the ARC to be considered to be overflowing
844 if it exceeds the target size by 1/256th (0.3%) of the target size.
846 When the ARC is overflowing, new buffer allocations are stalled until
847 the reclaim thread catches up and the overflow condition no longer exists.
849 Default value: \fB8\fR.
856 \fBzfs_arc_p_min_shift\fR (int)
859 If set to a non zero value, this will update arc_p_min_shift (default 4)
861 arc_p_min_shift is used to shift of arc_c for calculating both min and max
864 Default value: \fB0\fR.
870 \fBzfs_arc_p_dampener_disable\fR (int)
873 Disable arc_p adapt dampener
875 Use \fB1\fR for yes (default) and \fB0\fR to disable.
881 \fBzfs_arc_shrink_shift\fR (int)
884 If set to a non zero value, this will update arc_shrink_shift (default 7)
887 Default value: \fB0\fR.
893 \fBzfs_arc_pc_percent\fR (uint)
896 Percent of pagecache to reclaim arc to
898 This tunable allows ZFS arc to play more nicely with the kernel's LRU
899 pagecache. It can guarantee that the arc size won't collapse under scanning
900 pressure on the pagecache, yet still allows arc to be reclaimed down to
901 zfs_arc_min if necessary. This value is specified as percent of pagecache
902 size (as measured by NR_FILE_PAGES) where that percent may exceed 100. This
903 only operates during memory pressure/reclaim.
905 Default value: \fB0\fR% (disabled).
911 \fBzfs_arc_sys_free\fR (ulong)
914 The target number of bytes the ARC should leave as free memory on the system.
915 Defaults to the larger of 1/64 of physical memory or 512K. Setting this
916 option to a non-zero value will override the default.
918 Default value: \fB0\fR.
924 \fBzfs_autoimport_disable\fR (int)
927 Disable pool import at module load by ignoring the cache file (typically \fB/etc/zfs/zpool.cache\fR).
929 Use \fB1\fR for yes (default) and \fB0\fR for no.
935 \fBzfs_checksums_per_second\fR (int)
938 Rate limit checksum events to this many per second. Note that this should
939 not be set below the zed thresholds (currently 10 checksums over 10 sec)
940 or else zed may not trigger any action.
948 \fBzfs_commit_timeout_pct\fR (int)
951 This controls the amount of time that a ZIL block (lwb) will remain "open"
952 when it isn't "full", and it has a thread waiting for it to be committed to
953 stable storage. The timeout is scaled based on a percentage of the last lwb
954 latency to avoid significantly impacting the latency of each individual
955 transaction record (itx).
957 Default value: \fB5\fR%.
963 \fBzfs_condense_indirect_vdevs_enable\fR (int)
966 Enable condensing indirect vdev mappings. When set to a non-zero value,
967 attempt to condense indirect vdev mappings if the mapping uses more than
968 \fBzfs_condense_min_mapping_bytes\fR bytes of memory and if the obsolete
969 space map object uses more than \fBzfs_condense_max_obsolete_bytes\fR
970 bytes on-disk. The condensing process is an attempt to save memory by
971 removing obsolete mappings.
973 Default value: \fB1\fR.
979 \fBzfs_condense_max_obsolete_bytes\fR (ulong)
982 Only attempt to condense indirect vdev mappings if the on-disk size
983 of the obsolete space map object is greater than this number of bytes
984 (see \fBfBzfs_condense_indirect_vdevs_enable\fR).
986 Default value: \fB1,073,741,824\fR.
992 \fBzfs_condense_min_mapping_bytes\fR (ulong)
995 Minimum size vdev mapping to attempt to condense (see
996 \fBzfs_condense_indirect_vdevs_enable\fR).
998 Default value: \fB131,072\fR.
1004 \fBzfs_dbgmsg_enable\fR (int)
1007 Internally ZFS keeps a small log to facilitate debugging. By default the log
1008 is disabled, to enable it set this option to 1. The contents of the log can
1009 be accessed by reading the /proc/spl/kstat/zfs/dbgmsg file. Writing 0 to
1010 this proc file clears the log.
1012 Default value: \fB0\fR.
1018 \fBzfs_dbgmsg_maxsize\fR (int)
1021 The maximum size in bytes of the internal ZFS debug log.
1023 Default value: \fB4M\fR.
1029 \fBzfs_dbuf_state_index\fR (int)
1032 This feature is currently unused. It is normally used for controlling what
1033 reporting is available under /proc/spl/kstat/zfs.
1035 Default value: \fB0\fR.
1041 \fBzfs_deadman_enabled\fR (int)
1044 When a pool sync operation takes longer than \fBzfs_deadman_synctime_ms\fR
1045 milliseconds, or when an individual I/O takes longer than
1046 \fBzfs_deadman_ziotime_ms\fR milliseconds, then the operation is considered to
1047 be "hung". If \fBzfs_deadman_enabled\fR is set then the deadman behavior is
1048 invoked as described by the \fBzfs_deadman_failmode\fR module option.
1049 By default the deadman is enabled and configured to \fBwait\fR which results
1050 in "hung" I/Os only being logged. The deadman is automatically disabled
1051 when a pool gets suspended.
1053 Default value: \fB1\fR.
1059 \fBzfs_deadman_failmode\fR (charp)
1062 Controls the failure behavior when the deadman detects a "hung" I/O. Valid
1063 values are \fBwait\fR, \fBcontinue\fR, and \fBpanic\fR.
1065 \fBwait\fR - Wait for a "hung" I/O to complete. For each "hung" I/O a
1066 "deadman" event will be posted describing that I/O.
1068 \fBcontinue\fR - Attempt to recover from a "hung" I/O by re-dispatching it
1069 to the I/O pipeline if possible.
1071 \fBpanic\fR - Panic the system. This can be used to facilitate an automatic
1072 fail-over to a properly configured fail-over partner.
1074 Default value: \fBwait\fR.
1080 \fBzfs_deadman_checktime_ms\fR (int)
1083 Check time in milliseconds. This defines the frequency at which we check
1084 for hung I/O and potentially invoke the \fBzfs_deadman_failmode\fR behavior.
1086 Default value: \fB60,000\fR.
1092 \fBzfs_deadman_synctime_ms\fR (ulong)
1095 Interval in milliseconds after which the deadman is triggered and also
1096 the interval after which a pool sync operation is considered to be "hung".
1097 Once this limit is exceeded the deadman will be invoked every
1098 \fBzfs_deadman_checktime_ms\fR milliseconds until the pool sync completes.
1100 Default value: \fB600,000\fR.
1106 \fBzfs_deadman_ziotime_ms\fR (ulong)
1109 Interval in milliseconds after which the deadman is triggered and an
1110 individual I/O operation is considered to be "hung". As long as the I/O
1111 remains "hung" the deadman will be invoked every \fBzfs_deadman_checktime_ms\fR
1112 milliseconds until the I/O completes.
1114 Default value: \fB300,000\fR.
1120 \fBzfs_dedup_prefetch\fR (int)
1123 Enable prefetching dedup-ed blks
1125 Use \fB1\fR for yes and \fB0\fR to disable (default).
1131 \fBzfs_delay_min_dirty_percent\fR (int)
1134 Start to delay each transaction once there is this amount of dirty data,
1135 expressed as a percentage of \fBzfs_dirty_data_max\fR.
1136 This value should be >= zfs_vdev_async_write_active_max_dirty_percent.
1137 See the section "ZFS TRANSACTION DELAY".
1139 Default value: \fB60\fR%.
1145 \fBzfs_delay_scale\fR (int)
1148 This controls how quickly the transaction delay approaches infinity.
1149 Larger values cause longer delays for a given amount of dirty data.
1151 For the smoothest delay, this value should be about 1 billion divided
1152 by the maximum number of operations per second. This will smoothly
1153 handle between 10x and 1/10th this number.
1155 See the section "ZFS TRANSACTION DELAY".
1157 Note: \fBzfs_delay_scale\fR * \fBzfs_dirty_data_max\fR must be < 2^64.
1159 Default value: \fB500,000\fR.
1165 \fBzfs_slow_io_events_per_second\fR (int)
1168 Rate limit delay zevents (which report slow I/Os) to this many per second.
1176 \fBzfs_unlink_suspend_progress\fR (uint)
1179 When enabled, files will not be asynchronously removed from the list of pending
1180 unlinks and the space they consume will be leaked. Once this option has been
1181 disabled and the dataset is remounted, the pending unlinks will be processed
1182 and the freed space returned to the pool.
1183 This option is used by the test suite to facilitate testing.
1185 Uses \fB0\fR (default) to allow progress and \fB1\fR to pause progress.
1191 \fBzfs_delete_blocks\fR (ulong)
1194 This is the used to define a large file for the purposes of delete. Files
1195 containing more than \fBzfs_delete_blocks\fR will be deleted asynchronously
1196 while smaller files are deleted synchronously. Decreasing this value will
1197 reduce the time spent in an unlink(2) system call at the expense of a longer
1198 delay before the freed space is available.
1200 Default value: \fB20,480\fR.
1206 \fBzfs_dirty_data_max\fR (int)
1209 Determines the dirty space limit in bytes. Once this limit is exceeded, new
1210 writes are halted until space frees up. This parameter takes precedence
1211 over \fBzfs_dirty_data_max_percent\fR.
1212 See the section "ZFS TRANSACTION DELAY".
1214 Default value: \fB10\fR% of physical RAM, capped at \fBzfs_dirty_data_max_max\fR.
1220 \fBzfs_dirty_data_max_max\fR (int)
1223 Maximum allowable value of \fBzfs_dirty_data_max\fR, expressed in bytes.
1224 This limit is only enforced at module load time, and will be ignored if
1225 \fBzfs_dirty_data_max\fR is later changed. This parameter takes
1226 precedence over \fBzfs_dirty_data_max_max_percent\fR. See the section
1227 "ZFS TRANSACTION DELAY".
1229 Default value: \fB25\fR% of physical RAM.
1235 \fBzfs_dirty_data_max_max_percent\fR (int)
1238 Maximum allowable value of \fBzfs_dirty_data_max\fR, expressed as a
1239 percentage of physical RAM. This limit is only enforced at module load
1240 time, and will be ignored if \fBzfs_dirty_data_max\fR is later changed.
1241 The parameter \fBzfs_dirty_data_max_max\fR takes precedence over this
1242 one. See the section "ZFS TRANSACTION DELAY".
1244 Default value: \fB25\fR%.
1250 \fBzfs_dirty_data_max_percent\fR (int)
1253 Determines the dirty space limit, expressed as a percentage of all
1254 memory. Once this limit is exceeded, new writes are halted until space frees
1255 up. The parameter \fBzfs_dirty_data_max\fR takes precedence over this
1256 one. See the section "ZFS TRANSACTION DELAY".
1258 Default value: \fB10\fR%, subject to \fBzfs_dirty_data_max_max\fR.
1264 \fBzfs_dirty_data_sync_percent\fR (int)
1267 Start syncing out a transaction group if there's at least this much dirty data
1268 as a percentage of \fBzfs_dirty_data_max\fR. This should be less than
1269 \fBzfs_vdev_async_write_active_min_dirty_percent\fR.
1271 Default value: \fB20\fR% of \fBzfs_dirty_data_max\fR.
1277 \fBzfs_fletcher_4_impl\fR (string)
1280 Select a fletcher 4 implementation.
1282 Supported selectors are: \fBfastest\fR, \fBscalar\fR, \fBsse2\fR, \fBssse3\fR,
1283 \fBavx2\fR, \fBavx512f\fR, and \fBaarch64_neon\fR.
1284 All of the selectors except \fBfastest\fR and \fBscalar\fR require instruction
1285 set extensions to be available and will only appear if ZFS detects that they are
1286 present at runtime. If multiple implementations of fletcher 4 are available,
1287 the \fBfastest\fR will be chosen using a micro benchmark. Selecting \fBscalar\fR
1288 results in the original, CPU based calculation, being used. Selecting any option
1289 other than \fBfastest\fR and \fBscalar\fR results in vector instructions from
1290 the respective CPU instruction set being used.
1292 Default value: \fBfastest\fR.
1298 \fBzfs_free_bpobj_enabled\fR (int)
1301 Enable/disable the processing of the free_bpobj object.
1303 Default value: \fB1\fR.
1309 \fBzfs_async_block_max_blocks\fR (ulong)
1312 Maximum number of blocks freed in a single txg.
1314 Default value: \fB100,000\fR.
1320 \fBzfs_override_estimate_recordsize\fR (ulong)
1323 Record size calculation override for zfs send estimates.
1325 Default value: \fB0\fR.
1331 \fBzfs_vdev_async_read_max_active\fR (int)
1334 Maximum asynchronous read I/Os active to each device.
1335 See the section "ZFS I/O SCHEDULER".
1337 Default value: \fB3\fR.
1343 \fBzfs_vdev_async_read_min_active\fR (int)
1346 Minimum asynchronous read I/Os active to each device.
1347 See the section "ZFS I/O SCHEDULER".
1349 Default value: \fB1\fR.
1355 \fBzfs_vdev_async_write_active_max_dirty_percent\fR (int)
1358 When the pool has more than
1359 \fBzfs_vdev_async_write_active_max_dirty_percent\fR dirty data, use
1360 \fBzfs_vdev_async_write_max_active\fR to limit active async writes. If
1361 the dirty data is between min and max, the active I/O limit is linearly
1362 interpolated. See the section "ZFS I/O SCHEDULER".
1364 Default value: \fB60\fR%.
1370 \fBzfs_vdev_async_write_active_min_dirty_percent\fR (int)
1373 When the pool has less than
1374 \fBzfs_vdev_async_write_active_min_dirty_percent\fR dirty data, use
1375 \fBzfs_vdev_async_write_min_active\fR to limit active async writes. If
1376 the dirty data is between min and max, the active I/O limit is linearly
1377 interpolated. See the section "ZFS I/O SCHEDULER".
1379 Default value: \fB30\fR%.
1385 \fBzfs_vdev_async_write_max_active\fR (int)
1388 Maximum asynchronous write I/Os active to each device.
1389 See the section "ZFS I/O SCHEDULER".
1391 Default value: \fB10\fR.
1397 \fBzfs_vdev_async_write_min_active\fR (int)
1400 Minimum asynchronous write I/Os active to each device.
1401 See the section "ZFS I/O SCHEDULER".
1403 Lower values are associated with better latency on rotational media but poorer
1404 resilver performance. The default value of 2 was chosen as a compromise. A
1405 value of 3 has been shown to improve resilver performance further at a cost of
1406 further increasing latency.
1408 Default value: \fB2\fR.
1414 \fBzfs_vdev_initializing_max_active\fR (int)
1417 Maximum initializing I/Os active to each device.
1418 See the section "ZFS I/O SCHEDULER".
1420 Default value: \fB1\fR.
1426 \fBzfs_vdev_initializing_min_active\fR (int)
1429 Minimum initializing I/Os active to each device.
1430 See the section "ZFS I/O SCHEDULER".
1432 Default value: \fB1\fR.
1438 \fBzfs_vdev_max_active\fR (int)
1441 The maximum number of I/Os active to each device. Ideally, this will be >=
1442 the sum of each queue's max_active. It must be at least the sum of each
1443 queue's min_active. See the section "ZFS I/O SCHEDULER".
1445 Default value: \fB1,000\fR.
1451 \fBzfs_vdev_removal_max_active\fR (int)
1454 Maximum removal I/Os active to each device.
1455 See the section "ZFS I/O SCHEDULER".
1457 Default value: \fB2\fR.
1463 \fBzfs_vdev_removal_min_active\fR (int)
1466 Minimum removal I/Os active to each device.
1467 See the section "ZFS I/O SCHEDULER".
1469 Default value: \fB1\fR.
1475 \fBzfs_vdev_scrub_max_active\fR (int)
1478 Maximum scrub I/Os active to each device.
1479 See the section "ZFS I/O SCHEDULER".
1481 Default value: \fB2\fR.
1487 \fBzfs_vdev_scrub_min_active\fR (int)
1490 Minimum scrub I/Os active to each device.
1491 See the section "ZFS I/O SCHEDULER".
1493 Default value: \fB1\fR.
1499 \fBzfs_vdev_sync_read_max_active\fR (int)
1502 Maximum synchronous read I/Os active to each device.
1503 See the section "ZFS I/O SCHEDULER".
1505 Default value: \fB10\fR.
1511 \fBzfs_vdev_sync_read_min_active\fR (int)
1514 Minimum synchronous read I/Os active to each device.
1515 See the section "ZFS I/O SCHEDULER".
1517 Default value: \fB10\fR.
1523 \fBzfs_vdev_sync_write_max_active\fR (int)
1526 Maximum synchronous write I/Os active to each device.
1527 See the section "ZFS I/O SCHEDULER".
1529 Default value: \fB10\fR.
1535 \fBzfs_vdev_sync_write_min_active\fR (int)
1538 Minimum synchronous write I/Os active to each device.
1539 See the section "ZFS I/O SCHEDULER".
1541 Default value: \fB10\fR.
1547 \fBzfs_vdev_trim_max_active\fR (int)
1550 Maximum trim/discard I/Os active to each device.
1551 See the section "ZFS I/O SCHEDULER".
1553 Default value: \fB2\fR.
1559 \fBzfs_vdev_trim_min_active\fR (int)
1562 Minimum trim/discard I/Os active to each device.
1563 See the section "ZFS I/O SCHEDULER".
1565 Default value: \fB1\fR.
1571 \fBzfs_vdev_queue_depth_pct\fR (int)
1574 Maximum number of queued allocations per top-level vdev expressed as
1575 a percentage of \fBzfs_vdev_async_write_max_active\fR which allows the
1576 system to detect devices that are more capable of handling allocations
1577 and to allocate more blocks to those devices. It allows for dynamic
1578 allocation distribution when devices are imbalanced as fuller devices
1579 will tend to be slower than empty devices.
1581 See also \fBzio_dva_throttle_enabled\fR.
1583 Default value: \fB1000\fR%.
1589 \fBzfs_expire_snapshot\fR (int)
1592 Seconds to expire .zfs/snapshot
1594 Default value: \fB300\fR.
1600 \fBzfs_admin_snapshot\fR (int)
1603 Allow the creation, removal, or renaming of entries in the .zfs/snapshot
1604 directory to cause the creation, destruction, or renaming of snapshots.
1605 When enabled this functionality works both locally and over NFS exports
1606 which have the 'no_root_squash' option set. This functionality is disabled
1609 Use \fB1\fR for yes and \fB0\fR for no (default).
1615 \fBzfs_flags\fR (int)
1618 Set additional debugging flags. The following flags may be bitwise-or'd
1630 Enable dprintf entries in the debug log.
1632 2 ZFS_DEBUG_DBUF_VERIFY *
1633 Enable extra dbuf verifications.
1635 4 ZFS_DEBUG_DNODE_VERIFY *
1636 Enable extra dnode verifications.
1638 8 ZFS_DEBUG_SNAPNAMES
1639 Enable snapshot name verification.
1642 Check for illegally modified ARC buffers.
1644 64 ZFS_DEBUG_ZIO_FREE
1645 Enable verification of block frees.
1647 128 ZFS_DEBUG_HISTOGRAM_VERIFY
1648 Enable extra spacemap histogram verifications.
1650 256 ZFS_DEBUG_METASLAB_VERIFY
1651 Verify space accounting on disk matches in-core range_trees.
1653 512 ZFS_DEBUG_SET_ERROR
1654 Enable SET_ERROR and dprintf entries in the debug log.
1656 1024 ZFS_DEBUG_INDIRECT_REMAP
1657 Verify split blocks created by device removal.
1660 Verify TRIM ranges are always within the allocatable range tree.
1663 * Requires debug build.
1665 Default value: \fB0\fR.
1671 \fBzfs_free_leak_on_eio\fR (int)
1674 If destroy encounters an EIO while reading metadata (e.g. indirect
1675 blocks), space referenced by the missing metadata can not be freed.
1676 Normally this causes the background destroy to become "stalled", as
1677 it is unable to make forward progress. While in this stalled state,
1678 all remaining space to free from the error-encountering filesystem is
1679 "temporarily leaked". Set this flag to cause it to ignore the EIO,
1680 permanently leak the space from indirect blocks that can not be read,
1681 and continue to free everything else that it can.
1683 The default, "stalling" behavior is useful if the storage partially
1684 fails (i.e. some but not all i/os fail), and then later recovers. In
1685 this case, we will be able to continue pool operations while it is
1686 partially failed, and when it recovers, we can continue to free the
1687 space, with no leaks. However, note that this case is actually
1690 Typically pools either (a) fail completely (but perhaps temporarily,
1691 e.g. a top-level vdev going offline), or (b) have localized,
1692 permanent errors (e.g. disk returns the wrong data due to bit flip or
1693 firmware bug). In case (a), this setting does not matter because the
1694 pool will be suspended and the sync thread will not be able to make
1695 forward progress regardless. In case (b), because the error is
1696 permanent, the best we can do is leak the minimum amount of space,
1697 which is what setting this flag will do. Therefore, it is reasonable
1698 for this flag to normally be set, but we chose the more conservative
1699 approach of not setting it, so that there is no possibility of
1700 leaking space in the "partial temporary" failure case.
1702 Default value: \fB0\fR.
1708 \fBzfs_free_min_time_ms\fR (int)
1711 During a \fBzfs destroy\fR operation using \fBfeature@async_destroy\fR a minimum
1712 of this much time will be spent working on freeing blocks per txg.
1714 Default value: \fB1,000\fR.
1720 \fBzfs_immediate_write_sz\fR (long)
1723 Largest data block to write to zil. Larger blocks will be treated as if the
1724 dataset being written to had the property setting \fBlogbias=throughput\fR.
1726 Default value: \fB32,768\fR.
1732 \fBzfs_initialize_value\fR (ulong)
1735 Pattern written to vdev free space by \fBzpool initialize\fR.
1737 Default value: \fB16,045,690,984,833,335,022\fR (0xdeadbeefdeadbeee).
1743 \fBzfs_lua_max_instrlimit\fR (ulong)
1746 The maximum execution time limit that can be set for a ZFS channel program,
1747 specified as a number of Lua instructions.
1749 Default value: \fB100,000,000\fR.
1755 \fBzfs_lua_max_memlimit\fR (ulong)
1758 The maximum memory limit that can be set for a ZFS channel program, specified
1761 Default value: \fB104,857,600\fR.
1767 \fBzfs_max_dataset_nesting\fR (int)
1770 The maximum depth of nested datasets. This value can be tuned temporarily to
1771 fix existing datasets that exceed the predefined limit.
1773 Default value: \fB50\fR.
1779 \fBzfs_max_recordsize\fR (int)
1782 We currently support block sizes from 512 bytes to 16MB. The benefits of
1783 larger blocks, and thus larger I/O, need to be weighed against the cost of
1784 COWing a giant block to modify one byte. Additionally, very large blocks
1785 can have an impact on i/o latency, and also potentially on the memory
1786 allocator. Therefore, we do not allow the recordsize to be set larger than
1787 zfs_max_recordsize (default 1MB). Larger blocks can be created by changing
1788 this tunable, and pools with larger blocks can always be imported and used,
1789 regardless of this setting.
1791 Default value: \fB1,048,576\fR.
1797 \fBzfs_metaslab_fragmentation_threshold\fR (int)
1800 Allow metaslabs to keep their active state as long as their fragmentation
1801 percentage is less than or equal to this value. An active metaslab that
1802 exceeds this threshold will no longer keep its active status allowing
1803 better metaslabs to be selected.
1805 Default value: \fB70\fR.
1811 \fBzfs_mg_fragmentation_threshold\fR (int)
1814 Metaslab groups are considered eligible for allocations if their
1815 fragmentation metric (measured as a percentage) is less than or equal to
1816 this value. If a metaslab group exceeds this threshold then it will be
1817 skipped unless all metaslab groups within the metaslab class have also
1818 crossed this threshold.
1820 Default value: \fB85\fR.
1826 \fBzfs_mg_noalloc_threshold\fR (int)
1829 Defines a threshold at which metaslab groups should be eligible for
1830 allocations. The value is expressed as a percentage of free space
1831 beyond which a metaslab group is always eligible for allocations.
1832 If a metaslab group's free space is less than or equal to the
1833 threshold, the allocator will avoid allocating to that group
1834 unless all groups in the pool have reached the threshold. Once all
1835 groups have reached the threshold, all groups are allowed to accept
1836 allocations. The default value of 0 disables the feature and causes
1837 all metaslab groups to be eligible for allocations.
1839 This parameter allows one to deal with pools having heavily imbalanced
1840 vdevs such as would be the case when a new vdev has been added.
1841 Setting the threshold to a non-zero percentage will stop allocations
1842 from being made to vdevs that aren't filled to the specified percentage
1843 and allow lesser filled vdevs to acquire more allocations than they
1844 otherwise would under the old \fBzfs_mg_alloc_failures\fR facility.
1846 Default value: \fB0\fR.
1852 \fBzfs_ddt_data_is_special\fR (int)
1855 If enabled, ZFS will place DDT data into the special allocation class.
1857 Default value: \fB1\fR.
1863 \fBzfs_user_indirect_is_special\fR (int)
1866 If enabled, ZFS will place user data (both file and zvol) indirect blocks
1867 into the special allocation class.
1869 Default value: \fB1\fR.
1875 \fBzfs_multihost_history\fR (int)
1878 Historical statistics for the last N multihost updates will be available in
1879 \fB/proc/spl/kstat/zfs/<pool>/multihost\fR
1881 Default value: \fB0\fR.
1887 \fBzfs_multihost_interval\fR (ulong)
1890 Used to control the frequency of multihost writes which are performed when the
1891 \fBmultihost\fR pool property is on. This is one factor used to determine the
1892 length of the activity check during import.
1894 The multihost write period is \fBzfs_multihost_interval / leaf-vdevs\fR
1895 milliseconds. On average a multihost write will be issued for each leaf vdev
1896 every \fBzfs_multihost_interval\fR milliseconds. In practice, the observed
1897 period can vary with the I/O load and this observed value is the delay which is
1898 stored in the uberblock.
1900 Default value: \fB1000\fR.
1906 \fBzfs_multihost_import_intervals\fR (uint)
1909 Used to control the duration of the activity test on import. Smaller values of
1910 \fBzfs_multihost_import_intervals\fR will reduce the import time but increase
1911 the risk of failing to detect an active pool. The total activity check time is
1912 never allowed to drop below one second.
1914 On import the activity check waits a minimum amount of time determined by
1915 \fBzfs_multihost_interval * zfs_multihost_import_intervals\fR, or the same
1916 product computed on the host which last had the pool imported (whichever is
1917 greater). The activity check time may be further extended if the value of mmp
1918 delay found in the best uberblock indicates actual multihost updates happened
1919 at longer intervals than \fBzfs_multihost_interval\fR. A minimum value of
1920 \fB100ms\fR is enforced.
1922 A value of 0 is ignored and treated as if it was set to 1.
1924 Default value: \fB20\fR.
1930 \fBzfs_multihost_fail_intervals\fR (uint)
1933 Controls the behavior of the pool when multihost write failures or delays are
1936 When \fBzfs_multihost_fail_intervals = 0\fR, multihost write failures or delays
1937 are ignored. The failures will still be reported to the ZED which depending on
1938 its configuration may take action such as suspending the pool or offlining a
1942 When \fBzfs_multihost_fail_intervals > 0\fR, the pool will be suspended if
1943 \fBzfs_multihost_fail_intervals * zfs_multihost_interval\fR milliseconds pass
1944 without a successful mmp write. This guarantees the activity test will see
1945 mmp writes if the pool is imported. A value of 1 is ignored and treated as
1946 if it was set to 2. This is necessary to prevent the pool from being suspended
1947 due to normal, small I/O latency variations.
1950 Default value: \fB10\fR.
1956 \fBzfs_no_scrub_io\fR (int)
1959 Set for no scrub I/O. This results in scrubs not actually scrubbing data and
1960 simply doing a metadata crawl of the pool instead.
1962 Use \fB1\fR for yes and \fB0\fR for no (default).
1968 \fBzfs_no_scrub_prefetch\fR (int)
1971 Set to disable block prefetching for scrubs.
1973 Use \fB1\fR for yes and \fB0\fR for no (default).
1979 \fBzfs_nocacheflush\fR (int)
1982 Disable cache flush operations on disks when writing. Setting this will
1983 cause pool corruption on power loss if a volatile out-of-order write cache
1986 Use \fB1\fR for yes and \fB0\fR for no (default).
1992 \fBzfs_nopwrite_enabled\fR (int)
1997 Use \fB1\fR for yes (default) and \fB0\fR to disable.
2003 \fBzfs_dmu_offset_next_sync\fR (int)
2006 Enable forcing txg sync to find holes. When enabled forces ZFS to act
2007 like prior versions when SEEK_HOLE or SEEK_DATA flags are used, which
2008 when a dnode is dirty causes txg's to be synced so that this data can be
2011 Use \fB1\fR for yes and \fB0\fR to disable (default).
2017 \fBzfs_pd_bytes_max\fR (int)
2020 The number of bytes which should be prefetched during a pool traversal
2021 (eg: \fBzfs send\fR or other data crawling operations)
2023 Default value: \fB52,428,800\fR.
2029 \fBzfs_per_txg_dirty_frees_percent \fR (ulong)
2032 Tunable to control percentage of dirtied indirect blocks from frees allowed
2033 into one TXG. After this threshold is crossed, additional frees will wait until
2035 A value of zero will disable this throttle.
2037 Default value: \fB5\fR, set to \fB0\fR to disable.
2043 \fBzfs_prefetch_disable\fR (int)
2046 This tunable disables predictive prefetch. Note that it leaves "prescient"
2047 prefetch (e.g. prefetch for zfs send) intact. Unlike predictive prefetch,
2048 prescient prefetch never issues i/os that end up not being needed, so it
2049 can't hurt performance.
2051 Use \fB1\fR for yes and \fB0\fR for no (default).
2057 \fBzfs_qat_checksum_disable\fR (int)
2060 This tunable disables qat hardware acceleration for sha256 checksums. It
2061 may be set after the zfs modules have been loaded to initialize the qat
2062 hardware as long as support is compiled in and the qat driver is present.
2064 Use \fB1\fR for yes and \fB0\fR for no (default).
2070 \fBzfs_qat_compress_disable\fR (int)
2073 This tunable disables qat hardware acceleration for gzip compression. It
2074 may be set after the zfs modules have been loaded to initialize the qat
2075 hardware as long as support is compiled in and the qat driver is present.
2077 Use \fB1\fR for yes and \fB0\fR for no (default).
2083 \fBzfs_qat_encrypt_disable\fR (int)
2086 This tunable disables qat hardware acceleration for AES-GCM encryption. It
2087 may be set after the zfs modules have been loaded to initialize the qat
2088 hardware as long as support is compiled in and the qat driver is present.
2090 Use \fB1\fR for yes and \fB0\fR for no (default).
2096 \fBzfs_read_chunk_size\fR (long)
2099 Bytes to read per chunk
2101 Default value: \fB1,048,576\fR.
2107 \fBzfs_read_history\fR (int)
2110 Historical statistics for the last N reads will be available in
2111 \fB/proc/spl/kstat/zfs/<pool>/reads\fR
2113 Default value: \fB0\fR (no data is kept).
2119 \fBzfs_read_history_hits\fR (int)
2122 Include cache hits in read history
2124 Use \fB1\fR for yes and \fB0\fR for no (default).
2130 \fBzfs_reconstruct_indirect_combinations_max\fR (int)
2133 If an indirect split block contains more than this many possible unique
2134 combinations when being reconstructed, consider it too computationally
2135 expensive to check them all. Instead, try at most
2136 \fBzfs_reconstruct_indirect_combinations_max\fR randomly-selected
2137 combinations each time the block is accessed. This allows all segment
2138 copies to participate fairly in the reconstruction when all combinations
2139 cannot be checked and prevents repeated use of one bad copy.
2141 Default value: \fB4096\fR.
2147 \fBzfs_recover\fR (int)
2150 Set to attempt to recover from fatal errors. This should only be used as a
2151 last resort, as it typically results in leaked space, or worse.
2153 Use \fB1\fR for yes and \fB0\fR for no (default).
2159 \fBzfs_removal_ignore_errors\fR (int)
2163 Ignore hard IO errors during device removal. When set, if a device encounters
2164 a hard IO error during the removal process the removal will not be cancelled.
2165 This can result in a normally recoverable block becoming permanently damaged
2166 and is not recommended. This should only be used as a last resort when the
2167 pool cannot be returned to a healthy state prior to removing the device.
2169 Default value: \fB0\fR.
2175 \fBzfs_resilver_min_time_ms\fR (int)
2178 Resilvers are processed by the sync thread. While resilvering it will spend
2179 at least this much time working on a resilver between txg flushes.
2181 Default value: \fB3,000\fR.
2187 \fBzfs_scan_ignore_errors\fR (int)
2190 If set to a nonzero value, remove the DTL (dirty time list) upon
2191 completion of a pool scan (scrub) even if there were unrepairable
2192 errors. It is intended to be used during pool repair or recovery to
2193 stop resilvering when the pool is next imported.
2195 Default value: \fB0\fR.
2201 \fBzfs_scrub_min_time_ms\fR (int)
2204 Scrubs are processed by the sync thread. While scrubbing it will spend
2205 at least this much time working on a scrub between txg flushes.
2207 Default value: \fB1,000\fR.
2213 \fBzfs_scan_checkpoint_intval\fR (int)
2216 To preserve progress across reboots the sequential scan algorithm periodically
2217 needs to stop metadata scanning and issue all the verifications I/Os to disk.
2218 The frequency of this flushing is determined by the
2219 \fBzfs_scan_checkpoint_intval\fR tunable.
2221 Default value: \fB7200\fR seconds (every 2 hours).
2227 \fBzfs_scan_fill_weight\fR (int)
2230 This tunable affects how scrub and resilver I/O segments are ordered. A higher
2231 number indicates that we care more about how filled in a segment is, while a
2232 lower number indicates we care more about the size of the extent without
2233 considering the gaps within a segment. This value is only tunable upon module
2234 insertion. Changing the value afterwards will have no affect on scrub or
2235 resilver performance.
2237 Default value: \fB3\fR.
2243 \fBzfs_scan_issue_strategy\fR (int)
2246 Determines the order that data will be verified while scrubbing or resilvering.
2247 If set to \fB1\fR, data will be verified as sequentially as possible, given the
2248 amount of memory reserved for scrubbing (see \fBzfs_scan_mem_lim_fact\fR). This
2249 may improve scrub performance if the pool's data is very fragmented. If set to
2250 \fB2\fR, the largest mostly-contiguous chunk of found data will be verified
2251 first. By deferring scrubbing of small segments, we may later find adjacent data
2252 to coalesce and increase the segment size. If set to \fB0\fR, zfs will use
2253 strategy \fB1\fR during normal verification and strategy \fB2\fR while taking a
2256 Default value: \fB0\fR.
2262 \fBzfs_scan_legacy\fR (int)
2265 A value of 0 indicates that scrubs and resilvers will gather metadata in
2266 memory before issuing sequential I/O. A value of 1 indicates that the legacy
2267 algorithm will be used where I/O is initiated as soon as it is discovered.
2268 Changing this value to 0 will not affect scrubs or resilvers that are already
2271 Default value: \fB0\fR.
2277 \fBzfs_scan_max_ext_gap\fR (int)
2280 Indicates the largest gap in bytes between scrub / resilver I/Os that will still
2281 be considered sequential for sorting purposes. Changing this value will not
2282 affect scrubs or resilvers that are already in progress.
2284 Default value: \fB2097152 (2 MB)\fR.
2290 \fBzfs_scan_mem_lim_fact\fR (int)
2293 Maximum fraction of RAM used for I/O sorting by sequential scan algorithm.
2294 This tunable determines the hard limit for I/O sorting memory usage.
2295 When the hard limit is reached we stop scanning metadata and start issuing
2296 data verification I/O. This is done until we get below the soft limit.
2298 Default value: \fB20\fR which is 5% of RAM (1/20).
2304 \fBzfs_scan_mem_lim_soft_fact\fR (int)
2307 The fraction of the hard limit used to determined the soft limit for I/O sorting
2308 by the sequential scan algorithm. When we cross this limit from bellow no action
2309 is taken. When we cross this limit from above it is because we are issuing
2310 verification I/O. In this case (unless the metadata scan is done) we stop
2311 issuing verification I/O and start scanning metadata again until we get to the
2314 Default value: \fB20\fR which is 5% of the hard limit (1/20).
2320 \fBzfs_scan_vdev_limit\fR (int)
2323 Maximum amount of data that can be concurrently issued at once for scrubs and
2324 resilvers per leaf device, given in bytes.
2326 Default value: \fB41943040\fR.
2332 \fBzfs_send_corrupt_data\fR (int)
2335 Allow sending of corrupt data (ignore read/checksum errors when sending data)
2337 Use \fB1\fR for yes and \fB0\fR for no (default).
2343 \fBzfs_send_unmodified_spill_blocks\fR (int)
2346 Include unmodified spill blocks in the send stream. Under certain circumstances
2347 previous versions of ZFS could incorrectly remove the spill block from an
2348 existing object. Including unmodified copies of the spill blocks creates a
2349 backwards compatible stream which will recreate a spill block if it was
2350 incorrectly removed.
2352 Use \fB1\fR for yes (default) and \fB0\fR for no.
2358 \fBzfs_send_queue_length\fR (int)
2361 The maximum number of bytes allowed in the \fBzfs send\fR queue. This value
2362 must be at least twice the maximum block size in use.
2364 Default value: \fB16,777,216\fR.
2370 \fBzfs_recv_queue_length\fR (int)
2373 The maximum number of bytes allowed in the \fBzfs receive\fR queue. This value
2374 must be at least twice the maximum block size in use.
2376 Default value: \fB16,777,216\fR.
2382 \fBzfs_sync_pass_deferred_free\fR (int)
2385 Flushing of data to disk is done in passes. Defer frees starting in this pass
2387 Default value: \fB2\fR.
2393 \fBzfs_spa_discard_memory_limit\fR (int)
2396 Maximum memory used for prefetching a checkpoint's space map on each
2397 vdev while discarding the checkpoint.
2399 Default value: \fB16,777,216\fR.
2405 \fBzfs_sync_pass_dont_compress\fR (int)
2408 Don't compress starting in this pass
2410 Default value: \fB5\fR.
2416 \fBzfs_sync_pass_rewrite\fR (int)
2419 Rewrite new block pointers starting in this pass
2421 Default value: \fB2\fR.
2427 \fBzfs_sync_taskq_batch_pct\fR (int)
2430 This controls the number of threads used by the dp_sync_taskq. The default
2431 value of 75% will create a maximum of one thread per cpu.
2433 Default value: \fB75\fR%.
2439 \fBzfs_trim_extent_bytes_max\fR (unsigned int)
2442 Maximum size of TRIM command. Ranges larger than this will be split in to
2443 chunks no larger than \fBzfs_trim_extent_bytes_max\fR bytes before being
2444 issued to the device.
2446 Default value: \fB134,217,728\fR.
2452 \fBzfs_trim_extent_bytes_min\fR (unsigned int)
2455 Minimum size of TRIM commands. TRIM ranges smaller than this will be skipped
2456 unless they're part of a larger range which was broken in to chunks. This is
2457 done because it's common for these small TRIMs to negatively impact overall
2458 performance. This value can be set to 0 to TRIM all unallocated space.
2460 Default value: \fB32,768\fR.
2466 \fBzfs_trim_metaslab_skip\fR (unsigned int)
2469 Skip uninitialized metaslabs during the TRIM process. This option is useful
2470 for pools constructed from large thinly-provisioned devices where TRIM
2471 operations are slow. As a pool ages an increasing fraction of the pools
2472 metaslabs will be initialized progressively degrading the usefulness of
2473 this option. This setting is stored when starting a manual TRIM and will
2474 persist for the duration of the requested TRIM.
2476 Default value: \fB0\fR.
2482 \fBzfs_trim_queue_limit\fR (unsigned int)
2485 Maximum number of queued TRIMs outstanding per leaf vdev. The number of
2486 concurrent TRIM commands issued to the device is controlled by the
2487 \fBzfs_vdev_trim_min_active\fR and \fBzfs_vdev_trim_max_active\fR module
2490 Default value: \fB10\fR.
2496 \fBzfs_trim_txg_batch\fR (unsigned int)
2499 The number of transaction groups worth of frees which should be aggregated
2500 before TRIM operations are issued to the device. This setting represents a
2501 trade-off between issuing larger, more efficient TRIM operations and the
2502 delay before the recently trimmed space is available for use by the device.
2504 Increasing this value will allow frees to be aggregated for a longer time.
2505 This will result is larger TRIM operations and potentially increased memory
2506 usage. Decreasing this value will have the opposite effect. The default
2507 value of 32 was determined to be a reasonable compromise.
2509 Default value: \fB32\fR.
2515 \fBzfs_txg_history\fR (int)
2518 Historical statistics for the last N txgs will be available in
2519 \fB/proc/spl/kstat/zfs/<pool>/txgs\fR
2521 Default value: \fB0\fR.
2527 \fBzfs_txg_timeout\fR (int)
2530 Flush dirty data to disk at least every N seconds (maximum txg duration)
2532 Default value: \fB5\fR.
2538 \fBzfs_vdev_aggregate_trim\fR (int)
2541 Allow TRIM I/Os to be aggregated. This is normally not helpful because
2542 the extents to be trimmed will have been already been aggregated by the
2543 metaslab. This option is provided for debugging and performance analysis.
2545 Default value: \fB0\fR.
2551 \fBzfs_vdev_aggregation_limit\fR (int)
2554 Max vdev I/O aggregation size
2556 Default value: \fB1,048,576\fR.
2562 \fBzfs_vdev_aggregation_limit_non_rotating\fR (int)
2565 Max vdev I/O aggregation size for non-rotating media
2567 Default value: \fB131,072\fR.
2573 \fBzfs_vdev_cache_bshift\fR (int)
2576 Shift size to inflate reads too
2578 Default value: \fB16\fR (effectively 65536).
2584 \fBzfs_vdev_cache_max\fR (int)
2587 Inflate reads smaller than this value to meet the \fBzfs_vdev_cache_bshift\fR
2590 Default value: \fB16384\fR.
2596 \fBzfs_vdev_cache_size\fR (int)
2599 Total size of the per-disk cache in bytes.
2601 Currently this feature is disabled as it has been found to not be helpful
2602 for performance and in some cases harmful.
2604 Default value: \fB0\fR.
2610 \fBzfs_vdev_mirror_rotating_inc\fR (int)
2613 A number by which the balancing algorithm increments the load calculation for
2614 the purpose of selecting the least busy mirror member when an I/O immediately
2615 follows its predecessor on rotational vdevs for the purpose of making decisions
2618 Default value: \fB0\fR.
2624 \fBzfs_vdev_mirror_rotating_seek_inc\fR (int)
2627 A number by which the balancing algorithm increments the load calculation for
2628 the purpose of selecting the least busy mirror member when an I/O lacks
2629 locality as defined by the zfs_vdev_mirror_rotating_seek_offset. I/Os within
2630 this that are not immediately following the previous I/O are incremented by
2633 Default value: \fB5\fR.
2639 \fBzfs_vdev_mirror_rotating_seek_offset\fR (int)
2642 The maximum distance for the last queued I/O in which the balancing algorithm
2643 considers an I/O to have locality.
2644 See the section "ZFS I/O SCHEDULER".
2646 Default value: \fB1048576\fR.
2652 \fBzfs_vdev_mirror_non_rotating_inc\fR (int)
2655 A number by which the balancing algorithm increments the load calculation for
2656 the purpose of selecting the least busy mirror member on non-rotational vdevs
2657 when I/Os do not immediately follow one another.
2659 Default value: \fB0\fR.
2665 \fBzfs_vdev_mirror_non_rotating_seek_inc\fR (int)
2668 A number by which the balancing algorithm increments the load calculation for
2669 the purpose of selecting the least busy mirror member when an I/O lacks
2670 locality as defined by the zfs_vdev_mirror_rotating_seek_offset. I/Os within
2671 this that are not immediately following the previous I/O are incremented by
2674 Default value: \fB1\fR.
2680 \fBzfs_vdev_read_gap_limit\fR (int)
2683 Aggregate read I/O operations if the gap on-disk between them is within this
2686 Default value: \fB32,768\fR.
2692 \fBzfs_vdev_scheduler\fR (charp)
2695 Set the Linux I/O scheduler on whole disk vdevs to this scheduler. Valid options
2696 are noop, cfq, bfq & deadline
2698 Default value: \fBnoop\fR.
2704 \fBzfs_vdev_write_gap_limit\fR (int)
2707 Aggregate write I/O over gap
2709 Default value: \fB4,096\fR.
2715 \fBzfs_vdev_raidz_impl\fR (string)
2718 Parameter for selecting raidz parity implementation to use.
2720 Options marked (always) below may be selected on module load as they are
2721 supported on all systems.
2722 The remaining options may only be set after the module is loaded, as they
2723 are available only if the implementations are compiled in and supported
2724 on the running system.
2726 Once the module is loaded, the content of
2727 /sys/module/zfs/parameters/zfs_vdev_raidz_impl will show available options
2728 with the currently selected one enclosed in [].
2729 Possible options are:
2730 fastest - (always) implementation selected using built-in benchmark
2731 original - (always) original raidz implementation
2732 scalar - (always) scalar raidz implementation
2733 sse2 - implementation using SSE2 instruction set (64bit x86 only)
2734 ssse3 - implementation using SSSE3 instruction set (64bit x86 only)
2735 avx2 - implementation using AVX2 instruction set (64bit x86 only)
2736 avx512f - implementation using AVX512F instruction set (64bit x86 only)
2737 avx512bw - implementation using AVX512F & AVX512BW instruction sets (64bit x86 only)
2738 aarch64_neon - implementation using NEON (Aarch64/64 bit ARMv8 only)
2739 aarch64_neonx2 - implementation using NEON with more unrolling (Aarch64/64 bit ARMv8 only)
2741 Default value: \fBfastest\fR.
2747 \fBzfs_zevent_cols\fR (int)
2750 When zevents are logged to the console use this as the word wrap width.
2752 Default value: \fB80\fR.
2758 \fBzfs_zevent_console\fR (int)
2761 Log events to the console
2763 Use \fB1\fR for yes and \fB0\fR for no (default).
2769 \fBzfs_zevent_len_max\fR (int)
2772 Max event queue length. A value of 0 will result in a calculated value which
2773 increases with the number of CPUs in the system (minimum 64 events). Events
2774 in the queue can be viewed with the \fBzpool events\fR command.
2776 Default value: \fB0\fR.
2782 \fBzfs_zil_clean_taskq_maxalloc\fR (int)
2785 The maximum number of taskq entries that are allowed to be cached. When this
2786 limit is exceeded transaction records (itxs) will be cleaned synchronously.
2788 Default value: \fB1048576\fR.
2794 \fBzfs_zil_clean_taskq_minalloc\fR (int)
2797 The number of taskq entries that are pre-populated when the taskq is first
2798 created and are immediately available for use.
2800 Default value: \fB1024\fR.
2806 \fBzfs_zil_clean_taskq_nthr_pct\fR (int)
2809 This controls the number of threads used by the dp_zil_clean_taskq. The default
2810 value of 100% will create a maximum of one thread per cpu.
2812 Default value: \fB100\fR%.
2818 \fBzil_nocacheflush\fR (int)
2821 Disable the cache flush commands that are normally sent to the disk(s) by
2822 the ZIL after an LWB write has completed. Setting this will cause ZIL
2823 corruption on power loss if a volatile out-of-order write cache is enabled.
2825 Use \fB1\fR for yes and \fB0\fR for no (default).
2831 \fBzil_replay_disable\fR (int)
2834 Disable intent logging replay. Can be disabled for recovery from corrupted
2837 Use \fB1\fR for yes and \fB0\fR for no (default).
2843 \fBzil_slog_bulk\fR (ulong)
2846 Limit SLOG write size per commit executed with synchronous priority.
2847 Any writes above that will be executed with lower (asynchronous) priority
2848 to limit potential SLOG device abuse by single active ZIL writer.
2850 Default value: \fB786,432\fR.
2856 \fBzio_deadman_log_all\fR (int)
2859 If non-zero, the zio deadman will produce debugging messages (see
2860 \fBzfs_dbgmsg_enable\fR) for all zios, rather than only for leaf
2861 zios possessing a vdev. This is meant to be used by developers to gain
2862 diagnostic information for hang conditions which don't involve a mutex
2863 or other locking primitive; typically conditions in which a thread in
2864 the zio pipeline is looping indefinitely.
2866 Default value: \fB0\fR.
2872 \fBzio_decompress_fail_fraction\fR (int)
2875 If non-zero, this value represents the denominator of the probability that zfs
2876 should induce a decompression failure. For instance, for a 5% decompression
2877 failure rate, this value should be set to 20.
2879 Default value: \fB0\fR.
2885 \fBzio_slow_io_ms\fR (int)
2888 When an I/O operation takes more than \fBzio_slow_io_ms\fR milliseconds to
2889 complete is marked as a slow I/O. Each slow I/O causes a delay zevent. Slow
2890 I/O counters can be seen with "zpool status -s".
2893 Default value: \fB30,000\fR.
2899 \fBzio_dva_throttle_enabled\fR (int)
2902 Throttle block allocations in the I/O pipeline. This allows for
2903 dynamic allocation distribution when devices are imbalanced.
2904 When enabled, the maximum number of pending allocations per top-level vdev
2905 is limited by \fBzfs_vdev_queue_depth_pct\fR.
2907 Default value: \fB1\fR.
2913 \fBzio_requeue_io_start_cut_in_line\fR (int)
2916 Prioritize requeued I/O
2918 Default value: \fB0\fR.
2924 \fBzio_taskq_batch_pct\fR (uint)
2927 Percentage of online CPUs (or CPU cores, etc) which will run a worker thread
2928 for I/O. These workers are responsible for I/O work such as compression and
2929 checksum calculations. Fractional number of CPUs will be rounded down.
2931 The default value of 75 was chosen to avoid using all CPUs which can result in
2932 latency issues and inconsistent application performance, especially when high
2933 compression is enabled.
2935 Default value: \fB75\fR.
2941 \fBzvol_inhibit_dev\fR (uint)
2944 Do not create zvol device nodes. This may slightly improve startup time on
2945 systems with a very large number of zvols.
2947 Use \fB1\fR for yes and \fB0\fR for no (default).
2953 \fBzvol_major\fR (uint)
2956 Major number for zvol block devices
2958 Default value: \fB230\fR.
2964 \fBzvol_max_discard_blocks\fR (ulong)
2967 Discard (aka TRIM) operations done on zvols will be done in batches of this
2968 many blocks, where block size is determined by the \fBvolblocksize\fR property
2971 Default value: \fB16,384\fR.
2977 \fBzvol_prefetch_bytes\fR (uint)
2980 When adding a zvol to the system prefetch \fBzvol_prefetch_bytes\fR
2981 from the start and end of the volume. Prefetching these regions
2982 of the volume is desirable because they are likely to be accessed
2983 immediately by \fBblkid(8)\fR or by the kernel scanning for a partition
2986 Default value: \fB131,072\fR.
2992 \fBzvol_request_sync\fR (uint)
2995 When processing I/O requests for a zvol submit them synchronously. This
2996 effectively limits the queue depth to 1 for each I/O submitter. When set
2997 to 0 requests are handled asynchronously by a thread pool. The number of
2998 requests which can be handled concurrently is controller by \fBzvol_threads\fR.
3000 Default value: \fB0\fR.
3006 \fBzvol_threads\fR (uint)
3009 Max number of threads which can handle zvol I/O requests concurrently.
3011 Default value: \fB32\fR.
3017 \fBzvol_volmode\fR (uint)
3020 Defines zvol block devices behaviour when \fBvolmode\fR is set to \fBdefault\fR.
3021 Valid values are \fB1\fR (full), \fB2\fR (dev) and \fB3\fR (none).
3023 Default value: \fB1\fR.
3026 .SH ZFS I/O SCHEDULER
3027 ZFS issues I/O operations to leaf vdevs to satisfy and complete I/Os.
3028 The I/O scheduler determines when and in what order those operations are
3029 issued. The I/O scheduler divides operations into five I/O classes
3030 prioritized in the following order: sync read, sync write, async read,
3031 async write, and scrub/resilver. Each queue defines the minimum and
3032 maximum number of concurrent operations that may be issued to the
3033 device. In addition, the device has an aggregate maximum,
3034 \fBzfs_vdev_max_active\fR. Note that the sum of the per-queue minimums
3035 must not exceed the aggregate maximum. If the sum of the per-queue
3036 maximums exceeds the aggregate maximum, then the number of active I/Os
3037 may reach \fBzfs_vdev_max_active\fR, in which case no further I/Os will
3038 be issued regardless of whether all per-queue minimums have been met.
3040 For many physical devices, throughput increases with the number of
3041 concurrent operations, but latency typically suffers. Further, physical
3042 devices typically have a limit at which more concurrent operations have no
3043 effect on throughput or can actually cause it to decrease.
3045 The scheduler selects the next operation to issue by first looking for an
3046 I/O class whose minimum has not been satisfied. Once all are satisfied and
3047 the aggregate maximum has not been hit, the scheduler looks for classes
3048 whose maximum has not been satisfied. Iteration through the I/O classes is
3049 done in the order specified above. No further operations are issued if the
3050 aggregate maximum number of concurrent operations has been hit or if there
3051 are no operations queued for an I/O class that has not hit its maximum.
3052 Every time an I/O is queued or an operation completes, the I/O scheduler
3053 looks for new operations to issue.
3055 In general, smaller max_active's will lead to lower latency of synchronous
3056 operations. Larger max_active's may lead to higher overall throughput,
3057 depending on underlying storage.
3059 The ratio of the queues' max_actives determines the balance of performance
3060 between reads, writes, and scrubs. E.g., increasing
3061 \fBzfs_vdev_scrub_max_active\fR will cause the scrub or resilver to complete
3062 more quickly, but reads and writes to have higher latency and lower throughput.
3064 All I/O classes have a fixed maximum number of outstanding operations
3065 except for the async write class. Asynchronous writes represent the data
3066 that is committed to stable storage during the syncing stage for
3067 transaction groups. Transaction groups enter the syncing state
3068 periodically so the number of queued async writes will quickly burst up
3069 and then bleed down to zero. Rather than servicing them as quickly as
3070 possible, the I/O scheduler changes the maximum number of active async
3071 write I/Os according to the amount of dirty data in the pool. Since
3072 both throughput and latency typically increase with the number of
3073 concurrent operations issued to physical devices, reducing the
3074 burstiness in the number of concurrent operations also stabilizes the
3075 response time of operations from other -- and in particular synchronous
3076 -- queues. In broad strokes, the I/O scheduler will issue more
3077 concurrent operations from the async write queue as there's more dirty
3082 The number of concurrent operations issued for the async write I/O class
3083 follows a piece-wise linear function defined by a few adjustable points.
3086 | o---------| <-- zfs_vdev_async_write_max_active
3093 |-------o | | <-- zfs_vdev_async_write_min_active
3094 0|_______^______|_________|
3095 0% | | 100% of zfs_dirty_data_max
3097 | `-- zfs_vdev_async_write_active_max_dirty_percent
3098 `--------- zfs_vdev_async_write_active_min_dirty_percent
3101 Until the amount of dirty data exceeds a minimum percentage of the dirty
3102 data allowed in the pool, the I/O scheduler will limit the number of
3103 concurrent operations to the minimum. As that threshold is crossed, the
3104 number of concurrent operations issued increases linearly to the maximum at
3105 the specified maximum percentage of the dirty data allowed in the pool.
3107 Ideally, the amount of dirty data on a busy pool will stay in the sloped
3108 part of the function between \fBzfs_vdev_async_write_active_min_dirty_percent\fR
3109 and \fBzfs_vdev_async_write_active_max_dirty_percent\fR. If it exceeds the
3110 maximum percentage, this indicates that the rate of incoming data is
3111 greater than the rate that the backend storage can handle. In this case, we
3112 must further throttle incoming writes, as described in the next section.
3114 .SH ZFS TRANSACTION DELAY
3115 We delay transactions when we've determined that the backend storage
3116 isn't able to accommodate the rate of incoming writes.
3118 If there is already a transaction waiting, we delay relative to when
3119 that transaction will finish waiting. This way the calculated delay time
3120 is independent of the number of threads concurrently executing
3123 If we are the only waiter, wait relative to when the transaction
3124 started, rather than the current time. This credits the transaction for
3125 "time already served", e.g. reading indirect blocks.
3127 The minimum time for a transaction to take is calculated as:
3129 min_time = zfs_delay_scale * (dirty - min) / (max - dirty)
3130 min_time is then capped at 100 milliseconds.
3133 The delay has two degrees of freedom that can be adjusted via tunables. The
3134 percentage of dirty data at which we start to delay is defined by
3135 \fBzfs_delay_min_dirty_percent\fR. This should typically be at or above
3136 \fBzfs_vdev_async_write_active_max_dirty_percent\fR so that we only start to
3137 delay after writing at full speed has failed to keep up with the incoming write
3138 rate. The scale of the curve is defined by \fBzfs_delay_scale\fR. Roughly speaking,
3139 this variable determines the amount of delay at the midpoint of the curve.
3143 10ms +-------------------------------------------------------------*+
3159 2ms + (midpoint) * +
3162 | zfs_delay_scale ----------> ******** |
3163 0 +-------------------------------------*********----------------+
3164 0% <- zfs_dirty_data_max -> 100%
3167 Note that since the delay is added to the outstanding time remaining on the
3168 most recent transaction, the delay is effectively the inverse of IOPS.
3169 Here the midpoint of 500us translates to 2000 IOPS. The shape of the curve
3170 was chosen such that small changes in the amount of accumulated dirty data
3171 in the first 3/4 of the curve yield relatively small differences in the
3174 The effects can be easier to understand when the amount of delay is
3175 represented on a log scale:
3179 100ms +-------------------------------------------------------------++
3188 + zfs_delay_scale ----------> ***** +
3199 +--------------------------------------------------------------+
3200 0% <- zfs_dirty_data_max -> 100%
3203 Note here that only as the amount of dirty data approaches its limit does
3204 the delay start to increase rapidly. The goal of a properly tuned system
3205 should be to keep the amount of dirty data out of that range by first
3206 ensuring that the appropriate limits are set for the I/O scheduler to reach
3207 optimal throughput on the backend storage, and then by changing the value
3208 of \fBzfs_delay_scale\fR to increase the steepness of the curve.
projects / mirror_zfs.git / blob