]>
Commit | Line | Data |
---|---|---|
a08ee875 LG |
1 | '\" te |
2 | .\" Copyright (c) 2013 by Turbo Fredriksson <turbo@bayour.com>. All rights reserved. | |
3 | .\" The contents of this file are subject to the terms of the Common Development | |
4 | .\" and Distribution License (the "License"). You may not use this file except | |
5 | .\" in compliance with the License. You can obtain a copy of the license at | |
6 | .\" usr/src/OPENSOLARIS.LICENSE or http://www.opensolaris.org/os/licensing. | |
7 | .\" | |
8 | .\" See the License for the specific language governing permissions and | |
9 | .\" limitations under the License. When distributing Covered Code, include this | |
10 | .\" CDDL HEADER in each file and include the License file at | |
11 | .\" usr/src/OPENSOLARIS.LICENSE. If applicable, add the following below this | |
12 | .\" CDDL HEADER, with the fields enclosed by brackets "[]" replaced with your | |
13 | .\" own identifying information: | |
14 | .\" Portions Copyright [yyyy] [name of copyright owner] | |
41d74433 | 15 | .TH ZFS-MODULE-PARAMETERS 5 "Oct 28, 2017" |
a08ee875 LG |
16 | .SH NAME |
17 | zfs\-module\-parameters \- ZFS module parameters | |
18 | .SH DESCRIPTION | |
19 | .sp | |
20 | .LP | |
21 | Description of the different parameters to the ZFS module. | |
22 | ||
23 | .SS "Module parameters" | |
24 | .sp | |
25 | .LP | |
26 | ||
87dac73d AX |
27 | .sp |
28 | .ne 2 | |
29 | .na | |
30 | \fBignore_hole_birth\fR (int) | |
31 | .ad | |
32 | .RS 12n | |
33 | When set, the hole_birth optimization will not be used, and all holes will | |
34 | always be sent on zfs send. Useful if you suspect your datasets are affected | |
35 | by a bug in hole_birth. | |
36 | .sp | |
cae5b340 | 37 | Use \fB1\fR for on (default) and \fB0\fR for off. |
87dac73d AX |
38 | .RE |
39 | ||
a08ee875 LG |
40 | .sp |
41 | .ne 2 | |
42 | .na | |
43 | \fBl2arc_feed_again\fR (int) | |
44 | .ad | |
45 | .RS 12n | |
cae5b340 AX |
46 | Turbo L2ARC warm-up. When the L2ARC is cold the fill interval will be set as |
47 | fast as possible. | |
a08ee875 LG |
48 | .sp |
49 | Use \fB1\fR for yes (default) and \fB0\fR to disable. | |
50 | .RE | |
51 | ||
52 | .sp | |
53 | .ne 2 | |
54 | .na | |
55 | \fBl2arc_feed_min_ms\fR (ulong) | |
56 | .ad | |
57 | .RS 12n | |
cae5b340 AX |
58 | Min feed interval in milliseconds. Requires \fBl2arc_feed_again=1\fR and only |
59 | applicable in related situations. | |
a08ee875 LG |
60 | .sp |
61 | Default value: \fB200\fR. | |
62 | .RE | |
63 | ||
64 | .sp | |
65 | .ne 2 | |
66 | .na | |
67 | \fBl2arc_feed_secs\fR (ulong) | |
68 | .ad | |
69 | .RS 12n | |
70 | Seconds between L2ARC writing | |
71 | .sp | |
72 | Default value: \fB1\fR. | |
73 | .RE | |
74 | ||
75 | .sp | |
76 | .ne 2 | |
77 | .na | |
78 | \fBl2arc_headroom\fR (ulong) | |
79 | .ad | |
80 | .RS 12n | |
cae5b340 AX |
81 | How far through the ARC lists to search for L2ARC cacheable content, expressed |
82 | as a multiplier of \fBl2arc_write_max\fR | |
a08ee875 LG |
83 | .sp |
84 | Default value: \fB2\fR. | |
85 | .RE | |
86 | ||
87 | .sp | |
88 | .ne 2 | |
89 | .na | |
90 | \fBl2arc_headroom_boost\fR (ulong) | |
91 | .ad | |
92 | .RS 12n | |
cae5b340 AX |
93 | Scales \fBl2arc_headroom\fR by this percentage when L2ARC contents are being |
94 | successfully compressed before writing. A value of 100 disables this feature. | |
a08ee875 LG |
95 | .sp |
96 | Default value: \fB200\fR. | |
97 | .RE | |
98 | ||
a08ee875 LG |
99 | .sp |
100 | .ne 2 | |
101 | .na | |
102 | \fBl2arc_noprefetch\fR (int) | |
103 | .ad | |
104 | .RS 12n | |
cae5b340 AX |
105 | Do not write buffers to L2ARC if they were prefetched but not used by |
106 | applications | |
a08ee875 LG |
107 | .sp |
108 | Use \fB1\fR for yes (default) and \fB0\fR to disable. | |
109 | .RE | |
110 | ||
111 | .sp | |
112 | .ne 2 | |
113 | .na | |
114 | \fBl2arc_norw\fR (int) | |
115 | .ad | |
116 | .RS 12n | |
117 | No reads during writes | |
118 | .sp | |
119 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
120 | .RE | |
121 | ||
122 | .sp | |
123 | .ne 2 | |
124 | .na | |
125 | \fBl2arc_write_boost\fR (ulong) | |
126 | .ad | |
127 | .RS 12n | |
cae5b340 AX |
128 | Cold L2ARC devices will have \fBl2arc_write_max\fR increased by this amount |
129 | while they remain cold. | |
a08ee875 LG |
130 | .sp |
131 | Default value: \fB8,388,608\fR. | |
132 | .RE | |
133 | ||
134 | .sp | |
135 | .ne 2 | |
136 | .na | |
137 | \fBl2arc_write_max\fR (ulong) | |
138 | .ad | |
139 | .RS 12n | |
140 | Max write bytes per interval | |
141 | .sp | |
142 | Default value: \fB8,388,608\fR. | |
143 | .RE | |
144 | ||
e10b0808 AX |
145 | .sp |
146 | .ne 2 | |
147 | .na | |
148 | \fBmetaslab_aliquot\fR (ulong) | |
149 | .ad | |
150 | .RS 12n | |
151 | Metaslab granularity, in bytes. This is roughly similar to what would be | |
152 | referred to as the "stripe size" in traditional RAID arrays. In normal | |
153 | operation, ZFS will try to write this amount of data to a top-level vdev | |
154 | before moving on to the next one. | |
155 | .sp | |
156 | Default value: \fB524,288\fR. | |
157 | .RE | |
158 | ||
a08ee875 LG |
159 | .sp |
160 | .ne 2 | |
161 | .na | |
ea04106b | 162 | \fBmetaslab_bias_enabled\fR (int) |
a08ee875 LG |
163 | .ad |
164 | .RS 12n | |
ea04106b AX |
165 | Enable metaslab group biasing based on its vdev's over- or under-utilization |
166 | relative to the pool. | |
167 | .sp | |
168 | Use \fB1\fR for yes (default) and \fB0\fR for no. | |
169 | .RE | |
170 | ||
cae5b340 AX |
171 | .sp |
172 | .ne 2 | |
173 | .na | |
174 | \fBzfs_metaslab_segment_weight_enabled\fR (int) | |
175 | .ad | |
176 | .RS 12n | |
177 | Enable/disable segment-based metaslab selection. | |
178 | .sp | |
179 | Use \fB1\fR for yes (default) and \fB0\fR for no. | |
180 | .RE | |
181 | ||
182 | .sp | |
183 | .ne 2 | |
184 | .na | |
185 | \fBzfs_metaslab_switch_threshold\fR (int) | |
186 | .ad | |
187 | .RS 12n | |
188 | When using segment-based metaslab selection, continue allocating | |
189 | from the active metaslab until \fBzfs_metaslab_switch_threshold\fR | |
190 | worth of buckets have been exhausted. | |
191 | .sp | |
192 | Default value: \fB2\fR. | |
193 | .RE | |
194 | ||
ea04106b AX |
195 | .sp |
196 | .ne 2 | |
197 | .na | |
198 | \fBmetaslab_debug_load\fR (int) | |
199 | .ad | |
200 | .RS 12n | |
201 | Load all metaslabs during pool import. | |
202 | .sp | |
203 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
204 | .RE | |
205 | ||
206 | .sp | |
207 | .ne 2 | |
208 | .na | |
209 | \fBmetaslab_debug_unload\fR (int) | |
210 | .ad | |
211 | .RS 12n | |
212 | Prevent metaslabs from being unloaded. | |
a08ee875 LG |
213 | .sp |
214 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
215 | .RE | |
216 | ||
ea04106b AX |
217 | .sp |
218 | .ne 2 | |
219 | .na | |
220 | \fBmetaslab_fragmentation_factor_enabled\fR (int) | |
221 | .ad | |
222 | .RS 12n | |
223 | Enable use of the fragmentation metric in computing metaslab weights. | |
224 | .sp | |
225 | Use \fB1\fR for yes (default) and \fB0\fR for no. | |
226 | .RE | |
227 | ||
228 | .sp | |
229 | .ne 2 | |
230 | .na | |
231 | \fBmetaslabs_per_vdev\fR (int) | |
232 | .ad | |
233 | .RS 12n | |
234 | When a vdev is added, it will be divided into approximately (but no more than) this number of metaslabs. | |
235 | .sp | |
236 | Default value: \fB200\fR. | |
237 | .RE | |
238 | ||
239 | .sp | |
240 | .ne 2 | |
241 | .na | |
242 | \fBmetaslab_preload_enabled\fR (int) | |
243 | .ad | |
244 | .RS 12n | |
245 | Enable metaslab group preloading. | |
246 | .sp | |
247 | Use \fB1\fR for yes (default) and \fB0\fR for no. | |
248 | .RE | |
249 | ||
250 | .sp | |
251 | .ne 2 | |
252 | .na | |
253 | \fBmetaslab_lba_weighting_enabled\fR (int) | |
254 | .ad | |
255 | .RS 12n | |
256 | Give more weight to metaslabs with lower LBAs, assuming they have | |
257 | greater bandwidth as is typically the case on a modern constant | |
258 | angular velocity disk drive. | |
259 | .sp | |
260 | Use \fB1\fR for yes (default) and \fB0\fR for no. | |
261 | .RE | |
262 | ||
a08ee875 LG |
263 | .sp |
264 | .ne 2 | |
265 | .na | |
266 | \fBspa_config_path\fR (charp) | |
267 | .ad | |
268 | .RS 12n | |
269 | SPA config file | |
270 | .sp | |
271 | Default value: \fB/etc/zfs/zpool.cache\fR. | |
272 | .RE | |
273 | ||
274 | .sp | |
275 | .ne 2 | |
276 | .na | |
277 | \fBspa_asize_inflation\fR (int) | |
278 | .ad | |
279 | .RS 12n | |
280 | Multiplication factor used to estimate actual disk consumption from the | |
281 | size of data being written. The default value is a worst case estimate, | |
282 | but lower values may be valid for a given pool depending on its | |
283 | configuration. Pool administrators who understand the factors involved | |
284 | may wish to specify a more realistic inflation factor, particularly if | |
285 | they operate close to quota or capacity limits. | |
286 | .sp | |
cae5b340 | 287 | Default value: \fB24\fR. |
a08ee875 LG |
288 | .RE |
289 | ||
ea04106b AX |
290 | .sp |
291 | .ne 2 | |
292 | .na | |
293 | \fBspa_load_verify_data\fR (int) | |
294 | .ad | |
295 | .RS 12n | |
296 | Whether to traverse data blocks during an "extreme rewind" (\fB-X\fR) | |
297 | import. Use 0 to disable and 1 to enable. | |
298 | ||
299 | An extreme rewind import normally performs a full traversal of all | |
300 | blocks in the pool for verification. If this parameter is set to 0, | |
301 | the traversal skips non-metadata blocks. It can be toggled once the | |
302 | import has started to stop or start the traversal of non-metadata blocks. | |
303 | .sp | |
cae5b340 | 304 | Default value: \fB1\fR. |
ea04106b AX |
305 | .RE |
306 | ||
307 | .sp | |
308 | .ne 2 | |
309 | .na | |
310 | \fBspa_load_verify_metadata\fR (int) | |
311 | .ad | |
312 | .RS 12n | |
313 | Whether to traverse blocks during an "extreme rewind" (\fB-X\fR) | |
314 | pool import. Use 0 to disable and 1 to enable. | |
315 | ||
316 | An extreme rewind import normally performs a full traversal of all | |
cae5b340 | 317 | blocks in the pool for verification. If this parameter is set to 0, |
ea04106b AX |
318 | the traversal is not performed. It can be toggled once the import has |
319 | started to stop or start the traversal. | |
320 | .sp | |
cae5b340 | 321 | Default value: \fB1\fR. |
ea04106b AX |
322 | .RE |
323 | ||
324 | .sp | |
325 | .ne 2 | |
326 | .na | |
327 | \fBspa_load_verify_maxinflight\fR (int) | |
328 | .ad | |
329 | .RS 12n | |
330 | Maximum concurrent I/Os during the traversal performed during an "extreme | |
331 | rewind" (\fB-X\fR) pool import. | |
332 | .sp | |
cae5b340 | 333 | Default value: \fB10000\fR. |
ea04106b AX |
334 | .RE |
335 | ||
e10b0808 AX |
336 | .sp |
337 | .ne 2 | |
338 | .na | |
339 | \fBspa_slop_shift\fR (int) | |
340 | .ad | |
341 | .RS 12n | |
342 | Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space | |
343 | in the pool to be consumed. This ensures that we don't run the pool | |
344 | completely out of space, due to unaccounted changes (e.g. to the MOS). | |
345 | It also limits the worst-case time to allocate space. If we have | |
346 | less than this amount of free space, most ZPL operations (e.g. write, | |
347 | create) will return ENOSPC. | |
348 | .sp | |
cae5b340 | 349 | Default value: \fB5\fR. |
e10b0808 AX |
350 | .RE |
351 | ||
a08ee875 LG |
352 | .sp |
353 | .ne 2 | |
354 | .na | |
355 | \fBzfetch_array_rd_sz\fR (ulong) | |
356 | .ad | |
357 | .RS 12n | |
ea04106b | 358 | If prefetching is enabled, disable prefetching for reads larger than this size. |
a08ee875 LG |
359 | .sp |
360 | Default value: \fB1,048,576\fR. | |
361 | .RE | |
362 | ||
363 | .sp | |
364 | .ne 2 | |
365 | .na | |
cae5b340 | 366 | \fBzfetch_max_distance\fR (uint) |
a08ee875 LG |
367 | .ad |
368 | .RS 12n | |
cae5b340 | 369 | Max bytes to prefetch per stream (default 8MB). |
a08ee875 | 370 | .sp |
cae5b340 | 371 | Default value: \fB8,388,608\fR. |
a08ee875 LG |
372 | .RE |
373 | ||
374 | .sp | |
375 | .ne 2 | |
376 | .na | |
377 | \fBzfetch_max_streams\fR (uint) | |
378 | .ad | |
379 | .RS 12n | |
ea04106b | 380 | Max number of streams per zfetch (prefetch streams per file). |
a08ee875 LG |
381 | .sp |
382 | Default value: \fB8\fR. | |
383 | .RE | |
384 | ||
385 | .sp | |
386 | .ne 2 | |
387 | .na | |
388 | \fBzfetch_min_sec_reap\fR (uint) | |
389 | .ad | |
390 | .RS 12n | |
ea04106b | 391 | Min time before an active prefetch stream can be reclaimed |
a08ee875 LG |
392 | .sp |
393 | Default value: \fB2\fR. | |
394 | .RE | |
395 | ||
cae5b340 AX |
396 | .sp |
397 | .ne 2 | |
398 | .na | |
399 | \fBzfs_arc_dnode_limit\fR (ulong) | |
400 | .ad | |
401 | .RS 12n | |
402 | When the number of bytes consumed by dnodes in the ARC exceeds this number of | |
403 | bytes, try to unpin some of it in response to demand for non-metadata. This | |
404 | value acts as a ceiling to the amount of dnode metadata, and defaults to 0 which | |
405 | indicates that a percent which is based on \fBzfs_arc_dnode_limit_percent\fR of | |
406 | the ARC meta buffers that may be used for dnodes. | |
407 | ||
408 | See also \fBzfs_arc_meta_prune\fR which serves a similar purpose but is used | |
409 | when the amount of metadata in the ARC exceeds \fBzfs_arc_meta_limit\fR rather | |
410 | than in response to overall demand for non-metadata. | |
411 | ||
412 | .sp | |
413 | Default value: \fB0\fR. | |
414 | .RE | |
415 | ||
416 | .sp | |
417 | .ne 2 | |
418 | .na | |
419 | \fBzfs_arc_dnode_limit_percent\fR (ulong) | |
420 | .ad | |
421 | .RS 12n | |
422 | Percentage that can be consumed by dnodes of ARC meta buffers. | |
423 | .sp | |
424 | See also \fBzfs_arc_dnode_limit\fR which serves a similar purpose but has a | |
425 | higher priority if set to nonzero value. | |
426 | .sp | |
427 | Default value: \fB10\fR. | |
428 | .RE | |
429 | ||
430 | .sp | |
431 | .ne 2 | |
432 | .na | |
433 | \fBzfs_arc_dnode_reduce_percent\fR (ulong) | |
434 | .ad | |
435 | .RS 12n | |
436 | Percentage of ARC dnodes to try to scan in response to demand for non-metadata | |
437 | when the number of bytes consumed by dnodes exceeds \fBzfs_arc_dnode_limit\fR. | |
438 | ||
439 | .sp | |
440 | Default value: \fB10% of the number of dnodes in the ARC\fR. | |
441 | .RE | |
442 | ||
ea04106b AX |
443 | .sp |
444 | .ne 2 | |
445 | .na | |
446 | \fBzfs_arc_average_blocksize\fR (int) | |
447 | .ad | |
448 | .RS 12n | |
449 | The ARC's buffer hash table is sized based on the assumption of an average | |
450 | block size of \fBzfs_arc_average_blocksize\fR (default 8K). This works out | |
451 | to roughly 1MB of hash table per 1GB of physical memory with 8-byte pointers. | |
452 | For configurations with a known larger average block size this value can be | |
453 | increased to reduce the memory footprint. | |
454 | ||
455 | .sp | |
456 | Default value: \fB8192\fR. | |
457 | .RE | |
458 | ||
e10b0808 AX |
459 | .sp |
460 | .ne 2 | |
461 | .na | |
462 | \fBzfs_arc_evict_batch_limit\fR (int) | |
463 | .ad | |
464 | .RS 12n | |
465 | Number ARC headers to evict per sub-list before proceeding to another sub-list. | |
466 | This batch-style operation prevents entire sub-lists from being evicted at once | |
467 | but comes at a cost of additional unlocking and locking. | |
468 | .sp | |
469 | Default value: \fB10\fR. | |
470 | .RE | |
471 | ||
a08ee875 LG |
472 | .sp |
473 | .ne 2 | |
474 | .na | |
475 | \fBzfs_arc_grow_retry\fR (int) | |
476 | .ad | |
477 | .RS 12n | |
41d74433 AX |
478 | If set to a non zero value, it will replace the arc_grow_retry value with this value. |
479 | The arc_grow_retry value (default 5) is the number of seconds the ARC will wait before | |
480 | trying to resume growth after a memory pressure event. | |
a08ee875 | 481 | .sp |
41d74433 | 482 | Default value: \fB0\fR. |
a08ee875 LG |
483 | .RE |
484 | ||
485 | .sp | |
486 | .ne 2 | |
487 | .na | |
e10b0808 | 488 | \fBzfs_arc_lotsfree_percent\fR (int) |
a08ee875 LG |
489 | .ad |
490 | .RS 12n | |
e10b0808 AX |
491 | Throttle I/O when free system memory drops below this percentage of total |
492 | system memory. Setting this value to 0 will disable the throttle. | |
a08ee875 | 493 | .sp |
e10b0808 | 494 | Default value: \fB10\fR. |
a08ee875 LG |
495 | .RE |
496 | ||
497 | .sp | |
498 | .ne 2 | |
499 | .na | |
e10b0808 | 500 | \fBzfs_arc_max\fR (ulong) |
a08ee875 LG |
501 | .ad |
502 | .RS 12n | |
cae5b340 AX |
503 | Max arc size of ARC in bytes. If set to 0 then it will consume 1/2 of system |
504 | RAM. This value must be at least 67108864 (64 megabytes). | |
505 | .sp | |
506 | This value can be changed dynamically with some caveats. It cannot be set back | |
507 | to 0 while running and reducing it below the current ARC size will not cause | |
508 | the ARC to shrink without memory pressure to induce shrinking. | |
a08ee875 | 509 | .sp |
e10b0808 | 510 | Default value: \fB0\fR. |
a08ee875 LG |
511 | .RE |
512 | ||
41d74433 AX |
513 | .sp |
514 | .ne 2 | |
515 | .na | |
516 | \fBzfs_arc_meta_adjust_restarts\fR (ulong) | |
517 | .ad | |
518 | .RS 12n | |
519 | The number of restart passes to make while scanning the ARC attempting | |
520 | the free buffers in order to stay below the \fBzfs_arc_meta_limit\fR. | |
521 | This value should not need to be tuned but is available to facilitate | |
522 | performance analysis. | |
523 | .sp | |
524 | Default value: \fB4096\fR. | |
525 | .RE | |
526 | ||
a08ee875 LG |
527 | .sp |
528 | .ne 2 | |
529 | .na | |
530 | \fBzfs_arc_meta_limit\fR (ulong) | |
531 | .ad | |
532 | .RS 12n | |
ea04106b AX |
533 | The maximum allowed size in bytes that meta data buffers are allowed to |
534 | consume in the ARC. When this limit is reached meta data buffers will | |
535 | be reclaimed even if the overall arc_c_max has not been reached. This | |
cae5b340 AX |
536 | value defaults to 0 which indicates that a percent which is based on |
537 | \fBzfs_arc_meta_limit_percent\fR of the ARC may be used for meta data. | |
538 | .sp | |
539 | This value my be changed dynamically except that it cannot be set back to 0 | |
540 | for a specific percent of the ARC; it must be set to an explicit value. | |
a08ee875 LG |
541 | .sp |
542 | Default value: \fB0\fR. | |
543 | .RE | |
544 | ||
cae5b340 AX |
545 | .sp |
546 | .ne 2 | |
547 | .na | |
548 | \fBzfs_arc_meta_limit_percent\fR (ulong) | |
549 | .ad | |
550 | .RS 12n | |
551 | Percentage of ARC buffers that can be used for meta data. | |
552 | ||
553 | See also \fBzfs_arc_meta_limit\fR which serves a similar purpose but has a | |
554 | higher priority if set to nonzero value. | |
555 | ||
556 | .sp | |
557 | Default value: \fB75\fR. | |
558 | .RE | |
559 | ||
e10b0808 AX |
560 | .sp |
561 | .ne 2 | |
562 | .na | |
563 | \fBzfs_arc_meta_min\fR (ulong) | |
564 | .ad | |
565 | .RS 12n | |
566 | The minimum allowed size in bytes that meta data buffers may consume in | |
567 | the ARC. This value defaults to 0 which disables a floor on the amount | |
568 | of the ARC devoted meta data. | |
569 | .sp | |
570 | Default value: \fB0\fR. | |
571 | .RE | |
572 | ||
a08ee875 LG |
573 | .sp |
574 | .ne 2 | |
575 | .na | |
576 | \fBzfs_arc_meta_prune\fR (int) | |
577 | .ad | |
578 | .RS 12n | |
ea04106b AX |
579 | The number of dentries and inodes to be scanned looking for entries |
580 | which can be dropped. This may be required when the ARC reaches the | |
581 | \fBzfs_arc_meta_limit\fR because dentries and inodes can pin buffers | |
582 | in the ARC. Increasing this value will cause to dentry and inode caches | |
583 | to be pruned more aggressively. Setting this value to 0 will disable | |
584 | pruning the inode and dentry caches. | |
a08ee875 | 585 | .sp |
ea04106b AX |
586 | Default value: \fB10,000\fR. |
587 | .RE | |
588 | ||
589 | .sp | |
590 | .ne 2 | |
591 | .na | |
41d74433 | 592 | \fBzfs_arc_meta_strategy\fR (int) |
ea04106b AX |
593 | .ad |
594 | .RS 12n | |
41d74433 AX |
595 | Define the strategy for ARC meta data buffer eviction (meta reclaim strategy). |
596 | A value of 0 (META_ONLY) will evict only the ARC meta data buffers. | |
597 | A value of 1 (BALANCED) indicates that additional data buffers may be evicted if | |
598 | that is required to in order to evict the required number of meta data buffers. | |
ea04106b | 599 | .sp |
41d74433 | 600 | Default value: \fB1\fR. |
a08ee875 LG |
601 | .RE |
602 | ||
603 | .sp | |
604 | .ne 2 | |
605 | .na | |
606 | \fBzfs_arc_min\fR (ulong) | |
607 | .ad | |
608 | .RS 12n | |
41d74433 AX |
609 | Min arc size of ARC in bytes. If set to 0 then arc_c_min will default to |
610 | consuming the larger of 32M or 1/32 of total system memory. | |
a08ee875 | 611 | .sp |
41d74433 | 612 | Default value: \fB0\fR. |
a08ee875 LG |
613 | .RE |
614 | ||
615 | .sp | |
616 | .ne 2 | |
617 | .na | |
618 | \fBzfs_arc_min_prefetch_lifespan\fR (int) | |
619 | .ad | |
620 | .RS 12n | |
cae5b340 AX |
621 | Minimum time prefetched blocks are locked in the ARC, specified in jiffies. |
622 | A value of 0 will default to 1 second. | |
a08ee875 | 623 | .sp |
cae5b340 | 624 | Default value: \fB0\fR. |
a08ee875 LG |
625 | .RE |
626 | ||
e10b0808 AX |
627 | .sp |
628 | .ne 2 | |
629 | .na | |
cae5b340 | 630 | \fBzfs_multilist_num_sublists\fR (int) |
e10b0808 AX |
631 | .ad |
632 | .RS 12n | |
633 | To allow more fine-grained locking, each ARC state contains a series | |
634 | of lists for both data and meta data objects. Locking is performed at | |
635 | the level of these "sub-lists". This parameters controls the number of | |
cae5b340 AX |
636 | sub-lists per ARC state, and also applies to other uses of the |
637 | multilist data structure. | |
e10b0808 | 638 | .sp |
cae5b340 | 639 | Default value: \fB4\fR or the number of online CPUs, whichever is greater |
e10b0808 AX |
640 | .RE |
641 | ||
642 | .sp | |
643 | .ne 2 | |
644 | .na | |
645 | \fBzfs_arc_overflow_shift\fR (int) | |
646 | .ad | |
647 | .RS 12n | |
648 | The ARC size is considered to be overflowing if it exceeds the current | |
649 | ARC target size (arc_c) by a threshold determined by this parameter. | |
650 | The threshold is calculated as a fraction of arc_c using the formula | |
651 | "arc_c >> \fBzfs_arc_overflow_shift\fR". | |
652 | ||
653 | The default value of 8 causes the ARC to be considered to be overflowing | |
654 | if it exceeds the target size by 1/256th (0.3%) of the target size. | |
655 | ||
656 | When the ARC is overflowing, new buffer allocations are stalled until | |
657 | the reclaim thread catches up and the overflow condition no longer exists. | |
658 | .sp | |
659 | Default value: \fB8\fR. | |
660 | .RE | |
661 | ||
662 | .sp | |
663 | .ne 2 | |
664 | .na | |
665 | ||
666 | \fBzfs_arc_p_min_shift\fR (int) | |
667 | .ad | |
668 | .RS 12n | |
41d74433 AX |
669 | If set to a non zero value, this will update arc_p_min_shift (default 4) |
670 | with the new value. | |
671 | arc_p_min_shift is used to shift of arc_c for calculating both min and max | |
672 | max arc_p | |
e10b0808 | 673 | .sp |
41d74433 | 674 | Default value: \fB0\fR. |
e10b0808 AX |
675 | .RE |
676 | ||
ea04106b AX |
677 | .sp |
678 | .ne 2 | |
679 | .na | |
680 | \fBzfs_arc_p_dampener_disable\fR (int) | |
681 | .ad | |
682 | .RS 12n | |
683 | Disable arc_p adapt dampener | |
684 | .sp | |
685 | Use \fB1\fR for yes (default) and \fB0\fR to disable. | |
a08ee875 LG |
686 | .RE |
687 | ||
688 | .sp | |
689 | .ne 2 | |
690 | .na | |
691 | \fBzfs_arc_shrink_shift\fR (int) | |
692 | .ad | |
693 | .RS 12n | |
41d74433 AX |
694 | If set to a non zero value, this will update arc_shrink_shift (default 7) |
695 | with the new value. | |
a08ee875 | 696 | .sp |
41d74433 | 697 | Default value: \fB0\fR. |
a08ee875 LG |
698 | .RE |
699 | ||
cae5b340 AX |
700 | .sp |
701 | .ne 2 | |
702 | .na | |
703 | \fBzfs_arc_pc_percent\fR (uint) | |
704 | .ad | |
705 | .RS 12n | |
706 | Percent of pagecache to reclaim arc to | |
707 | ||
708 | This tunable allows ZFS arc to play more nicely with the kernel's LRU | |
709 | pagecache. It can guarantee that the arc size won't collapse under scanning | |
710 | pressure on the pagecache, yet still allows arc to be reclaimed down to | |
711 | zfs_arc_min if necessary. This value is specified as percent of pagecache | |
712 | size (as measured by NR_FILE_PAGES) where that percent may exceed 100. This | |
713 | only operates during memory pressure/reclaim. | |
714 | .sp | |
715 | Default value: \fB0\fR (disabled). | |
716 | .RE | |
717 | ||
e10b0808 AX |
718 | .sp |
719 | .ne 2 | |
720 | .na | |
721 | \fBzfs_arc_sys_free\fR (ulong) | |
722 | .ad | |
723 | .RS 12n | |
724 | The target number of bytes the ARC should leave as free memory on the system. | |
725 | Defaults to the larger of 1/64 of physical memory or 512K. Setting this | |
726 | option to a non-zero value will override the default. | |
727 | .sp | |
728 | Default value: \fB0\fR. | |
729 | .RE | |
730 | ||
a08ee875 LG |
731 | .sp |
732 | .ne 2 | |
733 | .na | |
734 | \fBzfs_autoimport_disable\fR (int) | |
735 | .ad | |
736 | .RS 12n | |
ea04106b | 737 | Disable pool import at module load by ignoring the cache file (typically \fB/etc/zfs/zpool.cache\fR). |
a08ee875 | 738 | .sp |
e10b0808 AX |
739 | Use \fB1\fR for yes (default) and \fB0\fR for no. |
740 | .RE | |
741 | ||
42f7b73b AX |
742 | .sp |
743 | .ne 2 | |
744 | .na | |
745 | \fBzfs_checksums_per_second\fR (int) | |
746 | .ad | |
747 | .RS 12n | |
748 | Rate limit checksum events to this many per second. Note that this should | |
749 | not be set below the zed thresholds (currently 10 checksums over 10 sec) | |
750 | or else zed may not trigger any action. | |
751 | .sp | |
752 | Default value: 20 | |
753 | .RE | |
754 | ||
755 | .sp | |
756 | .ne 2 | |
757 | .na | |
758 | \fBzfs_commit_timeout_pct\fR (int) | |
759 | .ad | |
760 | .RS 12n | |
761 | This controls the amount of time that a ZIL block (lwb) will remain "open" | |
762 | when it isn't "full", and it has a thread waiting for it to be committed to | |
763 | stable storage. The timeout is scaled based on a percentage of the last lwb | |
764 | latency to avoid significantly impacting the latency of each individual | |
765 | transaction record (itx). | |
766 | .sp | |
767 | Default value: \fB5\fR%. | |
768 | .RE | |
769 | ||
e10b0808 AX |
770 | .sp |
771 | .ne 2 | |
772 | .na | |
773 | \fBzfs_dbgmsg_enable\fR (int) | |
774 | .ad | |
775 | .RS 12n | |
776 | Internally ZFS keeps a small log to facilitate debugging. By default the log | |
777 | is disabled, to enable it set this option to 1. The contents of the log can | |
778 | be accessed by reading the /proc/spl/kstat/zfs/dbgmsg file. Writing 0 to | |
779 | this proc file clears the log. | |
780 | .sp | |
781 | Default value: \fB0\fR. | |
782 | .RE | |
783 | ||
784 | .sp | |
785 | .ne 2 | |
786 | .na | |
787 | \fBzfs_dbgmsg_maxsize\fR (int) | |
788 | .ad | |
789 | .RS 12n | |
790 | The maximum size in bytes of the internal ZFS debug log. | |
791 | .sp | |
792 | Default value: \fB4M\fR. | |
a08ee875 LG |
793 | .RE |
794 | ||
795 | .sp | |
796 | .ne 2 | |
797 | .na | |
798 | \fBzfs_dbuf_state_index\fR (int) | |
799 | .ad | |
800 | .RS 12n | |
cae5b340 AX |
801 | This feature is currently unused. It is normally used for controlling what |
802 | reporting is available under /proc/spl/kstat/zfs. | |
a08ee875 LG |
803 | .sp |
804 | Default value: \fB0\fR. | |
805 | .RE | |
806 | ||
807 | .sp | |
808 | .ne 2 | |
809 | .na | |
810 | \fBzfs_deadman_enabled\fR (int) | |
811 | .ad | |
812 | .RS 12n | |
cae5b340 AX |
813 | When a pool sync operation takes longer than \fBzfs_deadman_synctime_ms\fR |
814 | milliseconds, a "slow spa_sync" message is logged to the debug log | |
815 | (see \fBzfs_dbgmsg_enable\fR). If \fBzfs_deadman_enabled\fR is set, | |
816 | all pending IO operations are also checked and if any haven't completed | |
817 | within \fBzfs_deadman_synctime_ms\fR milliseconds, a "SLOW IO" message | |
818 | is logged to the debug log and a "delay" system event with the details of | |
819 | the hung IO is posted. | |
a08ee875 | 820 | .sp |
cae5b340 AX |
821 | Use \fB1\fR (default) to enable the slow IO check and \fB0\fR to disable. |
822 | .RE | |
823 | ||
824 | .sp | |
825 | .ne 2 | |
826 | .na | |
827 | \fBzfs_deadman_checktime_ms\fR (int) | |
828 | .ad | |
829 | .RS 12n | |
830 | Once a pool sync operation has taken longer than | |
831 | \fBzfs_deadman_synctime_ms\fR milliseconds, continue to check for slow | |
832 | operations every \fBzfs_deadman_checktime_ms\fR milliseconds. | |
833 | .sp | |
834 | Default value: \fB5,000\fR. | |
a08ee875 LG |
835 | .RE |
836 | ||
837 | .sp | |
838 | .ne 2 | |
839 | .na | |
840 | \fBzfs_deadman_synctime_ms\fR (ulong) | |
841 | .ad | |
842 | .RS 12n | |
cae5b340 AX |
843 | Interval in milliseconds after which the deadman is triggered and also |
844 | the interval after which an IO operation is considered to be "hung" | |
845 | if \fBzfs_deadman_enabled\fR is set. | |
846 | ||
847 | See \fBzfs_deadman_enabled\fR. | |
a08ee875 LG |
848 | .sp |
849 | Default value: \fB1,000,000\fR. | |
850 | .RE | |
851 | ||
852 | .sp | |
853 | .ne 2 | |
854 | .na | |
855 | \fBzfs_dedup_prefetch\fR (int) | |
856 | .ad | |
857 | .RS 12n | |
858 | Enable prefetching dedup-ed blks | |
859 | .sp | |
ea04106b | 860 | Use \fB1\fR for yes and \fB0\fR to disable (default). |
a08ee875 LG |
861 | .RE |
862 | ||
863 | .sp | |
864 | .ne 2 | |
865 | .na | |
866 | \fBzfs_delay_min_dirty_percent\fR (int) | |
867 | .ad | |
868 | .RS 12n | |
869 | Start to delay each transaction once there is this amount of dirty data, | |
870 | expressed as a percentage of \fBzfs_dirty_data_max\fR. | |
871 | This value should be >= zfs_vdev_async_write_active_max_dirty_percent. | |
872 | See the section "ZFS TRANSACTION DELAY". | |
873 | .sp | |
874 | Default value: \fB60\fR. | |
875 | .RE | |
876 | ||
877 | .sp | |
878 | .ne 2 | |
879 | .na | |
880 | \fBzfs_delay_scale\fR (int) | |
881 | .ad | |
882 | .RS 12n | |
883 | This controls how quickly the transaction delay approaches infinity. | |
884 | Larger values cause longer delays for a given amount of dirty data. | |
885 | .sp | |
886 | For the smoothest delay, this value should be about 1 billion divided | |
887 | by the maximum number of operations per second. This will smoothly | |
888 | handle between 10x and 1/10th this number. | |
889 | .sp | |
890 | See the section "ZFS TRANSACTION DELAY". | |
891 | .sp | |
892 | Note: \fBzfs_delay_scale\fR * \fBzfs_dirty_data_max\fR must be < 2^64. | |
893 | .sp | |
894 | Default value: \fB500,000\fR. | |
895 | .RE | |
896 | ||
42f7b73b AX |
897 | .sp |
898 | .ne 2 | |
899 | .na | |
900 | \fBzfs_delays_per_second\fR (int) | |
901 | .ad | |
902 | .RS 12n | |
903 | Rate limit IO delay events to this many per second. | |
904 | .sp | |
905 | Default value: 20 | |
906 | .RE | |
907 | ||
cae5b340 AX |
908 | .sp |
909 | .ne 2 | |
910 | .na | |
911 | \fBzfs_delete_blocks\fR (ulong) | |
912 | .ad | |
913 | .RS 12n | |
914 | This is the used to define a large file for the purposes of delete. Files | |
915 | containing more than \fBzfs_delete_blocks\fR will be deleted asynchronously | |
916 | while smaller files are deleted synchronously. Decreasing this value will | |
917 | reduce the time spent in an unlink(2) system call at the expense of a longer | |
918 | delay before the freed space is available. | |
919 | .sp | |
920 | Default value: \fB20,480\fR. | |
921 | .RE | |
922 | ||
a08ee875 LG |
923 | .sp |
924 | .ne 2 | |
925 | .na | |
926 | \fBzfs_dirty_data_max\fR (int) | |
927 | .ad | |
928 | .RS 12n | |
929 | Determines the dirty space limit in bytes. Once this limit is exceeded, new | |
930 | writes are halted until space frees up. This parameter takes precedence | |
931 | over \fBzfs_dirty_data_max_percent\fR. | |
932 | See the section "ZFS TRANSACTION DELAY". | |
933 | .sp | |
934 | Default value: 10 percent of all memory, capped at \fBzfs_dirty_data_max_max\fR. | |
935 | .RE | |
936 | ||
937 | .sp | |
938 | .ne 2 | |
939 | .na | |
940 | \fBzfs_dirty_data_max_max\fR (int) | |
941 | .ad | |
942 | .RS 12n | |
943 | Maximum allowable value of \fBzfs_dirty_data_max\fR, expressed in bytes. | |
944 | This limit is only enforced at module load time, and will be ignored if | |
945 | \fBzfs_dirty_data_max\fR is later changed. This parameter takes | |
946 | precedence over \fBzfs_dirty_data_max_max_percent\fR. See the section | |
947 | "ZFS TRANSACTION DELAY". | |
948 | .sp | |
949 | Default value: 25% of physical RAM. | |
950 | .RE | |
951 | ||
952 | .sp | |
953 | .ne 2 | |
954 | .na | |
955 | \fBzfs_dirty_data_max_max_percent\fR (int) | |
956 | .ad | |
957 | .RS 12n | |
958 | Maximum allowable value of \fBzfs_dirty_data_max\fR, expressed as a | |
959 | percentage of physical RAM. This limit is only enforced at module load | |
960 | time, and will be ignored if \fBzfs_dirty_data_max\fR is later changed. | |
961 | The parameter \fBzfs_dirty_data_max_max\fR takes precedence over this | |
962 | one. See the section "ZFS TRANSACTION DELAY". | |
963 | .sp | |
cae5b340 | 964 | Default value: \fB25\fR. |
a08ee875 LG |
965 | .RE |
966 | ||
967 | .sp | |
968 | .ne 2 | |
969 | .na | |
970 | \fBzfs_dirty_data_max_percent\fR (int) | |
971 | .ad | |
972 | .RS 12n | |
973 | Determines the dirty space limit, expressed as a percentage of all | |
974 | memory. Once this limit is exceeded, new writes are halted until space frees | |
975 | up. The parameter \fBzfs_dirty_data_max\fR takes precedence over this | |
976 | one. See the section "ZFS TRANSACTION DELAY". | |
977 | .sp | |
978 | Default value: 10%, subject to \fBzfs_dirty_data_max_max\fR. | |
979 | .RE | |
980 | ||
981 | .sp | |
982 | .ne 2 | |
983 | .na | |
984 | \fBzfs_dirty_data_sync\fR (int) | |
985 | .ad | |
986 | .RS 12n | |
987 | Start syncing out a transaction group if there is at least this much dirty data. | |
988 | .sp | |
989 | Default value: \fB67,108,864\fR. | |
990 | .RE | |
991 | ||
cae5b340 AX |
992 | .sp |
993 | .ne 2 | |
994 | .na | |
995 | \fBzfs_fletcher_4_impl\fR (string) | |
996 | .ad | |
997 | .RS 12n | |
998 | Select a fletcher 4 implementation. | |
999 | .sp | |
1000 | Supported selectors are: \fBfastest\fR, \fBscalar\fR, \fBsse2\fR, \fBssse3\fR, | |
1001 | \fBavx2\fR, \fBavx512f\fR, and \fBaarch64_neon\fR. | |
1002 | All of the selectors except \fBfastest\fR and \fBscalar\fR require instruction | |
1003 | set extensions to be available and will only appear if ZFS detects that they are | |
1004 | present at runtime. If multiple implementations of fletcher 4 are available, | |
1005 | the \fBfastest\fR will be chosen using a micro benchmark. Selecting \fBscalar\fR | |
1006 | results in the original, CPU based calculation, being used. Selecting any option | |
1007 | other than \fBfastest\fR and \fBscalar\fR results in vector instructions from | |
1008 | the respective CPU instruction set being used. | |
1009 | .sp | |
1010 | Default value: \fBfastest\fR. | |
1011 | .RE | |
1012 | ||
1013 | .sp | |
1014 | .ne 2 | |
1015 | .na | |
1016 | \fBzfs_free_bpobj_enabled\fR (int) | |
1017 | .ad | |
1018 | .RS 12n | |
1019 | Enable/disable the processing of the free_bpobj object. | |
1020 | .sp | |
1021 | Default value: \fB1\fR. | |
1022 | .RE | |
1023 | ||
ea04106b AX |
1024 | .sp |
1025 | .ne 2 | |
1026 | .na | |
1027 | \fBzfs_free_max_blocks\fR (ulong) | |
1028 | .ad | |
1029 | .RS 12n | |
1030 | Maximum number of blocks freed in a single txg. | |
1031 | .sp | |
1032 | Default value: \fB100,000\fR. | |
1033 | .RE | |
1034 | ||
a08ee875 LG |
1035 | .sp |
1036 | .ne 2 | |
1037 | .na | |
1038 | \fBzfs_vdev_async_read_max_active\fR (int) | |
1039 | .ad | |
1040 | .RS 12n | |
cae5b340 | 1041 | Maximum asynchronous read I/Os active to each device. |
a08ee875 LG |
1042 | See the section "ZFS I/O SCHEDULER". |
1043 | .sp | |
1044 | Default value: \fB3\fR. | |
1045 | .RE | |
1046 | ||
1047 | .sp | |
1048 | .ne 2 | |
1049 | .na | |
1050 | \fBzfs_vdev_async_read_min_active\fR (int) | |
1051 | .ad | |
1052 | .RS 12n | |
1053 | Minimum asynchronous read I/Os active to each device. | |
1054 | See the section "ZFS I/O SCHEDULER". | |
1055 | .sp | |
1056 | Default value: \fB1\fR. | |
1057 | .RE | |
1058 | ||
1059 | .sp | |
1060 | .ne 2 | |
1061 | .na | |
1062 | \fBzfs_vdev_async_write_active_max_dirty_percent\fR (int) | |
1063 | .ad | |
1064 | .RS 12n | |
1065 | When the pool has more than | |
1066 | \fBzfs_vdev_async_write_active_max_dirty_percent\fR dirty data, use | |
1067 | \fBzfs_vdev_async_write_max_active\fR to limit active async writes. If | |
1068 | the dirty data is between min and max, the active I/O limit is linearly | |
1069 | interpolated. See the section "ZFS I/O SCHEDULER". | |
1070 | .sp | |
1071 | Default value: \fB60\fR. | |
1072 | .RE | |
1073 | ||
1074 | .sp | |
1075 | .ne 2 | |
1076 | .na | |
1077 | \fBzfs_vdev_async_write_active_min_dirty_percent\fR (int) | |
1078 | .ad | |
1079 | .RS 12n | |
1080 | When the pool has less than | |
1081 | \fBzfs_vdev_async_write_active_min_dirty_percent\fR dirty data, use | |
1082 | \fBzfs_vdev_async_write_min_active\fR to limit active async writes. If | |
1083 | the dirty data is between min and max, the active I/O limit is linearly | |
1084 | interpolated. See the section "ZFS I/O SCHEDULER". | |
1085 | .sp | |
1086 | Default value: \fB30\fR. | |
1087 | .RE | |
1088 | ||
1089 | .sp | |
1090 | .ne 2 | |
1091 | .na | |
1092 | \fBzfs_vdev_async_write_max_active\fR (int) | |
1093 | .ad | |
1094 | .RS 12n | |
cae5b340 | 1095 | Maximum asynchronous write I/Os active to each device. |
a08ee875 LG |
1096 | See the section "ZFS I/O SCHEDULER". |
1097 | .sp | |
1098 | Default value: \fB10\fR. | |
1099 | .RE | |
1100 | ||
1101 | .sp | |
1102 | .ne 2 | |
1103 | .na | |
1104 | \fBzfs_vdev_async_write_min_active\fR (int) | |
1105 | .ad | |
1106 | .RS 12n | |
1107 | Minimum asynchronous write I/Os active to each device. | |
1108 | See the section "ZFS I/O SCHEDULER". | |
1109 | .sp | |
cae5b340 AX |
1110 | Lower values are associated with better latency on rotational media but poorer |
1111 | resilver performance. The default value of 2 was chosen as a compromise. A | |
1112 | value of 3 has been shown to improve resilver performance further at a cost of | |
1113 | further increasing latency. | |
1114 | .sp | |
1115 | Default value: \fB2\fR. | |
a08ee875 LG |
1116 | .RE |
1117 | ||
1118 | .sp | |
1119 | .ne 2 | |
1120 | .na | |
1121 | \fBzfs_vdev_max_active\fR (int) | |
1122 | .ad | |
1123 | .RS 12n | |
1124 | The maximum number of I/Os active to each device. Ideally, this will be >= | |
1125 | the sum of each queue's max_active. It must be at least the sum of each | |
1126 | queue's min_active. See the section "ZFS I/O SCHEDULER". | |
1127 | .sp | |
1128 | Default value: \fB1,000\fR. | |
1129 | .RE | |
1130 | ||
1131 | .sp | |
1132 | .ne 2 | |
1133 | .na | |
1134 | \fBzfs_vdev_scrub_max_active\fR (int) | |
1135 | .ad | |
1136 | .RS 12n | |
cae5b340 | 1137 | Maximum scrub I/Os active to each device. |
a08ee875 LG |
1138 | See the section "ZFS I/O SCHEDULER". |
1139 | .sp | |
1140 | Default value: \fB2\fR. | |
1141 | .RE | |
1142 | ||
1143 | .sp | |
1144 | .ne 2 | |
1145 | .na | |
1146 | \fBzfs_vdev_scrub_min_active\fR (int) | |
1147 | .ad | |
1148 | .RS 12n | |
1149 | Minimum scrub I/Os active to each device. | |
1150 | See the section "ZFS I/O SCHEDULER". | |
1151 | .sp | |
1152 | Default value: \fB1\fR. | |
1153 | .RE | |
1154 | ||
1155 | .sp | |
1156 | .ne 2 | |
1157 | .na | |
1158 | \fBzfs_vdev_sync_read_max_active\fR (int) | |
1159 | .ad | |
1160 | .RS 12n | |
cae5b340 | 1161 | Maximum synchronous read I/Os active to each device. |
a08ee875 LG |
1162 | See the section "ZFS I/O SCHEDULER". |
1163 | .sp | |
1164 | Default value: \fB10\fR. | |
1165 | .RE | |
1166 | ||
1167 | .sp | |
1168 | .ne 2 | |
1169 | .na | |
1170 | \fBzfs_vdev_sync_read_min_active\fR (int) | |
1171 | .ad | |
1172 | .RS 12n | |
1173 | Minimum synchronous read I/Os active to each device. | |
1174 | See the section "ZFS I/O SCHEDULER". | |
1175 | .sp | |
1176 | Default value: \fB10\fR. | |
1177 | .RE | |
1178 | ||
1179 | .sp | |
1180 | .ne 2 | |
1181 | .na | |
1182 | \fBzfs_vdev_sync_write_max_active\fR (int) | |
1183 | .ad | |
1184 | .RS 12n | |
cae5b340 | 1185 | Maximum synchronous write I/Os active to each device. |
a08ee875 LG |
1186 | See the section "ZFS I/O SCHEDULER". |
1187 | .sp | |
1188 | Default value: \fB10\fR. | |
1189 | .RE | |
1190 | ||
1191 | .sp | |
1192 | .ne 2 | |
1193 | .na | |
1194 | \fBzfs_vdev_sync_write_min_active\fR (int) | |
1195 | .ad | |
1196 | .RS 12n | |
1197 | Minimum synchronous write I/Os active to each device. | |
1198 | See the section "ZFS I/O SCHEDULER". | |
1199 | .sp | |
1200 | Default value: \fB10\fR. | |
1201 | .RE | |
1202 | ||
cae5b340 AX |
1203 | .sp |
1204 | .ne 2 | |
1205 | .na | |
1206 | \fBzfs_vdev_queue_depth_pct\fR (int) | |
1207 | .ad | |
1208 | .RS 12n | |
1209 | Maximum number of queued allocations per top-level vdev expressed as | |
1210 | a percentage of \fBzfs_vdev_async_write_max_active\fR which allows the | |
1211 | system to detect devices that are more capable of handling allocations | |
1212 | and to allocate more blocks to those devices. It allows for dynamic | |
1213 | allocation distribution when devices are imbalanced as fuller devices | |
1214 | will tend to be slower than empty devices. | |
1215 | ||
1216 | See also \fBzio_dva_throttle_enabled\fR. | |
1217 | .sp | |
1218 | Default value: \fB1000\fR. | |
1219 | .RE | |
1220 | ||
a08ee875 LG |
1221 | .sp |
1222 | .ne 2 | |
1223 | .na | |
1224 | \fBzfs_disable_dup_eviction\fR (int) | |
1225 | .ad | |
1226 | .RS 12n | |
1227 | Disable duplicate buffer eviction | |
1228 | .sp | |
1229 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
1230 | .RE | |
1231 | ||
1232 | .sp | |
1233 | .ne 2 | |
1234 | .na | |
1235 | \fBzfs_expire_snapshot\fR (int) | |
1236 | .ad | |
1237 | .RS 12n | |
1238 | Seconds to expire .zfs/snapshot | |
1239 | .sp | |
1240 | Default value: \fB300\fR. | |
1241 | .RE | |
1242 | ||
e10b0808 AX |
1243 | .sp |
1244 | .ne 2 | |
1245 | .na | |
1246 | \fBzfs_admin_snapshot\fR (int) | |
1247 | .ad | |
1248 | .RS 12n | |
1249 | Allow the creation, removal, or renaming of entries in the .zfs/snapshot | |
1250 | directory to cause the creation, destruction, or renaming of snapshots. | |
1251 | When enabled this functionality works both locally and over NFS exports | |
1252 | which have the 'no_root_squash' option set. This functionality is disabled | |
1253 | by default. | |
1254 | .sp | |
1255 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
1256 | .RE | |
1257 | ||
a08ee875 LG |
1258 | .sp |
1259 | .ne 2 | |
1260 | .na | |
1261 | \fBzfs_flags\fR (int) | |
1262 | .ad | |
1263 | .RS 12n | |
ea04106b AX |
1264 | Set additional debugging flags. The following flags may be bitwise-or'd |
1265 | together. | |
1266 | .sp | |
1267 | .TS | |
1268 | box; | |
1269 | rB lB | |
1270 | lB lB | |
1271 | r l. | |
1272 | Value Symbolic Name | |
1273 | Description | |
1274 | _ | |
1275 | 1 ZFS_DEBUG_DPRINTF | |
1276 | Enable dprintf entries in the debug log. | |
1277 | _ | |
1278 | 2 ZFS_DEBUG_DBUF_VERIFY * | |
1279 | Enable extra dbuf verifications. | |
1280 | _ | |
1281 | 4 ZFS_DEBUG_DNODE_VERIFY * | |
1282 | Enable extra dnode verifications. | |
1283 | _ | |
1284 | 8 ZFS_DEBUG_SNAPNAMES | |
1285 | Enable snapshot name verification. | |
1286 | _ | |
1287 | 16 ZFS_DEBUG_MODIFY | |
1288 | Check for illegally modified ARC buffers. | |
1289 | _ | |
1290 | 32 ZFS_DEBUG_SPA | |
1291 | Enable spa_dbgmsg entries in the debug log. | |
1292 | _ | |
1293 | 64 ZFS_DEBUG_ZIO_FREE | |
1294 | Enable verification of block frees. | |
1295 | _ | |
1296 | 128 ZFS_DEBUG_HISTOGRAM_VERIFY | |
1297 | Enable extra spacemap histogram verifications. | |
cae5b340 AX |
1298 | _ |
1299 | 256 ZFS_DEBUG_METASLAB_VERIFY | |
1300 | Verify space accounting on disk matches in-core range_trees. | |
1301 | _ | |
1302 | 512 ZFS_DEBUG_SET_ERROR | |
1303 | Enable SET_ERROR and dprintf entries in the debug log. | |
ea04106b AX |
1304 | .TE |
1305 | .sp | |
1306 | * Requires debug build. | |
a08ee875 | 1307 | .sp |
ea04106b AX |
1308 | Default value: \fB0\fR. |
1309 | .RE | |
1310 | ||
1311 | .sp | |
1312 | .ne 2 | |
1313 | .na | |
1314 | \fBzfs_free_leak_on_eio\fR (int) | |
1315 | .ad | |
1316 | .RS 12n | |
1317 | If destroy encounters an EIO while reading metadata (e.g. indirect | |
1318 | blocks), space referenced by the missing metadata can not be freed. | |
1319 | Normally this causes the background destroy to become "stalled", as | |
1320 | it is unable to make forward progress. While in this stalled state, | |
1321 | all remaining space to free from the error-encountering filesystem is | |
1322 | "temporarily leaked". Set this flag to cause it to ignore the EIO, | |
1323 | permanently leak the space from indirect blocks that can not be read, | |
1324 | and continue to free everything else that it can. | |
1325 | ||
1326 | The default, "stalling" behavior is useful if the storage partially | |
1327 | fails (i.e. some but not all i/os fail), and then later recovers. In | |
1328 | this case, we will be able to continue pool operations while it is | |
1329 | partially failed, and when it recovers, we can continue to free the | |
1330 | space, with no leaks. However, note that this case is actually | |
1331 | fairly rare. | |
1332 | ||
1333 | Typically pools either (a) fail completely (but perhaps temporarily, | |
1334 | e.g. a top-level vdev going offline), or (b) have localized, | |
1335 | permanent errors (e.g. disk returns the wrong data due to bit flip or | |
1336 | firmware bug). In case (a), this setting does not matter because the | |
1337 | pool will be suspended and the sync thread will not be able to make | |
1338 | forward progress regardless. In case (b), because the error is | |
1339 | permanent, the best we can do is leak the minimum amount of space, | |
1340 | which is what setting this flag will do. Therefore, it is reasonable | |
1341 | for this flag to normally be set, but we chose the more conservative | |
1342 | approach of not setting it, so that there is no possibility of | |
1343 | leaking space in the "partial temporary" failure case. | |
1344 | .sp | |
1345 | Default value: \fB0\fR. | |
a08ee875 LG |
1346 | .RE |
1347 | ||
1348 | .sp | |
1349 | .ne 2 | |
1350 | .na | |
1351 | \fBzfs_free_min_time_ms\fR (int) | |
1352 | .ad | |
1353 | .RS 12n | |
cae5b340 AX |
1354 | During a \fBzfs destroy\fR operation using \fBfeature@async_destroy\fR a minimum |
1355 | of this much time will be spent working on freeing blocks per txg. | |
a08ee875 LG |
1356 | .sp |
1357 | Default value: \fB1,000\fR. | |
1358 | .RE | |
1359 | ||
1360 | .sp | |
1361 | .ne 2 | |
1362 | .na | |
1363 | \fBzfs_immediate_write_sz\fR (long) | |
1364 | .ad | |
1365 | .RS 12n | |
cae5b340 AX |
1366 | Largest data block to write to zil. Larger blocks will be treated as if the |
1367 | dataset being written to had the property setting \fBlogbias=throughput\fR. | |
a08ee875 LG |
1368 | .sp |
1369 | Default value: \fB32,768\fR. | |
1370 | .RE | |
1371 | ||
e10b0808 AX |
1372 | .sp |
1373 | .ne 2 | |
1374 | .na | |
1375 | \fBzfs_max_recordsize\fR (int) | |
1376 | .ad | |
1377 | .RS 12n | |
1378 | We currently support block sizes from 512 bytes to 16MB. The benefits of | |
1379 | larger blocks, and thus larger IO, need to be weighed against the cost of | |
1380 | COWing a giant block to modify one byte. Additionally, very large blocks | |
1381 | can have an impact on i/o latency, and also potentially on the memory | |
1382 | allocator. Therefore, we do not allow the recordsize to be set larger than | |
1383 | zfs_max_recordsize (default 1MB). Larger blocks can be created by changing | |
1384 | this tunable, and pools with larger blocks can always be imported and used, | |
1385 | regardless of this setting. | |
1386 | .sp | |
1387 | Default value: \fB1,048,576\fR. | |
1388 | .RE | |
1389 | ||
a08ee875 LG |
1390 | .sp |
1391 | .ne 2 | |
1392 | .na | |
1393 | \fBzfs_mdcomp_disable\fR (int) | |
1394 | .ad | |
1395 | .RS 12n | |
1396 | Disable meta data compression | |
1397 | .sp | |
1398 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
1399 | .RE | |
1400 | ||
ea04106b AX |
1401 | .sp |
1402 | .ne 2 | |
1403 | .na | |
1404 | \fBzfs_metaslab_fragmentation_threshold\fR (int) | |
1405 | .ad | |
1406 | .RS 12n | |
1407 | Allow metaslabs to keep their active state as long as their fragmentation | |
1408 | percentage is less than or equal to this value. An active metaslab that | |
1409 | exceeds this threshold will no longer keep its active status allowing | |
1410 | better metaslabs to be selected. | |
1411 | .sp | |
1412 | Default value: \fB70\fR. | |
1413 | .RE | |
1414 | ||
1415 | .sp | |
1416 | .ne 2 | |
1417 | .na | |
1418 | \fBzfs_mg_fragmentation_threshold\fR (int) | |
1419 | .ad | |
1420 | .RS 12n | |
1421 | Metaslab groups are considered eligible for allocations if their | |
cae5b340 | 1422 | fragmentation metric (measured as a percentage) is less than or equal to |
ea04106b AX |
1423 | this value. If a metaslab group exceeds this threshold then it will be |
1424 | skipped unless all metaslab groups within the metaslab class have also | |
1425 | crossed this threshold. | |
1426 | .sp | |
1427 | Default value: \fB85\fR. | |
1428 | .RE | |
1429 | ||
1430 | .sp | |
1431 | .ne 2 | |
1432 | .na | |
1433 | \fBzfs_mg_noalloc_threshold\fR (int) | |
1434 | .ad | |
1435 | .RS 12n | |
1436 | Defines a threshold at which metaslab groups should be eligible for | |
1437 | allocations. The value is expressed as a percentage of free space | |
1438 | beyond which a metaslab group is always eligible for allocations. | |
1439 | If a metaslab group's free space is less than or equal to the | |
cae5b340 | 1440 | threshold, the allocator will avoid allocating to that group |
ea04106b AX |
1441 | unless all groups in the pool have reached the threshold. Once all |
1442 | groups have reached the threshold, all groups are allowed to accept | |
1443 | allocations. The default value of 0 disables the feature and causes | |
1444 | all metaslab groups to be eligible for allocations. | |
1445 | ||
cae5b340 | 1446 | This parameter allows one to deal with pools having heavily imbalanced |
ea04106b AX |
1447 | vdevs such as would be the case when a new vdev has been added. |
1448 | Setting the threshold to a non-zero percentage will stop allocations | |
1449 | from being made to vdevs that aren't filled to the specified percentage | |
1450 | and allow lesser filled vdevs to acquire more allocations than they | |
1451 | otherwise would under the old \fBzfs_mg_alloc_failures\fR facility. | |
1452 | .sp | |
1453 | Default value: \fB0\fR. | |
1454 | .RE | |
1455 | ||
cae5b340 AX |
1456 | .sp |
1457 | .ne 2 | |
1458 | .na | |
1459 | \fBzfs_multihost_history\fR (int) | |
1460 | .ad | |
1461 | .RS 12n | |
1462 | Historical statistics for the last N multihost updates will be available in | |
1463 | \fB/proc/spl/kstat/zfs/<pool>/multihost\fR | |
1464 | .sp | |
1465 | Default value: \fB0\fR. | |
1466 | .RE | |
1467 | ||
1468 | .sp | |
1469 | .ne 2 | |
1470 | .na | |
1471 | \fBzfs_multihost_interval\fR (ulong) | |
1472 | .ad | |
1473 | .RS 12n | |
1474 | Used to control the frequency of multihost writes which are performed when the | |
1475 | \fBmultihost\fR pool property is on. This is one factor used to determine | |
1476 | the length of the activity check during import. | |
1477 | .sp | |
1478 | The multihost write period is \fBzfs_multihost_interval / leaf-vdevs\fR milliseconds. | |
1479 | This means that on average a multihost write will be issued for each leaf vdev every | |
1480 | \fBzfs_multihost_interval\fR milliseconds. In practice, the observed period can | |
1481 | vary with the I/O load and this observed value is the delay which is stored in | |
1482 | the uberblock. | |
1483 | .sp | |
1484 | On import the activity check waits a minimum amount of time determined by | |
1485 | \fBzfs_multihost_interval * zfs_multihost_import_intervals\fR. The activity | |
1486 | check time may be further extended if the value of mmp delay found in the best | |
1487 | uberblock indicates actual multihost updates happened at longer intervals than | |
1488 | \fBzfs_multihost_interval\fR. A minimum value of \fB100ms\fR is enforced. | |
1489 | .sp | |
1490 | Default value: \fB1000\fR. | |
1491 | .RE | |
1492 | ||
1493 | .sp | |
1494 | .ne 2 | |
1495 | .na | |
1496 | \fBzfs_multihost_import_intervals\fR (uint) | |
1497 | .ad | |
1498 | .RS 12n | |
1499 | Used to control the duration of the activity test on import. Smaller values of | |
1500 | \fBzfs_multihost_import_intervals\fR will reduce the import time but increase | |
1501 | the risk of failing to detect an active pool. The total activity check time is | |
1502 | never allowed to drop below one second. A value of 0 is ignored and treated as | |
1503 | if it was set to 1 | |
1504 | .sp | |
1505 | Default value: \fB10\fR. | |
1506 | .RE | |
1507 | ||
1508 | .sp | |
1509 | .ne 2 | |
1510 | .na | |
1511 | \fBzfs_multihost_fail_intervals\fR (uint) | |
1512 | .ad | |
1513 | .RS 12n | |
1514 | Controls the behavior of the pool when multihost write failures are detected. | |
1515 | .sp | |
1516 | When \fBzfs_multihost_fail_intervals = 0\fR then multihost write failures are ignored. | |
1517 | The failures will still be reported to the ZED which depending on its | |
1518 | configuration may take action such as suspending the pool or offlining a device. | |
1519 | .sp | |
1520 | When \fBzfs_multihost_fail_intervals > 0\fR then sequential multihost write failures | |
1521 | will cause the pool to be suspended. This occurs when | |
1522 | \fBzfs_multihost_fail_intervals * zfs_multihost_interval\fR milliseconds have | |
1523 | passed since the last successful multihost write. This guarantees the activity test | |
1524 | will see multihost writes if the pool is imported. | |
1525 | .sp | |
1526 | Default value: \fB5\fR. | |
1527 | .RE | |
1528 | ||
a08ee875 LG |
1529 | .sp |
1530 | .ne 2 | |
1531 | .na | |
1532 | \fBzfs_no_scrub_io\fR (int) | |
1533 | .ad | |
1534 | .RS 12n | |
cae5b340 AX |
1535 | Set for no scrub I/O. This results in scrubs not actually scrubbing data and |
1536 | simply doing a metadata crawl of the pool instead. | |
a08ee875 LG |
1537 | .sp |
1538 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
1539 | .RE | |
1540 | ||
1541 | .sp | |
1542 | .ne 2 | |
1543 | .na | |
1544 | \fBzfs_no_scrub_prefetch\fR (int) | |
1545 | .ad | |
1546 | .RS 12n | |
cae5b340 | 1547 | Set to disable block prefetching for scrubs. |
a08ee875 LG |
1548 | .sp |
1549 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
1550 | .RE | |
1551 | ||
1552 | .sp | |
1553 | .ne 2 | |
1554 | .na | |
1555 | \fBzfs_nocacheflush\fR (int) | |
1556 | .ad | |
1557 | .RS 12n | |
cae5b340 AX |
1558 | Disable cache flush operations on disks when writing. Beware, this may cause |
1559 | corruption if disks re-order writes. | |
a08ee875 LG |
1560 | .sp |
1561 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
1562 | .RE | |
1563 | ||
1564 | .sp | |
1565 | .ne 2 | |
1566 | .na | |
1567 | \fBzfs_nopwrite_enabled\fR (int) | |
1568 | .ad | |
1569 | .RS 12n | |
1570 | Enable NOP writes | |
1571 | .sp | |
1572 | Use \fB1\fR for yes (default) and \fB0\fR to disable. | |
1573 | .RE | |
1574 | ||
cae5b340 AX |
1575 | .sp |
1576 | .ne 2 | |
1577 | .na | |
1578 | \fBzfs_dmu_offset_next_sync\fR (int) | |
1579 | .ad | |
1580 | .RS 12n | |
1581 | Enable forcing txg sync to find holes. When enabled forces ZFS to act | |
1582 | like prior versions when SEEK_HOLE or SEEK_DATA flags are used, which | |
1583 | when a dnode is dirty causes txg's to be synced so that this data can be | |
1584 | found. | |
1585 | .sp | |
1586 | Use \fB1\fR for yes and \fB0\fR to disable (default). | |
1587 | .RE | |
1588 | ||
a08ee875 LG |
1589 | .sp |
1590 | .ne 2 | |
1591 | .na | |
ea04106b | 1592 | \fBzfs_pd_bytes_max\fR (int) |
a08ee875 LG |
1593 | .ad |
1594 | .RS 12n | |
cae5b340 AX |
1595 | The number of bytes which should be prefetched during a pool traversal |
1596 | (eg: \fBzfs send\fR or other data crawling operations) | |
a08ee875 | 1597 | .sp |
ea04106b | 1598 | Default value: \fB52,428,800\fR. |
a08ee875 LG |
1599 | .RE |
1600 | ||
cae5b340 AX |
1601 | .sp |
1602 | .ne 2 | |
1603 | .na | |
1604 | \fBzfs_per_txg_dirty_frees_percent \fR (ulong) | |
1605 | .ad | |
1606 | .RS 12n | |
1607 | Tunable to control percentage of dirtied blocks from frees in one TXG. | |
1608 | After this threshold is crossed, additional dirty blocks from frees | |
1609 | wait until the next TXG. | |
1610 | A value of zero will disable this throttle. | |
1611 | .sp | |
1612 | Default value: \fB30\fR and \fB0\fR to disable. | |
1613 | .RE | |
1614 | ||
1615 | ||
1616 | ||
a08ee875 LG |
1617 | .sp |
1618 | .ne 2 | |
1619 | .na | |
1620 | \fBzfs_prefetch_disable\fR (int) | |
1621 | .ad | |
1622 | .RS 12n | |
cae5b340 AX |
1623 | This tunable disables predictive prefetch. Note that it leaves "prescient" |
1624 | prefetch (e.g. prefetch for zfs send) intact. Unlike predictive prefetch, | |
1625 | prescient prefetch never issues i/os that end up not being needed, so it | |
1626 | can't hurt performance. | |
a08ee875 LG |
1627 | .sp |
1628 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
1629 | .RE | |
1630 | ||
1631 | .sp | |
1632 | .ne 2 | |
1633 | .na | |
1634 | \fBzfs_read_chunk_size\fR (long) | |
1635 | .ad | |
1636 | .RS 12n | |
1637 | Bytes to read per chunk | |
1638 | .sp | |
1639 | Default value: \fB1,048,576\fR. | |
1640 | .RE | |
1641 | ||
1642 | .sp | |
1643 | .ne 2 | |
1644 | .na | |
1645 | \fBzfs_read_history\fR (int) | |
1646 | .ad | |
1647 | .RS 12n | |
cae5b340 AX |
1648 | Historical statistics for the last N reads will be available in |
1649 | \fB/proc/spl/kstat/zfs/<pool>/reads\fR | |
a08ee875 | 1650 | .sp |
cae5b340 | 1651 | Default value: \fB0\fR (no data is kept). |
a08ee875 LG |
1652 | .RE |
1653 | ||
1654 | .sp | |
1655 | .ne 2 | |
1656 | .na | |
1657 | \fBzfs_read_history_hits\fR (int) | |
1658 | .ad | |
1659 | .RS 12n | |
1660 | Include cache hits in read history | |
1661 | .sp | |
1662 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
1663 | .RE | |
1664 | ||
1665 | .sp | |
1666 | .ne 2 | |
1667 | .na | |
1668 | \fBzfs_recover\fR (int) | |
1669 | .ad | |
1670 | .RS 12n | |
1671 | Set to attempt to recover from fatal errors. This should only be used as a | |
1672 | last resort, as it typically results in leaked space, or worse. | |
1673 | .sp | |
1674 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
1675 | .RE | |
1676 | ||
1677 | .sp | |
1678 | .ne 2 | |
1679 | .na | |
1680 | \fBzfs_resilver_delay\fR (int) | |
1681 | .ad | |
1682 | .RS 12n | |
ea04106b AX |
1683 | Number of ticks to delay prior to issuing a resilver I/O operation when |
1684 | a non-resilver or non-scrub I/O operation has occurred within the past | |
1685 | \fBzfs_scan_idle\fR ticks. | |
a08ee875 LG |
1686 | .sp |
1687 | Default value: \fB2\fR. | |
1688 | .RE | |
1689 | ||
1690 | .sp | |
1691 | .ne 2 | |
1692 | .na | |
1693 | \fBzfs_resilver_min_time_ms\fR (int) | |
1694 | .ad | |
1695 | .RS 12n | |
cae5b340 AX |
1696 | Resilvers are processed by the sync thread. While resilvering it will spend |
1697 | at least this much time working on a resilver between txg flushes. | |
a08ee875 LG |
1698 | .sp |
1699 | Default value: \fB3,000\fR. | |
1700 | .RE | |
1701 | ||
42f7b73b AX |
1702 | .sp |
1703 | .ne 2 | |
1704 | .na | |
1705 | \fBzfs_scan_ignore_errors\fR (int) | |
1706 | .ad | |
1707 | .RS 12n | |
1708 | If set to a nonzero value, remove the DTL (dirty time list) upon | |
1709 | completion of a pool scan (scrub) even if there were unrepairable | |
1710 | errors. It is intended to be used during pool repair or recovery to | |
1711 | stop resilvering when the pool is next imported. | |
1712 | .sp | |
1713 | Default value: \fB0\fR. | |
1714 | .RE | |
1715 | ||
a08ee875 LG |
1716 | .sp |
1717 | .ne 2 | |
1718 | .na | |
1719 | \fBzfs_scan_idle\fR (int) | |
1720 | .ad | |
1721 | .RS 12n | |
ea04106b AX |
1722 | Idle window in clock ticks. During a scrub or a resilver, if |
1723 | a non-scrub or non-resilver I/O operation has occurred during this | |
1724 | window, the next scrub or resilver operation is delayed by, respectively | |
1725 | \fBzfs_scrub_delay\fR or \fBzfs_resilver_delay\fR ticks. | |
a08ee875 LG |
1726 | .sp |
1727 | Default value: \fB50\fR. | |
1728 | .RE | |
1729 | ||
1730 | .sp | |
1731 | .ne 2 | |
1732 | .na | |
1733 | \fBzfs_scan_min_time_ms\fR (int) | |
1734 | .ad | |
1735 | .RS 12n | |
cae5b340 AX |
1736 | Scrubs are processed by the sync thread. While scrubbing it will spend |
1737 | at least this much time working on a scrub between txg flushes. | |
a08ee875 LG |
1738 | .sp |
1739 | Default value: \fB1,000\fR. | |
1740 | .RE | |
1741 | ||
1742 | .sp | |
1743 | .ne 2 | |
1744 | .na | |
1745 | \fBzfs_scrub_delay\fR (int) | |
1746 | .ad | |
1747 | .RS 12n | |
ea04106b AX |
1748 | Number of ticks to delay prior to issuing a scrub I/O operation when |
1749 | a non-scrub or non-resilver I/O operation has occurred within the past | |
1750 | \fBzfs_scan_idle\fR ticks. | |
a08ee875 LG |
1751 | .sp |
1752 | Default value: \fB4\fR. | |
1753 | .RE | |
1754 | ||
1755 | .sp | |
1756 | .ne 2 | |
1757 | .na | |
1758 | \fBzfs_send_corrupt_data\fR (int) | |
1759 | .ad | |
1760 | .RS 12n | |
cae5b340 | 1761 | Allow sending of corrupt data (ignore read/checksum errors when sending data) |
a08ee875 LG |
1762 | .sp |
1763 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
1764 | .RE | |
1765 | ||
42f7b73b AX |
1766 | .sp |
1767 | .ne 2 | |
1768 | .na | |
1769 | \fBzfs_send_queue_length\fR (int) | |
1770 | .ad | |
1771 | .RS 12n | |
1772 | The maximum number of bytes allowed in the \fBzfs send\fR queue. This value | |
1773 | must be at least twice the maximum block size in use. | |
1774 | .sp | |
1775 | Default value: \fB16,777,216\fR. | |
1776 | .RE | |
1777 | ||
1778 | .sp | |
1779 | .ne 2 | |
1780 | .na | |
1781 | \fBzfs_recv_queue_length\fR (int) | |
1782 | .ad | |
1783 | .RS 12n | |
1784 | .sp | |
1785 | The maximum number of bytes allowed in the \fBzfs receive\fR queue. This value | |
1786 | must be at least twice the maximum block size in use. | |
1787 | .sp | |
1788 | Default value: \fB16,777,216\fR. | |
1789 | .RE | |
1790 | ||
a08ee875 LG |
1791 | .sp |
1792 | .ne 2 | |
1793 | .na | |
1794 | \fBzfs_sync_pass_deferred_free\fR (int) | |
1795 | .ad | |
1796 | .RS 12n | |
cae5b340 | 1797 | Flushing of data to disk is done in passes. Defer frees starting in this pass |
a08ee875 LG |
1798 | .sp |
1799 | Default value: \fB2\fR. | |
1800 | .RE | |
1801 | ||
1802 | .sp | |
1803 | .ne 2 | |
1804 | .na | |
1805 | \fBzfs_sync_pass_dont_compress\fR (int) | |
1806 | .ad | |
1807 | .RS 12n | |
1808 | Don't compress starting in this pass | |
1809 | .sp | |
1810 | Default value: \fB5\fR. | |
1811 | .RE | |
1812 | ||
1813 | .sp | |
1814 | .ne 2 | |
1815 | .na | |
1816 | \fBzfs_sync_pass_rewrite\fR (int) | |
1817 | .ad | |
1818 | .RS 12n | |
cae5b340 | 1819 | Rewrite new block pointers starting in this pass |
a08ee875 LG |
1820 | .sp |
1821 | Default value: \fB2\fR. | |
1822 | .RE | |
1823 | ||
1824 | .sp | |
1825 | .ne 2 | |
1826 | .na | |
1827 | \fBzfs_top_maxinflight\fR (int) | |
1828 | .ad | |
1829 | .RS 12n | |
cae5b340 AX |
1830 | Max concurrent I/Os per top-level vdev (mirrors or raidz arrays) allowed during |
1831 | scrub or resilver operations. | |
a08ee875 LG |
1832 | .sp |
1833 | Default value: \fB32\fR. | |
1834 | .RE | |
1835 | ||
1836 | .sp | |
1837 | .ne 2 | |
1838 | .na | |
1839 | \fBzfs_txg_history\fR (int) | |
1840 | .ad | |
1841 | .RS 12n | |
cae5b340 AX |
1842 | Historical statistics for the last N txgs will be available in |
1843 | \fB/proc/spl/kstat/zfs/<pool>/txgs\fR | |
a08ee875 LG |
1844 | .sp |
1845 | Default value: \fB0\fR. | |
1846 | .RE | |
1847 | ||
1848 | .sp | |
1849 | .ne 2 | |
1850 | .na | |
1851 | \fBzfs_txg_timeout\fR (int) | |
1852 | .ad | |
1853 | .RS 12n | |
cae5b340 | 1854 | Flush dirty data to disk at least every N seconds (maximum txg duration) |
a08ee875 LG |
1855 | .sp |
1856 | Default value: \fB5\fR. | |
1857 | .RE | |
1858 | ||
1859 | .sp | |
1860 | .ne 2 | |
1861 | .na | |
1862 | \fBzfs_vdev_aggregation_limit\fR (int) | |
1863 | .ad | |
1864 | .RS 12n | |
1865 | Max vdev I/O aggregation size | |
1866 | .sp | |
1867 | Default value: \fB131,072\fR. | |
1868 | .RE | |
1869 | ||
1870 | .sp | |
1871 | .ne 2 | |
1872 | .na | |
1873 | \fBzfs_vdev_cache_bshift\fR (int) | |
1874 | .ad | |
1875 | .RS 12n | |
1876 | Shift size to inflate reads too | |
1877 | .sp | |
cae5b340 | 1878 | Default value: \fB16\fR (effectively 65536). |
a08ee875 LG |
1879 | .RE |
1880 | ||
1881 | .sp | |
1882 | .ne 2 | |
1883 | .na | |
1884 | \fBzfs_vdev_cache_max\fR (int) | |
1885 | .ad | |
1886 | .RS 12n | |
41d74433 AX |
1887 | Inflate reads smaller than this value to meet the \fBzfs_vdev_cache_bshift\fR |
1888 | size (default 64k). | |
cae5b340 AX |
1889 | .sp |
1890 | Default value: \fB16384\fR. | |
a08ee875 LG |
1891 | .RE |
1892 | ||
1893 | .sp | |
1894 | .ne 2 | |
1895 | .na | |
1896 | \fBzfs_vdev_cache_size\fR (int) | |
1897 | .ad | |
1898 | .RS 12n | |
cae5b340 AX |
1899 | Total size of the per-disk cache in bytes. |
1900 | .sp | |
1901 | Currently this feature is disabled as it has been found to not be helpful | |
1902 | for performance and in some cases harmful. | |
a08ee875 LG |
1903 | .sp |
1904 | Default value: \fB0\fR. | |
1905 | .RE | |
1906 | ||
1907 | .sp | |
1908 | .ne 2 | |
1909 | .na | |
cae5b340 | 1910 | \fBzfs_vdev_mirror_rotating_inc\fR (int) |
a08ee875 LG |
1911 | .ad |
1912 | .RS 12n | |
cae5b340 AX |
1913 | A number by which the balancing algorithm increments the load calculation for |
1914 | the purpose of selecting the least busy mirror member when an I/O immediately | |
1915 | follows its predecessor on rotational vdevs for the purpose of making decisions | |
1916 | based on load. | |
a08ee875 | 1917 | .sp |
cae5b340 AX |
1918 | Default value: \fB0\fR. |
1919 | .RE | |
1920 | ||
1921 | .sp | |
1922 | .ne 2 | |
1923 | .na | |
1924 | \fBzfs_vdev_mirror_rotating_seek_inc\fR (int) | |
1925 | .ad | |
1926 | .RS 12n | |
1927 | A number by which the balancing algorithm increments the load calculation for | |
1928 | the purpose of selecting the least busy mirror member when an I/O lacks | |
1929 | locality as defined by the zfs_vdev_mirror_rotating_seek_offset. I/Os within | |
1930 | this that are not immediately following the previous I/O are incremented by | |
1931 | half. | |
1932 | .sp | |
1933 | Default value: \fB5\fR. | |
1934 | .RE | |
1935 | ||
1936 | .sp | |
1937 | .ne 2 | |
1938 | .na | |
1939 | \fBzfs_vdev_mirror_rotating_seek_offset\fR (int) | |
1940 | .ad | |
1941 | .RS 12n | |
1942 | The maximum distance for the last queued I/O in which the balancing algorithm | |
1943 | considers an I/O to have locality. | |
1944 | See the section "ZFS I/O SCHEDULER". | |
1945 | .sp | |
1946 | Default value: \fB1048576\fR. | |
1947 | .RE | |
1948 | ||
1949 | .sp | |
1950 | .ne 2 | |
1951 | .na | |
1952 | \fBzfs_vdev_mirror_non_rotating_inc\fR (int) | |
1953 | .ad | |
1954 | .RS 12n | |
1955 | A number by which the balancing algorithm increments the load calculation for | |
1956 | the purpose of selecting the least busy mirror member on non-rotational vdevs | |
1957 | when I/Os do not immediately follow one another. | |
1958 | .sp | |
1959 | Default value: \fB0\fR. | |
1960 | .RE | |
1961 | ||
1962 | .sp | |
1963 | .ne 2 | |
1964 | .na | |
1965 | \fBzfs_vdev_mirror_non_rotating_seek_inc\fR (int) | |
1966 | .ad | |
1967 | .RS 12n | |
1968 | A number by which the balancing algorithm increments the load calculation for | |
1969 | the purpose of selecting the least busy mirror member when an I/O lacks | |
1970 | locality as defined by the zfs_vdev_mirror_rotating_seek_offset. I/Os within | |
1971 | this that are not immediately following the previous I/O are incremented by | |
1972 | half. | |
1973 | .sp | |
1974 | Default value: \fB1\fR. | |
a08ee875 LG |
1975 | .RE |
1976 | ||
1977 | .sp | |
1978 | .ne 2 | |
1979 | .na | |
1980 | \fBzfs_vdev_read_gap_limit\fR (int) | |
1981 | .ad | |
1982 | .RS 12n | |
cae5b340 AX |
1983 | Aggregate read I/O operations if the gap on-disk between them is within this |
1984 | threshold. | |
a08ee875 LG |
1985 | .sp |
1986 | Default value: \fB32,768\fR. | |
1987 | .RE | |
1988 | ||
1989 | .sp | |
1990 | .ne 2 | |
1991 | .na | |
1992 | \fBzfs_vdev_scheduler\fR (charp) | |
1993 | .ad | |
1994 | .RS 12n | |
41d74433 AX |
1995 | Set the Linux I/O scheduler on whole disk vdevs to this scheduler. Valid options |
1996 | are noop, cfq, bfq & deadline | |
a08ee875 LG |
1997 | .sp |
1998 | Default value: \fBnoop\fR. | |
1999 | .RE | |
2000 | ||
2001 | .sp | |
2002 | .ne 2 | |
2003 | .na | |
2004 | \fBzfs_vdev_write_gap_limit\fR (int) | |
2005 | .ad | |
2006 | .RS 12n | |
2007 | Aggregate write I/O over gap | |
2008 | .sp | |
2009 | Default value: \fB4,096\fR. | |
2010 | .RE | |
2011 | ||
cae5b340 AX |
2012 | .sp |
2013 | .ne 2 | |
2014 | .na | |
2015 | \fBzfs_vdev_raidz_impl\fR (string) | |
2016 | .ad | |
2017 | .RS 12n | |
2018 | Parameter for selecting raidz parity implementation to use. | |
2019 | ||
2020 | Options marked (always) below may be selected on module load as they are | |
2021 | supported on all systems. | |
2022 | The remaining options may only be set after the module is loaded, as they | |
2023 | are available only if the implementations are compiled in and supported | |
2024 | on the running system. | |
2025 | ||
2026 | Once the module is loaded, the content of | |
2027 | /sys/module/zfs/parameters/zfs_vdev_raidz_impl will show available options | |
2028 | with the currently selected one enclosed in []. | |
2029 | Possible options are: | |
2030 | fastest - (always) implementation selected using built-in benchmark | |
2031 | original - (always) original raidz implementation | |
2032 | scalar - (always) scalar raidz implementation | |
2033 | sse2 - implementation using SSE2 instruction set (64bit x86 only) | |
2034 | ssse3 - implementation using SSSE3 instruction set (64bit x86 only) | |
2035 | avx2 - implementation using AVX2 instruction set (64bit x86 only) | |
2036 | avx512f - implementation using AVX512F instruction set (64bit x86 only) | |
2037 | avx512bw - implementation using AVX512F & AVX512BW instruction sets (64bit x86 only) | |
2038 | aarch64_neon - implementation using NEON (Aarch64/64 bit ARMv8 only) | |
2039 | aarch64_neonx2 - implementation using NEON with more unrolling (Aarch64/64 bit ARMv8 only) | |
2040 | .sp | |
2041 | Default value: \fBfastest\fR. | |
2042 | .RE | |
2043 | ||
a08ee875 LG |
2044 | .sp |
2045 | .ne 2 | |
2046 | .na | |
2047 | \fBzfs_zevent_cols\fR (int) | |
2048 | .ad | |
2049 | .RS 12n | |
cae5b340 | 2050 | When zevents are logged to the console use this as the word wrap width. |
a08ee875 LG |
2051 | .sp |
2052 | Default value: \fB80\fR. | |
2053 | .RE | |
2054 | ||
2055 | .sp | |
2056 | .ne 2 | |
2057 | .na | |
2058 | \fBzfs_zevent_console\fR (int) | |
2059 | .ad | |
2060 | .RS 12n | |
2061 | Log events to the console | |
2062 | .sp | |
2063 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
2064 | .RE | |
2065 | ||
2066 | .sp | |
2067 | .ne 2 | |
2068 | .na | |
2069 | \fBzfs_zevent_len_max\fR (int) | |
2070 | .ad | |
2071 | .RS 12n | |
cae5b340 AX |
2072 | Max event queue length. A value of 0 will result in a calculated value which |
2073 | increases with the number of CPUs in the system (minimum 64 events). Events | |
2074 | in the queue can be viewed with the \fBzpool events\fR command. | |
a08ee875 LG |
2075 | .sp |
2076 | Default value: \fB0\fR. | |
2077 | .RE | |
2078 | ||
2079 | .sp | |
2080 | .ne 2 | |
2081 | .na | |
2082 | \fBzil_replay_disable\fR (int) | |
2083 | .ad | |
2084 | .RS 12n | |
cae5b340 AX |
2085 | Disable intent logging replay. Can be disabled for recovery from corrupted |
2086 | ZIL | |
a08ee875 LG |
2087 | .sp |
2088 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
2089 | .RE | |
2090 | ||
2091 | .sp | |
2092 | .ne 2 | |
2093 | .na | |
cae5b340 | 2094 | \fBzil_slog_bulk\fR (ulong) |
a08ee875 LG |
2095 | .ad |
2096 | .RS 12n | |
cae5b340 AX |
2097 | Limit SLOG write size per commit executed with synchronous priority. |
2098 | Any writes above that will be executed with lower (asynchronous) priority | |
2099 | to limit potential SLOG device abuse by single active ZIL writer. | |
a08ee875 | 2100 | .sp |
cae5b340 | 2101 | Default value: \fB786,432\fR. |
a08ee875 LG |
2102 | .RE |
2103 | ||
a08ee875 LG |
2104 | .sp |
2105 | .ne 2 | |
2106 | .na | |
2107 | \fBzio_delay_max\fR (int) | |
2108 | .ad | |
2109 | .RS 12n | |
cae5b340 AX |
2110 | A zevent will be logged if a ZIO operation takes more than N milliseconds to |
2111 | complete. Note that this is only a logging facility, not a timeout on | |
2112 | operations. | |
a08ee875 LG |
2113 | .sp |
2114 | Default value: \fB30,000\fR. | |
2115 | .RE | |
2116 | ||
cae5b340 AX |
2117 | .sp |
2118 | .ne 2 | |
2119 | .na | |
2120 | \fBzio_dva_throttle_enabled\fR (int) | |
2121 | .ad | |
2122 | .RS 12n | |
2123 | Throttle block allocations in the ZIO pipeline. This allows for | |
2124 | dynamic allocation distribution when devices are imbalanced. | |
2125 | When enabled, the maximum number of pending allocations per top-level vdev | |
2126 | is limited by \fBzfs_vdev_queue_depth_pct\fR. | |
2127 | .sp | |
2128 | Default value: \fB1\fR. | |
2129 | .RE | |
2130 | ||
a08ee875 LG |
2131 | .sp |
2132 | .ne 2 | |
2133 | .na | |
2134 | \fBzio_requeue_io_start_cut_in_line\fR (int) | |
2135 | .ad | |
2136 | .RS 12n | |
2137 | Prioritize requeued I/O | |
2138 | .sp | |
2139 | Default value: \fB0\fR. | |
2140 | .RE | |
2141 | ||
94a40997 AX |
2142 | .sp |
2143 | .ne 2 | |
2144 | .na | |
2145 | \fBzio_taskq_batch_pct\fR (uint) | |
2146 | .ad | |
2147 | .RS 12n | |
2148 | Percentage of online CPUs (or CPU cores, etc) which will run a worker thread | |
2149 | for IO. These workers are responsible for IO work such as compression and | |
2150 | checksum calculations. Fractional number of CPUs will be rounded down. | |
2151 | .sp | |
2152 | The default value of 75 was chosen to avoid using all CPUs which can result in | |
2153 | latency issues and inconsistent application performance, especially when high | |
2154 | compression is enabled. | |
2155 | .sp | |
2156 | Default value: \fB75\fR. | |
2157 | .RE | |
2158 | ||
a08ee875 LG |
2159 | .sp |
2160 | .ne 2 | |
2161 | .na | |
2162 | \fBzvol_inhibit_dev\fR (uint) | |
2163 | .ad | |
2164 | .RS 12n | |
cae5b340 AX |
2165 | Do not create zvol device nodes. This may slightly improve startup time on |
2166 | systems with a very large number of zvols. | |
a08ee875 LG |
2167 | .sp |
2168 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
2169 | .RE | |
2170 | ||
2171 | .sp | |
2172 | .ne 2 | |
2173 | .na | |
2174 | \fBzvol_major\fR (uint) | |
2175 | .ad | |
2176 | .RS 12n | |
cae5b340 | 2177 | Major number for zvol block devices |
a08ee875 LG |
2178 | .sp |
2179 | Default value: \fB230\fR. | |
2180 | .RE | |
2181 | ||
2182 | .sp | |
2183 | .ne 2 | |
2184 | .na | |
2185 | \fBzvol_max_discard_blocks\fR (ulong) | |
2186 | .ad | |
2187 | .RS 12n | |
cae5b340 AX |
2188 | Discard (aka TRIM) operations done on zvols will be done in batches of this |
2189 | many blocks, where block size is determined by the \fBvolblocksize\fR property | |
2190 | of a zvol. | |
a08ee875 LG |
2191 | .sp |
2192 | Default value: \fB16,384\fR. | |
2193 | .RE | |
2194 | ||
2195 | .sp | |
2196 | .ne 2 | |
2197 | .na | |
e10b0808 | 2198 | \fBzvol_prefetch_bytes\fR (uint) |
a08ee875 LG |
2199 | .ad |
2200 | .RS 12n | |
e10b0808 AX |
2201 | When adding a zvol to the system prefetch \fBzvol_prefetch_bytes\fR |
2202 | from the start and end of the volume. Prefetching these regions | |
2203 | of the volume is desirable because they are likely to be accessed | |
2204 | immediately by \fBblkid(8)\fR or by the kernel scanning for a partition | |
2205 | table. | |
a08ee875 | 2206 | .sp |
e10b0808 | 2207 | Default value: \fB131,072\fR. |
a08ee875 LG |
2208 | .RE |
2209 | ||
cae5b340 AX |
2210 | .sp |
2211 | .ne 2 | |
2212 | .na | |
2213 | \fBzvol_request_sync\fR (uint) | |
2214 | .ad | |
2215 | .RS 12n | |
2216 | When processing I/O requests for a zvol submit them synchronously. This | |
2217 | effectively limits the queue depth to 1 for each I/O submitter. When set | |
2218 | to 0 requests are handled asynchronously by a thread pool. The number of | |
2219 | requests which can be handled concurrently is controller by \fBzvol_threads\fR. | |
2220 | .sp | |
2221 | Default value: \fB0\fR. | |
2222 | .RE | |
2223 | ||
2224 | .sp | |
2225 | .ne 2 | |
2226 | .na | |
2227 | \fBzvol_threads\fR (uint) | |
2228 | .ad | |
2229 | .RS 12n | |
2230 | Max number of threads which can handle zvol I/O requests concurrently. | |
2231 | .sp | |
2232 | Default value: \fB32\fR. | |
2233 | .RE | |
2234 | ||
2235 | .sp | |
2236 | .ne 2 | |
2237 | .na | |
2238 | \fBzvol_volmode\fR (uint) | |
2239 | .ad | |
2240 | .RS 12n | |
2241 | Defines zvol block devices behaviour when \fBvolmode\fR is set to \fBdefault\fR. | |
2242 | Valid values are \fB1\fR (full), \fB2\fR (dev) and \fB3\fR (none). | |
2243 | .sp | |
2244 | Default value: \fB1\fR. | |
2245 | .RE | |
2246 | ||
2247 | .sp | |
2248 | .ne 2 | |
2249 | .na | |
2250 | \fBzfs_qat_disable\fR (int) | |
2251 | .ad | |
2252 | .RS 12n | |
2253 | This tunable disables qat hardware acceleration for gzip compression. | |
2254 | It is available only if qat acceleration is compiled in and qat driver | |
2255 | is present. | |
2256 | .sp | |
2257 | Use \fB1\fR for yes and \fB0\fR for no (default). | |
2258 | .RE | |
2259 | ||
a08ee875 LG |
2260 | .SH ZFS I/O SCHEDULER |
2261 | ZFS issues I/O operations to leaf vdevs to satisfy and complete I/Os. | |
2262 | The I/O scheduler determines when and in what order those operations are | |
2263 | issued. The I/O scheduler divides operations into five I/O classes | |
2264 | prioritized in the following order: sync read, sync write, async read, | |
2265 | async write, and scrub/resilver. Each queue defines the minimum and | |
2266 | maximum number of concurrent operations that may be issued to the | |
2267 | device. In addition, the device has an aggregate maximum, | |
2268 | \fBzfs_vdev_max_active\fR. Note that the sum of the per-queue minimums | |
2269 | must not exceed the aggregate maximum. If the sum of the per-queue | |
2270 | maximums exceeds the aggregate maximum, then the number of active I/Os | |
2271 | may reach \fBzfs_vdev_max_active\fR, in which case no further I/Os will | |
2272 | be issued regardless of whether all per-queue minimums have been met. | |
2273 | .sp | |
2274 | For many physical devices, throughput increases with the number of | |
2275 | concurrent operations, but latency typically suffers. Further, physical | |
2276 | devices typically have a limit at which more concurrent operations have no | |
2277 | effect on throughput or can actually cause it to decrease. | |
2278 | .sp | |
2279 | The scheduler selects the next operation to issue by first looking for an | |
2280 | I/O class whose minimum has not been satisfied. Once all are satisfied and | |
2281 | the aggregate maximum has not been hit, the scheduler looks for classes | |
2282 | whose maximum has not been satisfied. Iteration through the I/O classes is | |
2283 | done in the order specified above. No further operations are issued if the | |
2284 | aggregate maximum number of concurrent operations has been hit or if there | |
2285 | are no operations queued for an I/O class that has not hit its maximum. | |
2286 | Every time an I/O is queued or an operation completes, the I/O scheduler | |
2287 | looks for new operations to issue. | |
2288 | .sp | |
2289 | In general, smaller max_active's will lead to lower latency of synchronous | |
2290 | operations. Larger max_active's may lead to higher overall throughput, | |
2291 | depending on underlying storage. | |
2292 | .sp | |
2293 | The ratio of the queues' max_actives determines the balance of performance | |
2294 | between reads, writes, and scrubs. E.g., increasing | |
2295 | \fBzfs_vdev_scrub_max_active\fR will cause the scrub or resilver to complete | |
2296 | more quickly, but reads and writes to have higher latency and lower throughput. | |
2297 | .sp | |
2298 | All I/O classes have a fixed maximum number of outstanding operations | |
2299 | except for the async write class. Asynchronous writes represent the data | |
2300 | that is committed to stable storage during the syncing stage for | |
2301 | transaction groups. Transaction groups enter the syncing state | |
2302 | periodically so the number of queued async writes will quickly burst up | |
2303 | and then bleed down to zero. Rather than servicing them as quickly as | |
2304 | possible, the I/O scheduler changes the maximum number of active async | |
2305 | write I/Os according to the amount of dirty data in the pool. Since | |
2306 | both throughput and latency typically increase with the number of | |
2307 | concurrent operations issued to physical devices, reducing the | |
2308 | burstiness in the number of concurrent operations also stabilizes the | |
2309 | response time of operations from other -- and in particular synchronous | |
2310 | -- queues. In broad strokes, the I/O scheduler will issue more | |
2311 | concurrent operations from the async write queue as there's more dirty | |
2312 | data in the pool. | |
2313 | .sp | |
2314 | Async Writes | |
2315 | .sp | |
2316 | The number of concurrent operations issued for the async write I/O class | |
2317 | follows a piece-wise linear function defined by a few adjustable points. | |
2318 | .nf | |
2319 | ||
2320 | | o---------| <-- zfs_vdev_async_write_max_active | |
2321 | ^ | /^ | | |
2322 | | | / | | | |
2323 | active | / | | | |
2324 | I/O | / | | | |
2325 | count | / | | | |
2326 | | / | | | |
2327 | |-------o | | <-- zfs_vdev_async_write_min_active | |
2328 | 0|_______^______|_________| | |
2329 | 0% | | 100% of zfs_dirty_data_max | |
2330 | | | | |
2331 | | `-- zfs_vdev_async_write_active_max_dirty_percent | |
2332 | `--------- zfs_vdev_async_write_active_min_dirty_percent | |
2333 | ||
2334 | .fi | |
2335 | Until the amount of dirty data exceeds a minimum percentage of the dirty | |
2336 | data allowed in the pool, the I/O scheduler will limit the number of | |
2337 | concurrent operations to the minimum. As that threshold is crossed, the | |
2338 | number of concurrent operations issued increases linearly to the maximum at | |
2339 | the specified maximum percentage of the dirty data allowed in the pool. | |
2340 | .sp | |
2341 | Ideally, the amount of dirty data on a busy pool will stay in the sloped | |
2342 | part of the function between \fBzfs_vdev_async_write_active_min_dirty_percent\fR | |
2343 | and \fBzfs_vdev_async_write_active_max_dirty_percent\fR. If it exceeds the | |
2344 | maximum percentage, this indicates that the rate of incoming data is | |
2345 | greater than the rate that the backend storage can handle. In this case, we | |
2346 | must further throttle incoming writes, as described in the next section. | |
2347 | ||
2348 | .SH ZFS TRANSACTION DELAY | |
2349 | We delay transactions when we've determined that the backend storage | |
2350 | isn't able to accommodate the rate of incoming writes. | |
2351 | .sp | |
2352 | If there is already a transaction waiting, we delay relative to when | |
2353 | that transaction will finish waiting. This way the calculated delay time | |
2354 | is independent of the number of threads concurrently executing | |
2355 | transactions. | |
2356 | .sp | |
2357 | If we are the only waiter, wait relative to when the transaction | |
2358 | started, rather than the current time. This credits the transaction for | |
2359 | "time already served", e.g. reading indirect blocks. | |
2360 | .sp | |
2361 | The minimum time for a transaction to take is calculated as: | |
2362 | .nf | |
2363 | min_time = zfs_delay_scale * (dirty - min) / (max - dirty) | |
2364 | min_time is then capped at 100 milliseconds. | |
2365 | .fi | |
2366 | .sp | |
2367 | The delay has two degrees of freedom that can be adjusted via tunables. The | |
2368 | percentage of dirty data at which we start to delay is defined by | |
2369 | \fBzfs_delay_min_dirty_percent\fR. This should typically be at or above | |
2370 | \fBzfs_vdev_async_write_active_max_dirty_percent\fR so that we only start to | |
2371 | delay after writing at full speed has failed to keep up with the incoming write | |
2372 | rate. The scale of the curve is defined by \fBzfs_delay_scale\fR. Roughly speaking, | |
2373 | this variable determines the amount of delay at the midpoint of the curve. | |
2374 | .sp | |
2375 | .nf | |
2376 | delay | |
2377 | 10ms +-------------------------------------------------------------*+ | |
2378 | | *| | |
2379 | 9ms + *+ | |
2380 | | *| | |
2381 | 8ms + *+ | |
2382 | | * | | |
2383 | 7ms + * + | |
2384 | | * | | |
2385 | 6ms + * + | |
2386 | | * | | |
2387 | 5ms + * + | |
2388 | | * | | |
2389 | 4ms + * + | |
2390 | | * | | |
2391 | 3ms + * + | |
2392 | | * | | |
2393 | 2ms + (midpoint) * + | |
2394 | | | ** | | |
2395 | 1ms + v *** + | |
2396 | | zfs_delay_scale ----------> ******** | | |
2397 | 0 +-------------------------------------*********----------------+ | |
2398 | 0% <- zfs_dirty_data_max -> 100% | |
2399 | .fi | |
2400 | .sp | |
2401 | Note that since the delay is added to the outstanding time remaining on the | |
2402 | most recent transaction, the delay is effectively the inverse of IOPS. | |
2403 | Here the midpoint of 500us translates to 2000 IOPS. The shape of the curve | |
2404 | was chosen such that small changes in the amount of accumulated dirty data | |
2405 | in the first 3/4 of the curve yield relatively small differences in the | |
2406 | amount of delay. | |
2407 | .sp | |
2408 | The effects can be easier to understand when the amount of delay is | |
2409 | represented on a log scale: | |
2410 | .sp | |
2411 | .nf | |
2412 | delay | |
2413 | 100ms +-------------------------------------------------------------++ | |
2414 | + + | |
2415 | | | | |
2416 | + *+ | |
2417 | 10ms + *+ | |
2418 | + ** + | |
2419 | | (midpoint) ** | | |
2420 | + | ** + | |
2421 | 1ms + v **** + | |
2422 | + zfs_delay_scale ----------> ***** + | |
2423 | | **** | | |
2424 | + **** + | |
2425 | 100us + ** + | |
2426 | + * + | |
2427 | | * | | |
2428 | + * + | |
2429 | 10us + * + | |
2430 | + + | |
2431 | | | | |
2432 | + + | |
2433 | +--------------------------------------------------------------+ | |
2434 | 0% <- zfs_dirty_data_max -> 100% | |
2435 | .fi | |
2436 | .sp | |
2437 | Note here that only as the amount of dirty data approaches its limit does | |
2438 | the delay start to increase rapidly. The goal of a properly tuned system | |
2439 | should be to keep the amount of dirty data out of that range by first | |
2440 | ensuring that the appropriate limits are set for the I/O scheduler to reach | |
2441 | optimal throughput on the backend storage, and then by changing the value | |
2442 | of \fBzfs_delay_scale\fR to increase the steepness of the curve. |