diff options
author | Niall Douglas (s [underscore] sourceforge {at} nedprod [dot] com) <spamtrap@nedprod.com> | 2017-08-18 01:38:05 +0300 |
---|---|---|
committer | Niall Douglas (s [underscore] sourceforge {at} nedprod [dot] com) <spamtrap@nedprod.com> | 2017-08-18 01:38:05 +0300 |
commit | 11193b589a8c54e12769a2cbb6c37b1777ec2613 (patch) | |
tree | 053865bd89825d133bfbc32e44d4aa575904a7ff | |
parent | 79b3600d1a89446d8b4f58a1cfcd7c237e77e18a (diff) |
New latency results with the fixed tool
-rw-r--r-- | programs/fs-probe/fs_probe_results.yaml | 667 | ||||
-rw-r--r-- | programs/fs-probe/latencies.xlsx | bin | 0 -> 9466 bytes | |||
-rw-r--r-- | release_notes.md | 8 |
3 files changed, 671 insertions, 4 deletions
diff --git a/programs/fs-probe/fs_probe_results.yaml b/programs/fs-probe/fs_probe_results.yaml index 37e37127..16d70da4 100644 --- a/programs/fs-probe/fs_probe_results.yaml +++ b/programs/fs-probe/fs_probe_results.yaml @@ -747,4 +747,671 @@ direct=1 sync=1: # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99.999% # of the time) 99.999%: 27010641 +--- +timestamp: 2017-08-17 19:00:15 +0000 +system: + os: + name: Linux + ver: 4.4.19-1-pve + cpu: + name: GenuineIntel Intel(R) Xeon(R) CPU E3-1230 v3 @ 3.30GHz + architecture: x86_64 + physical_cores: 8 + mem: + # Main memory bandwidth when accessed sequentially (1 CPU core) + max_bandwidth: 1046898278 + # Main memory bandwidth when 4Kb pages are accessed randomly (1 CPU core) + min_bandwidth: 1637456281 + quantity: 33696063488 + in_use: 0.958757 + timer: + ns_per_tick: 15 + ns_overhead: 17 +storage: + device: + min_io_size: 4096 + fs: + name: ext4 + config: todo + size: 368654848000 + in_use: 0.903985 +direct=0 sync=0: + concurrency: + # The i/o modify quantum guaranteed to be atomically visible to readers + # irrespective of rewrite quantity + atomic_rewrite_quantum: 1 + + # The maximum single aligned i/o modify quantity atomically visible to readers + # (can be [potentially unreliably] much larger than atomic_rewrite_quantum). A + # very common value on modern hardware with direct i/o thanks to PCIe DMA is + # 4096, don't trust values higher than this because of potentially discontiguous + # memory page mapping. + max_aligned_atomic_rewrite: 1 + + latency: + read: + qd1: + # The nanoseconds to read 4Kb at a queue depth of 1 (min) + min: 263 + # The nanoseconds to read 4Kb at a queue depth of 1 (arithmetic mean) + mean: 3128 + # The nanoseconds to read 4Kb at a queue depth of 1 (max) + max: 8732594 + # The nanoseconds to read 4Kb at a queue depth of 1 (50% of the time) + 50%: 701 + # The nanoseconds to read 4Kb at a queue depth of 1 (95% of the time) + 95%: 1019 + # The nanoseconds to read 4Kb at a queue depth of 1 (99% of the time) + 99%: 128253 + # The nanoseconds to read 4Kb at a queue depth of 1 (99.999% of the time) + 99.999%: 329352 + qd16: + # The nanoseconds to read 4Kb at a queue depth of 16 (min) + min: 296 + # The nanoseconds to read 4Kb at a queue depth of 16 (arithmetic mean) + mean: 3513 + # The nanoseconds to read 4Kb at a queue depth of 16 (max) + max: 55953395 + # The nanoseconds to read 4Kb at a queue depth of 16 (50% of the time) + 50%: 1602 + # The nanoseconds to read 4Kb at a queue depth of 16 (95% of the time) + 95%: 2095 + # The nanoseconds to read 4Kb at a queue depth of 16 (99% of the time) + 99%: 2523 + # The nanoseconds to read 4Kb at a queue depth of 16 (99.999% of the time) + 99.999%: 24012593 + write: + qd1: + # The nanoseconds to write 4Kb at a queue depth of 1 (min) + min: 490 + # The nanoseconds to write 4Kb at a queue depth of 1 (arithmetic mean) + mean: 1011 + # The nanoseconds to write 4Kb at a queue depth of 1 (max) + max: 270712217 + # The nanoseconds to write 4Kb at a queue depth of 1 (50% of the time) + 50%: 912 + # The nanoseconds to write 4Kb at a queue depth of 1 (95% of the time) + 95%: 1172 + # The nanoseconds to write 4Kb at a queue depth of 1 (99% of the time) + 99%: 1340 + # The nanoseconds to write 4Kb at a queue depth of 1 (99.999% of the time) + 99.999%: 20402 + qd16: + # The nanoseconds to write 4Kb at a queue depth of 16 (min) + min: 575 + # The nanoseconds to write 4Kb at a queue depth of 16 (arithmetic mean) + mean: 68915 + # The nanoseconds to write 4Kb at a queue depth of 16 (max) + max: 439389549 + # The nanoseconds to write 4Kb at a queue depth of 16 (50% of the time) + 50%: 1565 + # The nanoseconds to write 4Kb at a queue depth of 16 (95% of the time) + 95%: 2267 + # The nanoseconds to write 4Kb at a queue depth of 16 (99% of the time) + 99%: 2679 + # The nanoseconds to write 4Kb at a queue depth of 16 (99.999% of the time) + 99.999%: 142945583 + readwrite: + qd4: + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (min) + min: 266 + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 + # (arithmetic mean) + mean: 1031 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (max) + max: 259895949 + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (50% of + # the time) + 50%: 893 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (95% of + # the time) + 95%: 1369 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99% of + # the time) + 99%: 1738 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99.999% + # of the time) + 99.999%: 19824 + +direct=1 sync=0: + concurrency: + # The i/o modify quantum guaranteed to be atomically visible to readers + # irrespective of rewrite quantity + atomic_rewrite_quantum: 1048576 + + # The maximum single aligned i/o modify quantity atomically visible to readers + # (can be [potentially unreliably] much larger than atomic_rewrite_quantum). A + # very common value on modern hardware with direct i/o thanks to PCIe DMA is + # 4096, don't trust values higher than this because of potentially discontiguous + # memory page mapping. + max_aligned_atomic_rewrite: 1048576 + + latency: + read: + qd1: + # The nanoseconds to read 4Kb at a queue depth of 1 (min) + min: 61738 + # The nanoseconds to read 4Kb at a queue depth of 1 (arithmetic mean) + mean: 122085 + # The nanoseconds to read 4Kb at a queue depth of 1 (max) + max: 6293708 + # The nanoseconds to read 4Kb at a queue depth of 1 (50% of the time) + 50%: 123086 + # The nanoseconds to read 4Kb at a queue depth of 1 (95% of the time) + 95%: 131704 + # The nanoseconds to read 4Kb at a queue depth of 1 (99% of the time) + 99%: 157980 + # The nanoseconds to read 4Kb at a queue depth of 1 (99.999% of the time) + 99.999%: 6293708 + qd16: + # The nanoseconds to read 4Kb at a queue depth of 16 (min) + min: 113284 + # The nanoseconds to read 4Kb at a queue depth of 16 (arithmetic mean) + mean: 180199 + # The nanoseconds to read 4Kb at a queue depth of 16 (max) + max: 4141448 + # The nanoseconds to read 4Kb at a queue depth of 16 (50% of the time) + 50%: 163952 + # The nanoseconds to read 4Kb at a queue depth of 16 (95% of the time) + 95%: 270049 + # The nanoseconds to read 4Kb at a queue depth of 16 (99% of the time) + 99%: 377075 + # The nanoseconds to read 4Kb at a queue depth of 16 (99.999% of the time) + 99.999%: 4015285 + write: + qd1: + # The nanoseconds to write 4Kb at a queue depth of 1 (min) + min: 22200 + # The nanoseconds to write 4Kb at a queue depth of 1 (arithmetic mean) + mean: 32225 + # The nanoseconds to write 4Kb at a queue depth of 1 (max) + max: 82480959 + # The nanoseconds to write 4Kb at a queue depth of 1 (50% of the time) + 50%: 31447 + # The nanoseconds to write 4Kb at a queue depth of 1 (95% of the time) + 95%: 33064 + # The nanoseconds to write 4Kb at a queue depth of 1 (99% of the time) + 99%: 37890 + # The nanoseconds to write 4Kb at a queue depth of 1 (99.999% of the time) + 99.999%: 14437186 + qd16: + # The nanoseconds to write 4Kb at a queue depth of 16 (min) + min: 79400 + # The nanoseconds to write 4Kb at a queue depth of 16 (arithmetic mean) + mean: 197243 + # The nanoseconds to write 4Kb at a queue depth of 16 (max) + max: 74244118 + # The nanoseconds to write 4Kb at a queue depth of 16 (50% of the time) + 50%: 181580 + # The nanoseconds to write 4Kb at a queue depth of 16 (95% of the time) + 95%: 183902 + # The nanoseconds to write 4Kb at a queue depth of 16 (99% of the time) + 99%: 185247 + # The nanoseconds to write 4Kb at a queue depth of 16 (99.999% of the time) + 99.999%: 74229396 + readwrite: + qd4: + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (min) + min: 22382 + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 + # (arithmetic mean) + mean: 100774 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (max) + max: 75325324 + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (50% of + # the time) + 50%: 31812 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (95% of + # the time) + 95%: 488436 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99% of + # the time) + 99%: 1416093 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99.999% + # of the time) + 99.999%: 36661354 + +direct=0 sync=1: + concurrency: + # The i/o modify quantum guaranteed to be atomically visible to readers + # irrespective of rewrite quantity + atomic_rewrite_quantum: 1 + + # The maximum single aligned i/o modify quantity atomically visible to readers + # (can be [potentially unreliably] much larger than atomic_rewrite_quantum). A + # very common value on modern hardware with direct i/o thanks to PCIe DMA is + # 4096, don't trust values higher than this because of potentially discontiguous + # memory page mapping. + max_aligned_atomic_rewrite: 1 + + latency: + read: + qd1: + # The nanoseconds to read 4Kb at a queue depth of 1 (min) + min: 399 + # The nanoseconds to read 4Kb at a queue depth of 1 (arithmetic mean) + mean: 95953 + # The nanoseconds to read 4Kb at a queue depth of 1 (max) + max: 18161672 + # The nanoseconds to read 4Kb at a queue depth of 1 (50% of the time) + 50%: 121143 + # The nanoseconds to read 4Kb at a queue depth of 1 (95% of the time) + 95%: 152834 + # The nanoseconds to read 4Kb at a queue depth of 1 (99% of the time) + 99%: 222099 + # The nanoseconds to read 4Kb at a queue depth of 1 (99.999% of the time) + 99.999%: 16774083 + qd16: + # The nanoseconds to read 4Kb at a queue depth of 16 (min) + min: 286 + # The nanoseconds to read 4Kb at a queue depth of 16 (arithmetic mean) + mean: 43916 + # The nanoseconds to read 4Kb at a queue depth of 16 (max) + max: 40009989 + # The nanoseconds to read 4Kb at a queue depth of 16 (50% of the time) + 50%: 1595 + # The nanoseconds to read 4Kb at a queue depth of 16 (95% of the time) + 95%: 383963 + # The nanoseconds to read 4Kb at a queue depth of 16 (99% of the time) + 99%: 554586 + # The nanoseconds to read 4Kb at a queue depth of 16 (99.999% of the time) + 99.999%: 24011488 + write: + qd1: + # The nanoseconds to write 4Kb at a queue depth of 1 (min) + min: 5582633 + # The nanoseconds to write 4Kb at a queue depth of 1 (arithmetic mean) + mean: 8465934 + # The nanoseconds to write 4Kb at a queue depth of 1 (max) + max: 52623801 + # The nanoseconds to write 4Kb at a queue depth of 1 (50% of the time) + 50%: 7980693 + # The nanoseconds to write 4Kb at a queue depth of 1 (95% of the time) + 95%: 9869716 + # The nanoseconds to write 4Kb at a queue depth of 1 (99% of the time) + 99%: 33155731 + # The nanoseconds to write 4Kb at a queue depth of 1 (99.999% of the time) + 99.999%: 52623801 + qd16: + # The nanoseconds to write 4Kb at a queue depth of 16 (min) + min: 5886202 + # The nanoseconds to write 4Kb at a queue depth of 16 (arithmetic mean) + mean: 18568045 + # The nanoseconds to write 4Kb at a queue depth of 16 (max) + max: 113155915 + # The nanoseconds to write 4Kb at a queue depth of 16 (50% of the time) + 50%: 15798299 + # The nanoseconds to write 4Kb at a queue depth of 16 (95% of the time) + 95%: 50844549 + # The nanoseconds to write 4Kb at a queue depth of 16 (99% of the time) + 99%: 81729954 + # The nanoseconds to write 4Kb at a queue depth of 16 (99.999% of the time) + 99.999%: 113155915 + readwrite: + qd4: + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (min) + min: 264 + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 + # (arithmetic mean) + mean: 1178 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (max) + max: 51067492 + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (50% of + # the time) + 50%: 822 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (95% of + # the time) + 95%: 1102 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99% of + # the time) + 99%: 1295 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99.999% + # of the time) + 99.999%: 9242725 + +direct=1 sync=1: + concurrency: + # The i/o modify quantum guaranteed to be atomically visible to readers + # irrespective of rewrite quantity + atomic_rewrite_quantum: 1048576 + + # The maximum single aligned i/o modify quantity atomically visible to readers + # (can be [potentially unreliably] much larger than atomic_rewrite_quantum). A + # very common value on modern hardware with direct i/o thanks to PCIe DMA is + # 4096, don't trust values higher than this because of potentially discontiguous + # memory page mapping. + max_aligned_atomic_rewrite: 1048576 + + latency: + read: + qd1: + # The nanoseconds to read 4Kb at a queue depth of 1 (min) + min: 59893 + # The nanoseconds to read 4Kb at a queue depth of 1 (arithmetic mean) + mean: 115390 + # The nanoseconds to read 4Kb at a queue depth of 1 (max) + max: 777491 + # The nanoseconds to read 4Kb at a queue depth of 1 (50% of the time) + 50%: 120977 + # The nanoseconds to read 4Kb at a queue depth of 1 (95% of the time) + 95%: 132123 + # The nanoseconds to read 4Kb at a queue depth of 1 (99% of the time) + 99%: 172881 + # The nanoseconds to read 4Kb at a queue depth of 1 (99.999% of the time) + 99.999%: 777491 + qd16: + # The nanoseconds to read 4Kb at a queue depth of 16 (min) + min: 73887 + # The nanoseconds to read 4Kb at a queue depth of 16 (arithmetic mean) + mean: 170646 + # The nanoseconds to read 4Kb at a queue depth of 16 (max) + max: 9681719 + # The nanoseconds to read 4Kb at a queue depth of 16 (50% of the time) + 50%: 161299 + # The nanoseconds to read 4Kb at a queue depth of 16 (95% of the time) + 95%: 236432 + # The nanoseconds to read 4Kb at a queue depth of 16 (99% of the time) + 99%: 313559 + # The nanoseconds to read 4Kb at a queue depth of 16 (99.999% of the time) + 99.999%: 9555022 + write: + qd1: + # The nanoseconds to write 4Kb at a queue depth of 1 (min) + min: 5609401 + # The nanoseconds to write 4Kb at a queue depth of 1 (arithmetic mean) + mean: 8562943 + # The nanoseconds to write 4Kb at a queue depth of 1 (max) + max: 115307180 + # The nanoseconds to write 4Kb at a queue depth of 1 (50% of the time) + 50%: 8792387 + # The nanoseconds to write 4Kb at a queue depth of 1 (95% of the time) + 95%: 10124523 + # The nanoseconds to write 4Kb at a queue depth of 1 (99% of the time) + 99%: 32969389 + # The nanoseconds to write 4Kb at a queue depth of 1 (99.999% of the time) + 99.999%: 115307180 + qd16: + # The nanoseconds to write 4Kb at a queue depth of 16 (min) + min: 9421407 + # The nanoseconds to write 4Kb at a queue depth of 16 (arithmetic mean) + mean: 15979770 + # The nanoseconds to write 4Kb at a queue depth of 16 (max) + max: 172498199 + # The nanoseconds to write 4Kb at a queue depth of 16 (50% of the time) + 50%: 15586458 + # The nanoseconds to write 4Kb at a queue depth of 16 (95% of the time) + 95%: 18403763 + # The nanoseconds to write 4Kb at a queue depth of 16 (99% of the time) + 99%: 19257315 + # The nanoseconds to write 4Kb at a queue depth of 16 (99.999% of the time) + 99.999%: 172498199 + readwrite: + qd4: + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (min) + min: 66847 + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 + # (arithmetic mean) + mean: 3788865 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (max) + max: 74815380 + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (50% of + # the time) + 50%: 3900109 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (95% of + # the time) + 95%: 8695311 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99% of + # the time) + 99%: 10139375 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99.999% + # of the time) + 99.999%: 74815380 +--- +timestamp: 2017-08-17 19:34:30 +0000 +system: + os: + name: Linux + ver: 4.4.19-1-pve + cpu: + name: GenuineIntel Intel(R) Xeon(R) CPU E3-1230 v3 @ 3.30GHz + architecture: x86_64 + physical_cores: 8 + mem: + # Main memory bandwidth when accessed sequentially (1 CPU core) + max_bandwidth: 1073741824 + # Main memory bandwidth when 4Kb pages are accessed randomly (1 CPU core) + min_bandwidth: 1637456281 + quantity: 33696063488 + in_use: 0.415808 + timer: + ns_per_tick: 16 + ns_overhead: 17 +storage: + device: + min_io_size: 131072 + fs: + name: zfs + config: todo + size: 112312188928 + in_use: 0.312374 +direct=0 sync=0: + concurrency: + # The i/o modify quantum guaranteed to be atomically visible to readers + # irrespective of rewrite quantity + atomic_rewrite_quantum: 1048576 + + # The maximum single aligned i/o modify quantity atomically visible to readers + # (can be [potentially unreliably] much larger than atomic_rewrite_quantum). A + # very common value on modern hardware with direct i/o thanks to PCIe DMA is + # 4096, don't trust values higher than this because of potentially discontiguous + # memory page mapping. + max_aligned_atomic_rewrite: 1048576 + + latency: + read: + qd1: + # The nanoseconds to read 4Kb at a queue depth of 1 (min) + min: 1910 + # The nanoseconds to read 4Kb at a queue depth of 1 (arithmetic mean) + mean: 10839 + # The nanoseconds to read 4Kb at a queue depth of 1 (max) + max: 140898 + # The nanoseconds to read 4Kb at a queue depth of 1 (50% of the time) + 50%: 7661 + # The nanoseconds to read 4Kb at a queue depth of 1 (95% of the time) + 95%: 43156 + # The nanoseconds to read 4Kb at a queue depth of 1 (99% of the time) + 99%: 67452 + # The nanoseconds to read 4Kb at a queue depth of 1 (99.999% of the time) + 99.999%: 117609 + qd16: + # The nanoseconds to read 4Kb at a queue depth of 16 (min) + min: 9773 + # The nanoseconds to read 4Kb at a queue depth of 16 (arithmetic mean) + mean: 38659 + # The nanoseconds to read 4Kb at a queue depth of 16 (max) + max: 52024944 + # The nanoseconds to read 4Kb at a queue depth of 16 (50% of the time) + 50%: 18597 + # The nanoseconds to read 4Kb at a queue depth of 16 (95% of the time) + 95%: 20734 + # The nanoseconds to read 4Kb at a queue depth of 16 (99% of the time) + 99%: 52481 + # The nanoseconds to read 4Kb at a queue depth of 16 (99.999% of the time) + 99.999%: 36022797 + write: + qd1: + # The nanoseconds to write 4Kb at a queue depth of 1 (min) + min: 4528 + # The nanoseconds to write 4Kb at a queue depth of 1 (arithmetic mean) + mean: 21609 + # The nanoseconds to write 4Kb at a queue depth of 1 (max) + max: 12804173 + # The nanoseconds to write 4Kb at a queue depth of 1 (50% of the time) + 50%: 18726 + # The nanoseconds to write 4Kb at a queue depth of 1 (95% of the time) + 95%: 36876 + # The nanoseconds to write 4Kb at a queue depth of 1 (99% of the time) + 99%: 55380 + # The nanoseconds to write 4Kb at a queue depth of 1 (99.999% of the time) + 99.999%: 6876852 + qd16: + # The nanoseconds to write 4Kb at a queue depth of 16 (min) + min: 5410 + # The nanoseconds to write 4Kb at a queue depth of 16 (arithmetic mean) + mean: 714772 + # The nanoseconds to write 4Kb at a queue depth of 16 (max) + max: 1042809986 + # The nanoseconds to write 4Kb at a queue depth of 16 (50% of the time) + 50%: 38032 + # The nanoseconds to write 4Kb at a queue depth of 16 (95% of the time) + 95%: 1587465 + # The nanoseconds to write 4Kb at a queue depth of 16 (99% of the time) + 99%: 4511125 + # The nanoseconds to write 4Kb at a queue depth of 16 (99.999% of the time) + 99.999%: 1035504567 + readwrite: + qd4: + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (min) + min: 1708 + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 + # (arithmetic mean) + mean: 12766 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (max) + max: 383277956 + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (50% of + # the time) + 50%: 4634 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (95% of + # the time) + 95%: 59314 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99% of + # the time) + 99%: 98357 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99.999% + # of the time) + 99.999%: 16425840 + +direct=0 sync=1: + concurrency: + # The i/o modify quantum guaranteed to be atomically visible to readers + # irrespective of rewrite quantity + atomic_rewrite_quantum: 1048576 + + # The maximum single aligned i/o modify quantity atomically visible to readers + # (can be [potentially unreliably] much larger than atomic_rewrite_quantum). A + # very common value on modern hardware with direct i/o thanks to PCIe DMA is + # 4096, don't trust values higher than this because of potentially discontiguous + # memory page mapping. + max_aligned_atomic_rewrite: 1048576 + + latency: + read: + qd1: + # The nanoseconds to read 4Kb at a queue depth of 1 (min) + min: 2016 + # The nanoseconds to read 4Kb at a queue depth of 1 (arithmetic mean) + mean: 16474 + # The nanoseconds to read 4Kb at a queue depth of 1 (max) + max: 2550811 + # The nanoseconds to read 4Kb at a queue depth of 1 (50% of the time) + 50%: 7994 + # The nanoseconds to read 4Kb at a queue depth of 1 (95% of the time) + 95%: 82363 + # The nanoseconds to read 4Kb at a queue depth of 1 (99% of the time) + 99%: 83649 + # The nanoseconds to read 4Kb at a queue depth of 1 (99.999% of the time) + 99.999%: 151481 + qd16: + # The nanoseconds to read 4Kb at a queue depth of 16 (min) + min: 3049 + # The nanoseconds to read 4Kb at a queue depth of 16 (arithmetic mean) + mean: 23651 + # The nanoseconds to read 4Kb at a queue depth of 16 (max) + max: 17621139 + # The nanoseconds to read 4Kb at a queue depth of 16 (50% of the time) + 50%: 10949 + # The nanoseconds to read 4Kb at a queue depth of 16 (95% of the time) + 95%: 88773 + # The nanoseconds to read 4Kb at a queue depth of 16 (99% of the time) + 99%: 162771 + # The nanoseconds to read 4Kb at a queue depth of 16 (99.999% of the time) + 99.999%: 7522623 + write: + qd1: + # The nanoseconds to write 4Kb at a queue depth of 1 (min) + min: 1972303 + # The nanoseconds to write 4Kb at a queue depth of 1 (arithmetic mean) + mean: 6658579 + # The nanoseconds to write 4Kb at a queue depth of 1 (max) + max: 17094746 + # The nanoseconds to write 4Kb at a queue depth of 1 (50% of the time) + 50%: 6907000 + # The nanoseconds to write 4Kb at a queue depth of 1 (95% of the time) + 95%: 9404125 + # The nanoseconds to write 4Kb at a queue depth of 1 (99% of the time) + 99%: 10868124 + # The nanoseconds to write 4Kb at a queue depth of 1 (99.999% of the time) + 99.999%: 17094746 + qd16: + # The nanoseconds to write 4Kb at a queue depth of 16 (min) + min: 2866524 + # The nanoseconds to write 4Kb at a queue depth of 16 (arithmetic mean) + mean: 23800870 + # The nanoseconds to write 4Kb at a queue depth of 16 (max) + max: 42342138 + # The nanoseconds to write 4Kb at a queue depth of 16 (50% of the time) + 50%: 24274413 + # The nanoseconds to write 4Kb at a queue depth of 16 (95% of the time) + 95%: 28867300 + # The nanoseconds to write 4Kb at a queue depth of 16 (99% of the time) + 99%: 32000489 + # The nanoseconds to write 4Kb at a queue depth of 16 (99.999% of the time) + 99.999%: 42342138 + readwrite: + qd4: + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (min) + min: 2788 + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 + # (arithmetic mean) + mean: 1708983 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (max) + max: 21646924 + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (50% of + # the time) + 50%: 5467 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (95% of + # the time) + 95%: 9042384 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99% of + # the time) + 99%: 11145248 + + # The nanoseconds to 75% read 25% write 4Kb at a total queue depth of 4 (99.999% + # of the time) + 99.999%: 21646924 + diff --git a/programs/fs-probe/latencies.xlsx b/programs/fs-probe/latencies.xlsx Binary files differnew file mode 100644 index 00000000..0b23f91c --- /dev/null +++ b/programs/fs-probe/latencies.xlsx diff --git a/release_notes.md b/release_notes.md index bdb27516..ad822040 100644 --- a/release_notes.md +++ b/release_notes.md @@ -35,10 +35,10 @@ Manufacturer claimed 4Kb transfer latencies for the physical hardware: - 99.999% spinning rust hard drive latency: **TODO** - Average SATA flash drive latency: **TODO** - 99.999% SATA flash drive latency: **TODO** -- Average NVMe flash drive latency: **351us** (2,849 IOPS) +- Average NVMe flash drive latency: **362us** (2,762 IOPS) - 99.999% NVMe flash drive latency: **35,378us** (28 IOPS) -Lowest sustained 4Kb read latency benchmarked to date by author (page cache): **2us** (500,000 IOPS, approx 1.9Gb/sec) +Lowest sustained 4Kb read latency benchmarked to date by author (page cache): **7us** (142k IOPS @ QD1, approx 558Mb/sec) </td> <td valign="top" width="33%"> 75% read 25% write QD4 4Kb transfer latencies for the software with AFIO: @@ -46,10 +46,10 @@ Lowest sustained 4Kb read latency benchmarked to date by author (page cache): ** - 99.999% spinning rust hard drive latency: **TODO** - Average SATA flash drive latency: **TODO** - 99.999% SATA flash drive latency: **TODO** -- Average NVMe flash drive latency: **221us** (4,525 IOPS) +- Average NVMe flash drive latency: **232us** (4,310 IOPS) - 99.999% NVMe flash drive latency: **11,952us** (83 IOPS) -Lowest sustained 4Kb write latency benchmarked to date by author (page cache): **2us** (500,000 IOPS, approx 1.9Gb/sec) +Lowest sustained 4Kb write latency benchmarked to date by author (page cache): **7us** (142k IOPS @ QD1, approx 558Mb/sec) </td> </tr> </table> |