[LU-13669] improve mmap performances

• Type: Improvement
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.14.0
• Assignee: Wang Shilong (Inactive)
• Reporter: Wang Shilong (Inactive)

Description

We have observed slow performance on mmap read on some applications.

The problem is if access pattern is neither sequential nor stride reading, but
still adjacent in a small range and then seek a random position.

So the pattern could be something like this:

[1M data] [hole..] [0.5M data] [hole]......[0.7M data]......[1M data]

So every time application want to access some data, the data is not only 4K but
cosed to let's say 1M range, so if we could predict this kind of behavior and next time we hit page miss, we use last time range for prefetching.

Issue Links

• is related to: LUDOC-485 document llite.*.read_ahead_range_kb parameter
• is related to: LU-11416 Improve readahead for random read of small/medium files

Comments

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/41228/
Subject: LU-13669 llite: make readahead aware of hints
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 7542820698696ed5853ded30c9bf7fd5a78f0937

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/38916/
Subject: LU-13669 llite: try to improve mmap performance
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 0c5ad4b6df5bf35b291842fc6d42c2720246a026

[Gerrit Updater]

Wang Shilong ([e-mail]) uploaded a new patch: https://review.whamcloud.com/39926
Subject: LU-13669 readahead: increase ra window progressively
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: fc72ad754bc8fcc86ac96661329d9b074e3703e8

[Wang Shilong (Inactive)]

Here is a benchmarking from Ihara:

master
[root@amd01 ~]#  echo 3 > /proc/sys/vm/drop_caches; fio --name=randread --directory=/ai400/fio --rw=randread --ioengine=mmap --bs=128K --numjobs=32 --filesize=200G --filename=randread --time_based --status-interval=10s --runtime=30s --allow_file_create=1 --group_reporting --disable_lat=1 --disable_clat=1 --disable_slat=1 --disk_util=0 --aux-path=/tmp --randrepeat=0 --unique_filename=0 --fallocate=0 
[root@amd01 ~]# lctl get_param llite.*.stats
llite.ai400-ffff8d855cf6a800.stats=
snapshot_time             1594299686.416386775 secs.nsecs
open                      49 samples [usec] 0 1401 22183
close                     49 samples [usec] 6 246 1117
mmap                      32 samples [usec] 7 17 381
page_fault                1867904 samples [usec] 0 37698 956414944
getattr                   34 samples [usec] 1 25 107
inode_permission          214 samples [usec] 0 22 407
master + patch
[root@amd01 ~]# echo 3 > /proc/sys/vm/drop_caches; fio --name=randread --directory=/ai400/fio --rw=randread --ioengine=mmap --bs=128K --numjobs=32 --filesize=200G --filename=randread --time_based --status-interval=10s --runtime=30s --allow_file_create=1 --group_reporting --disable_lat=1 --disable_clat=1 --disable_slat=1 --disk_util=0 --aux-path=/tmp --randrepeat=0 --unique_filename=0 --fallocate=0
[root@amd01 ~]# lctl get_param llite.*.stats
llite.ai400-ffff8d862cbc6000.stats=
snapshot_time             1594299943.816443708 secs.nsecs
open                      63 samples [usec] 1 4007 118809
close                     63 samples [usec] 4 250 1410
mmap                      32 samples [usec] 8 51 659
page_fault                17773312 samples [usec] 0 6543 933220595
getattr                   34 samples [usec] 2 224 440
inode_permission          256 samples [usec] 0 276 1292

tested first 30sec, mostly no cached mmap
512usec vs 52usec
10x latency reduction.

max latency 37698usec vs 6543usec
max latency is also very small

[Gerrit Updater]

Wang Shilong ([e-mail]) uploaded a new patch: https://review.whamcloud.com/38916
Subject: LU-13669 llite: try to improve mmap performance
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 84116d4451917e861671079ddf7ef57c722565c2