Yu Zhao yuzhao@google.com 于2023年12月14日周四 11:09写道：
On Wed, Dec 13, 2023 at 12:59:14AM -0700, Yu Zhao wrote:
On Tue, Dec 12, 2023 at 8:03 PM Kairui Song ryncsn@gmail.com wrote:
Kairui Song ryncsn@gmail.com 于2023年12月12日周二 14:52写道：
Yu Zhao yuzhao@google.com 于2023年12月12日周二 06:07写道：
On Fri, Dec 8, 2023 at 1:24 AM Kairui Song ryncsn@gmail.com wrote:
Yu Zhao yuzhao@google.com 于2023年12月8日周五 14:14写道： > > Unmapped folios accessed through file descriptors can be > underprotected. Those folios are added to the oldest generation based > on: > 1. The fact that they are less costly to reclaim (no need to walk the > rmap and flush the TLB) and have less impact on performance (don't > cause major PFs and can be non-blocking if needed again). > 2. The observation that they are likely to be single-use. E.g., for > client use cases like Android, its apps parse configuration files > and store the data in heap (anon); for server use cases like MySQL, > it reads from InnoDB files and holds the cached data for tables in > buffer pools (anon). > > However, the oldest generation can be very short lived, and if so, it > doesn't provide the PID controller with enough time to respond to a > surge of refaults. (Note that the PID controller uses weighted > refaults and those from evicted generations only take a half of the > whole weight.) In other words, for a short lived generation, the > moving average smooths out the spike quickly. > > To fix the problem: > 1. For folios that are already on LRU, if they can be beyond the > tracking range of tiers, i.e., five accesses through file > descriptors, move them to the second oldest generation to give them > more time to age. (Note that tiers are used by the PID controller > to statistically determine whether folios accessed multiple times > through file descriptors are worth protecting.) > 2. When adding unmapped folios to LRU, adjust the placement of them so > that they are not too close to the tail. The effect of this is > similar to the above. > > On Android, launching 55 apps sequentially: > Before After Change > workingset_refault_anon 25641024 25598972 0% > workingset_refault_file 115016834 106178438 -8%
Hi Yu,
Thanks you for your amazing works on MGLRU.
I believe this is the similar issue I was trying to resolve previously: https://lwn.net/Articles/945266/ The idea is to use refault distance to decide if the page should be place in oldest generation or some other gen, which per my test, worked very well, and we have been using refault distance for MGLRU in multiple workloads.
There are a few issues left in my previous RFC series, like anon pages in MGLRU shouldn't be considered, I wanted to collect feedback or test cases, but unfortunately it seems didn't get too much attention upstream.
I think both this patch and my previous series are for solving the file pages underpertected issue, and I did a quick test using this series, for mongodb test, refault distance seems still a better solution (I'm not saying these two optimization are mutually exclusive though, just they do have some conflicts in implementation and solving similar problem):
Previous result:
Execution Results after 905 seconds
Executed Time (µs) RateSTOCK_LEVEL 2542 27121571486.2 0.09 txn/s
TOTAL 2542 27121571486.2 0.09 txn/s
This patch:
Execution Results after 900 seconds
Executed Time (µs) RateSTOCK_LEVEL 1594 27061522574.4 0.06 txn/s
TOTAL 1594 27061522574.4 0.06 txn/s
Unpatched version is always around ~500.
Thanks for the test results!
I think there are a few points here:
- Refault distance make use of page shadow so it can better
distinguish evicted pages of different access pattern (re-access distance).
- Throttled refault distance can help hold part of workingset when
memory is too small to hold the whole workingset.
So maybe part of this patch and the bits of previous series can be combined to work better on this issue, how do you think?
I'll try to find some time this week to look at your RFC. It'd be a
Hi Yu,
I'm working on V4 of the RFC now, which just update some comments, and skip anon page re-activation in refault path for mglru which was not very helpful, only some tiny adjustment. And I found it easier to test with fio, using following test script:
#!/bin/bash swapoff -a
modprobe brd rd_nr=1 rd_size=16777216 mkfs.ext4 /dev/ram0 mount /dev/ram0 /mnt
mkdir -p /sys/fs/cgroup/benchmark cd /sys/fs/cgroup/benchmark
echo 4G > memory.max echo $$ > cgroup.procs echo 3 > /proc/sys/vm/drop_caches
fio -name=mglru --numjobs=12 --directory=/mnt --size=1024m \ --buffered=1 --ioengine=io_uring --iodepth=128 \ --iodepth_batch_submit=32 --iodepth_batch_complete=32 \ --rw=randread --random_distribution=zipf:0.5 --norandommap \ --time_based --ramp_time=5m --runtime=5m --group_reporting
zipf:0.5 is used here to simulate a cached read with slight bias towards certain pages. Unpatched 6.7-rc4: Run status group 0 (all jobs): READ: bw=6548MiB/s (6866MB/s), 6548MiB/s-6548MiB/s (6866MB/s-6866MB/s), io=1918GiB (2060GB), run=300001-300001msec
Patched with RFC v4: Run status group 0 (all jobs): READ: bw=7270MiB/s (7623MB/s), 7270MiB/s-7270MiB/s (7623MB/s-7623MB/s), io=2130GiB (2287GB), run=300001-300001msec
Patched with this series: Run status group 0 (all jobs): READ: bw=7098MiB/s (7442MB/s), 7098MiB/s-7098MiB/s (7442MB/s-7442MB/s), io=2079GiB (2233GB), run=300002-300002msec
MGLRU off: Run status group 0 (all jobs): READ: bw=6525MiB/s (6842MB/s), 6525MiB/s-6525MiB/s (6842MB/s-6842MB/s), io=1912GiB (2052GB), run=300002-300002msec
- If I change zipf:0.5 to random:
Unpatched 6.7-rc4: Patched with this series: Run status group 0 (all jobs): READ: bw=5975MiB/s (6265MB/s), 5975MiB/s-5975MiB/s (6265MB/s-6265MB/s), io=1750GiB (1879GB), run=300002-300002msec
Patched with RFC v4: Run status group 0 (all jobs): READ: bw=5987MiB/s (6278MB/s), 5987MiB/s-5987MiB/s (6278MB/s-6278MB/s), io=1754GiB (1883GB), run=300001-300001msec
Patched with this series: Run status group 0 (all jobs): READ: bw=5839MiB/s (6123MB/s), 5839MiB/s-5839MiB/s (6123MB/s-6123MB/s), io=1711GiB (1837GB), run=300001-300001msec
MGLRU off: Run status group 0 (all jobs): READ: bw=5689MiB/s (5965MB/s), 5689MiB/s-5689MiB/s (5965MB/s-5965MB/s), io=1667GiB (1790GB), run=300003-300003msec
fio uses ramdisk so LRU accuracy will have smaller impact. The Mongodb test I provided before uses a SATA SSD so it will have a much higher impact. I'll provides a script to setup the test case and run it, it's more complex to setup than fio since involving setting up multiple replicas and auth and hundreds of GB of test fixtures, I'm currently occupied by some other tasks but will try best to send them out as soon as possible.
Thanks! Apparently your RFC did show better IOPS with both access patterns, which was a surprise to me because it had higher refaults and usually higher refautls result in worse performance.
So I'm still trying to figure out why it turned out the opposite. My current guess is that:
- It had a very small but stable inactive LRU list, which was able to
fit into the L3 cache entirely. 2. It counted few folios as workingset and therefore incurred less overhead from CONFIG_PSI and/or CONFIG_TASK_DELAY_ACCT.
Did you save workingset_refault_file when you ran the test? If so, can you check the difference between this series and your RFC?
It seems I was right about #1 above. After I scaled your test up by 20x, I saw my series performed ~5% faster with zipf and ~9% faster with random accesses.
Hi Yu,
Thank you so much for testing and sharing this result.
I'm not sure about #1, the ramdisk size, access data, are far larger than L3 (16M on my CPU) even in down scaled test, and both random/zipf shows similar result.
IOW, I made rd_size from 16GB to 320GB, memory.max from 4GB to 80GB, --numjobs from 12 to 60 and --size from 1GB to 4GB.
v6.7-c5 + this series
zipf
mglru: (groupid=0, jobs=60): err= 0: pid=12155: Wed Dec 13 17:50:36 2023 read: IOPS=5074k, BW=19.4GiB/s (20.8GB/s)(5807GiB/300007msec) slat (usec): min=36, max=109326, avg=363.67, stdev=1829.97 clat (nsec): min=783, max=113292k, avg=1136755.10, stdev=3162056.05 lat (usec): min=37, max=149232, avg=1500.43, stdev=3644.21 clat percentiles (usec): | 1.00th=[ 490], 5.00th=[ 519], 10.00th=[ 537], 20.00th=[ 553], | 30.00th=[ 570], 40.00th=[ 586], 50.00th=[ 627], 60.00th=[ 840], | 70.00th=[ 988], 80.00th=[ 1074], 90.00th=[ 1188], 95.00th=[ 1336], | 99.00th=[ 7308], 99.50th=[31327], 99.90th=[36963], 99.95th=[45351], | 99.99th=[53216] bw ( MiB/s): min= 8332, max=27116, per=100.00%, avg=19846.67, stdev=58.20, samples=35903 iops : min=2133165, max=6941826, avg=5080741.79, stdev=14899.13, samples=35903 lat (nsec) : 1000=0.01% lat (usec) : 4=0.01%, 10=0.01%, 20=0.01%, 50=0.01%, 100=0.01% lat (usec) : 250=0.01%, 500=1.76%, 750=52.94%, 1000=16.65% lat (msec) : 2=26.22%, 4=0.15%, 10=1.36%, 20=0.01%, 50=0.90% lat (msec) : 100=0.02%, 250=0.01% cpu : usr=5.42%, sys=87.59%, ctx=470315, majf=0, minf=2184 IO depths : 1=0.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=100.0% submit : 0=0.0%, 4=0.0%, 8=0.0%, 16=0.1%, 32=100.0%, 64=0.0%, >=64=0.0% complete : 0=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=100.0%, 64=0.0%, >=64=0.1% issued rwts: total=1522384845,0,0,0 short=0,0,0,0 dropped=0,0,0,0 latency : target=0, window=0, percentile=100.00%, depth=128
Run status group 0 (all jobs): READ: bw=19.4GiB/s (20.8GB/s), 19.4GiB/s-19.4GiB/s (20.8GB/s-20.8GB/s), io=5807GiB (6236GB), run=300007-300007msec
Disk stats (read/write): ram0: ios=0/0, sectors=0/0, merge=0/0, ticks=0/0, in_queue=0, util=0.00% mglru: (g=0): rw=randread, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=io_uring, iodepth=128
random
mglru: (groupid=0, jobs=60): err= 0: pid=12576: Wed Dec 13 18:00:50 2023 read: IOPS=3853k, BW=14.7GiB/s (15.8GB/s)(4410GiB/300014msec) slat (usec): min=58, max=118605, avg=486.45, stdev=2311.45 clat (usec): min=3, max=169810, avg=1496.60, stdev=3982.89 lat (usec): min=73, max=170019, avg=1983.06, stdev=4585.87 clat percentiles (usec): | 1.00th=[ 586], 5.00th=[ 627], 10.00th=[ 644], 20.00th=[ 668], | 30.00th=[ 693], 40.00th=[ 725], 50.00th=[ 816], 60.00th=[ 1123], | 70.00th=[ 1221], 80.00th=[ 1352], 90.00th=[ 1516], 95.00th=[ 1713], | 99.00th=[31851], 99.50th=[34866], 99.90th=[41681], 99.95th=[54264], | 99.99th=[61080] bw ( MiB/s): min= 6049, max=21328, per=100.00%, avg=15070.00, stdev=45.96, samples=35940 iops : min=1548543, max=5459997, avg=3857912.87, stdev=11765.30, samples=35940 lat (usec) : 4=0.01%, 10=0.01%, 20=0.01%, 100=0.01%, 250=0.01% lat (usec) : 500=0.01%, 750=44.64%, 1000=8.20% lat (msec) : 2=43.84%, 4=0.27%, 10=1.79%, 20=0.01%, 50=1.20% lat (msec) : 100=0.07%, 250=0.01% cpu : usr=3.19%, sys=89.87%, ctx=463840, majf=0, minf=2248 IO depths : 1=0.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=100.0% submit : 0=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=100.0%, 64=0.0%, >=64=0.0% complete : 0=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=100.0%, 64=0.0%, >=64=0.1% issued rwts: total=1155923744,0,0,0 short=0,0,0,0 dropped=0,0,0,0 latency : target=0, window=0, percentile=100.00%, depth=128
Run status group 0 (all jobs): READ: bw=14.7GiB/s (15.8GB/s), 14.7GiB/s-14.7GiB/s (15.8GB/s-15.8GB/s), io=4410GiB (4735GB), run=300014-300014msec
Disk stats (read/write): ram0: ios=0/0, sectors=0/0, merge=0/0, ticks=0/0, in_queue=0, util=0.00%
memcg 3 /zipf node 0 0 1521654 0 0x 0 0r 0e 0p 0 0 0 1 0r 0e 0p 0 0 0 2 0r 0e 0p 0 0 0 3 0r 0e 0p 0 0 0 0 0 0 0 0 0 1 1521654 0 21 0 0 0 0 1077016797r 1111542014e 0p 1 0 0 0 317997853r 324814007e 0p 2 0 0 0 68064253r 68866308e 124302p 3 0 0 0 0r 0e 12282816p 0 0 0 0 0 0 2 1521654 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 3 1521654 0 0 0 0R 0T 0 0R 0T 0 1 0R 0T 0 0R 0T 0 2 0R 0T 0 0R 0T 0 3 0R 0T 0 0R 0T 0 0L 0O 0Y 0N 0F 0A node 1 0 1521654 0 0 0 0r 0e 0p 0r 0e 0p 1 0r 0e 0p 0r 0e 0p 2 0r 0e 0p 0r 0e 0p 3 0r 0e 0p 0r 0e 0p 0 0 0 0 0 0 1 1521654 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 2 1521654 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 3 1521654 0 0 0 0R 0T 0 0R 0T 0 1 0R 0T 0 0R 0T 0 2 0R 0T 0 0R 0T 0 3 0R 0T 0 0R 0T 0 0L 0O 0Y 0N 0F 0A memcg 4 /random node 0 0 600431 0 0x 0 0r 0e 0p 0 0 0 1 0r 0e 0p 0 0 0 2 0r 0e 0p 0 0 0 3 0r 0e 0p 0 0 0 0 0 0 0 0 0 1 600431 0 11169201 0 0 0 0 1071724785r 1103937007e 0p 1 0 0 0 376193810r 384852629e 0p 2 0 0 0 77315518r 78596395e 0p 3 0 0 0 0r 0e 9593442p 0 0 0 0 0 0 2 600431 1 9593442 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 3 600431 36 754 0 0R 0T 0 0R 0T 0 1 0R 0T 0 0R 0T 0 2 0R 0T 0 0R 0T 0 3 0R 0T 0 0R 0T 0 0L 0O 0Y 0N 0F 0A node 1 0 600431 0 0 0 0r 0e 0p 0r 0e 0p 1 0r 0e 0p 0r 0e 0p 2 0r 0e 0p 0r 0e 0p 3 0r 0e 0p 0r 0e 0p 0 0 0 0 0 0 1 600431 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 2 600431 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 3 600431 0 0 0 0R 0T 0 0R 0T 0 1 0R 0T 0 0R 0T 0 2 0R 0T 0 0R 0T 0 3 0R 0T 0 0R 0T 0 0L 0O 0Y 0N 0F 0A
v6.7-c5 + RFC v3
zipf
mglru: (groupid=0, jobs=60): err= 0: pid=11600: Wed Dec 13 18:34:31 2023 read: IOPS=4816k, BW=18.4GiB/s (19.7GB/s)(5512GiB/300014msec) slat (usec): min=3, max=121722, avg=384.46, stdev=2066.10 clat (nsec): min=356, max=174717k, avg=1197513.60, stdev=3568734.58 lat (usec): min=3, max=174919, avg=1581.97, stdev=4112.49 clat percentiles (usec): | 1.00th=[ 486], 5.00th=[ 515], 10.00th=[ 529], 20.00th=[ 553], | 30.00th=[ 570], 40.00th=[ 594], 50.00th=[ 652], 60.00th=[ 898], | 70.00th=[ 988], 80.00th=[ 1139], 90.00th=[ 1254], 95.00th=[ 1369], | 99.00th=[ 6915], 99.50th=[35914], 99.90th=[42206], 99.95th=[52167], | 99.99th=[61604] bw ( MiB/s): min= 7716, max=26325, per=100.00%, avg=18836.65, stdev=57.20, samples=35880 iops : min=1975306, max=6739280, avg=4822176.85, stdev=14642.35, samples=35880 lat (nsec) : 500=0.01%, 750=0.01%, 1000=0.01% lat (usec) : 4=0.01%, 10=0.01%, 20=0.01%, 100=0.01%, 250=0.01% lat (usec) : 500=2.57%, 750=50.99%, 1000=17.56% lat (msec) : 2=26.41%, 4=0.16%, 10=1.41%, 20=0.01%, 50=0.84% lat (msec) : 100=0.05%, 250=0.01% cpu : usr=4.95%, sys=88.09%, ctx=457609, majf=0, minf=2184 IO depths : 1=0.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=100.0% submit : 0=0.0%, 4=0.1%, 8=0.0%, 16=0.0%, 32=100.0%, 64=0.0%, >=64=0.0% complete : 0=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=100.0%, 64=0.0%, >=64=0.1% issued rwts: total=1445015808,0,0,0 short=0,0,0,0 dropped=0,0,0,0 latency : target=0, window=0, percentile=100.00%, depth=128
Run status group 0 (all jobs): READ: bw=18.4GiB/s (19.7GB/s), 18.4GiB/s-18.4GiB/s (19.7GB/s-19.7GB/s), io=5512GiB (5919GB), run=300014-300014msec
Disk stats (read/write): ram0: ios=0/0, sectors=0/0, merge=0/0, ticks=0/0, in_queue=0, util=0.00% mglru: (g=0): rw=randread, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=io_uring, iodepth=128
random
mglru: (groupid=0, jobs=60): err= 0: pid=12024: Wed Dec 13 18:44:45 2023 read: IOPS=3519k, BW=13.4GiB/s (14.4GB/s)(4027GiB/300011msec) slat (usec): min=54, max=136278, avg=534.57, stdev=2738.72 clat (usec): min=3, max=176186, avg=1638.66, stdev=4714.55 lat (usec): min=78, max=176426, avg=2173.23, stdev=5426.40 clat percentiles (usec): | 1.00th=[ 627], 5.00th=[ 676], 10.00th=[ 693], 20.00th=[ 725], | 30.00th=[ 766], 40.00th=[ 816], 50.00th=[ 1090], 60.00th=[ 1205], | 70.00th=[ 1270], 80.00th=[ 1369], 90.00th=[ 1500], 95.00th=[ 1614], | 99.00th=[38536], 99.50th=[41681], 99.90th=[47973], 99.95th=[65799], | 99.99th=[72877] bw ( MiB/s): min= 5586, max=20476, per=100.00%, avg=13760.26, stdev=45.33, samples=35904 iops : min=1430070, max=5242110, avg=3522621.15, stdev=11604.46, samples=35904 lat (usec) : 4=0.01%, 10=0.01%, 20=0.01%, 100=0.01%, 250=0.01% lat (usec) : 500=0.01%, 750=26.33%, 1000=21.81% lat (msec) : 2=48.54%, 4=0.16%, 10=1.91%, 20=0.01%, 50=1.17% lat (msec) : 100=0.09%, 250=0.01% cpu : usr=2.74%, sys=90.35%, ctx=481356, majf=0, minf=2244 IO depths : 1=0.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=100.0% submit : 0=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=100.0%, 64=0.0%, >=64=0.0% complete : 0=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=100.0%, 64=0.0%, >=64=0.1% issued rwts: total=1055590880,0,0,0 short=0,0,0,0 dropped=0,0,0,0 latency : target=0, window=0, percentile=100.00%, depth=128
Run status group 0 (all jobs): READ: bw=13.4GiB/s (14.4GB/s), 13.4GiB/s-13.4GiB/s (14.4GB/s-14.4GB/s), io=4027GiB (4324GB), run=300011-300011msec
Disk stats (read/write): ram0: ios=0/0, sectors=0/0, merge=0/0, ticks=0/0, in_queue=0, util=0.00%
memcg 3 /zipf node 0 0 1522519 0 22 0 0r 0e 0p 996363383r 1092111170e 0p 1 0r 0e 0p 274581982r 235766575e 0p 2 0r 0e 0p 85176438r 71356676e 96114p 3 0r 0e 0p 12470364r 11510461e 221796p 0 0 0 0 0 0 1 1522519 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 2 1522519 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 3 1522519 0 0 0 0R 0T 0 0R 0T 0 1 0R 0T 0 0R 0T 0 2 0R 0T 0 0R 0T 0 3 0R 0T 0 0R 0T 0 0L 0O 0Y 0N 0F 0A node 1 0 1522519 0 0 0 0r 0e 0p 0r 0e 0p 1 0r 0e 0p 0r 0e 0p 2 0r 0e 0p 0r 0e 0p 3 0r 0e 0p 0r 0e 0p 0 0 0 0 0 0 1 1522519 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 2 1522519 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 3 1522519 0 0 0 0R 0T 0 0R 0T 0 1 0R 0T 0 0R 0T 0 2 0R 0T 0 0R 0T 0 3 0R 0T 0 0R 0T 0 0L 0O 0Y 0N 0F 0A memcg 4 /random node 0 0 600413 0 2289676 0 0r 0e 0p 875605725r 960492874e 0p 1 0r 0e 0p 411230731r 383704269e 0p 2 0r 0e 0p 112639317r 97774351e 0p 3 0r 0e 0p 2103334r 1766407e 0p 0 0 0 0 0 0 1 600413 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 2 600413 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 3 600413 35 18466878 0 0R 0T 0 0R 0T 0 1 0R 0T 0 0R 0T 0 2 0R 0T 0 0R 0T 0 3 0R 0T 0 0R 0T 0 0L 0O 0Y 0N 0F 0A node 1 0 600413 0 0 0 0r 0e 0p 0r 0e 0p 1 0r 0e 0p 0r 0e 0p 2 0r 0e 0p 0r 0e 0p 3 0r 0e 0p 0r 0e 0p 0 0 0 0 0 0 1 600413 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 2 600413 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 3 600413 0 0 0 0R 0T 0 0R 0T 0 1 0R 0T 0 0R 0T 0 2 0R 0T 0 0R 0T 0 3 0R 0T 0 0R 0T 0 0L 0O 0Y 0N 0F 0A
And I reran the scaled down zipf test again:
RFC: Jobs: 12 (f=12): [r(12)][100.0%][r=7267MiB/s][r=1860k IOPS][eta 00m:00s]7s]s] mglru: (groupid=0, jobs=12): err= 0: pid=5159: Thu Dec 14 23:57:01 2023 read: IOPS=1862k, BW=7274MiB/s (7628MB/s)(2131GiB/300001msec) slat (usec): min=60, max=4711, avg=195.05, stdev=138.41 clat (usec): min=2, max=5097, avg=619.70, stdev=215.90 lat (usec): min=112, max=5271, avg=814.78, stdev=237.75 clat percentiles (usec): | 1.00th=[ 388], 5.00th=[ 408], 10.00th=[ 424], 20.00th=[ 457], | 30.00th=[ 482], 40.00th=[ 502], 50.00th=[ 523], 60.00th=[ 545], | 70.00th=[ 603], 80.00th=[ 889], 90.00th=[ 988], 95.00th=[ 1037], | 99.00th=[ 1106], 99.50th=[ 1139], 99.90th=[ 1237], 99.95th=[ 1369], | 99.99th=[ 1483] bw ( MiB/s): min= 6526, max= 8474, per=100.00%, avg=7284.26, stdev=48.62, samples=7176 iops : min=1670753, max=2169575, avg=1864770.39, stdev=12446.01, samples=7176 lat (usec) : 4=0.01%, 10=0.01%, 250=0.01%, 500=38.35%, 750=33.88% lat (usec) : 1000=19.46% lat (msec) : 2=8.30%, 4=0.01%, 10=0.01% cpu : usr=8.62%, sys=91.24%, ctx=531703, majf=0, minf=700 IO depths : 1=0.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=100.0% submit : 0=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=100.0%, 64=0.0%, >=64=0.0% complete : 0=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=100.0%, 64=0.0%, >=64=0.1% issued rwts: total=558664800,0,0,0 short=0,0,0,0 dropped=0,0,0,0 latency : target=0, window=0, percentile=100.00%, depth=128
Run status group 0 (all jobs): READ: bw=7274MiB/s (7628MB/s), 7274MiB/s-7274MiB/s (7628MB/s-7628MB/s), io=2131GiB (2288GB), run=300001-300001msec
workingset_refault_file 628192729
memcg 73 /benchmark node 0 0 1092186 0 0x 0 0r 0e 0p 0 0 0· 1 0r 0e 0p 0 0 0· 2 0r 0e 0p 0 0 0· 3 0r 0e 0p 0 0 0· 0 0 0 0 0 0· 1 1092186 0 4283· 0 0 0 0 507816078r 511714221e 0p 1 0 0 0 4682206r 3201136e 0p 2 0 0 0 64762r 43587e 0p 3 0 0 0 0r 0e 0p 0 0 0 0 0 0· 2 1092186 0 0· 0 0 0 0 0 0 0· 1 0 0 0 0 0 0· 2 0 0 0 0 0 0· 3 0 0 0 0 0 0· 0 0 0 0 0 0· 3 1092186 0 750308· 0 0R 0T 0 49689099R 52516254T 0· 1 0R 0T 0 5786054R 5786054T 0· 2 0R 0T 0 1140749R 1140749T 0· 3 0R 0T 0 0R 0T 0· 0L 0O 0Y 0N 0F 0A
This series: Jobs: 12 (f=12): [r(12)][100.0%][r=6447MiB/s][r=1650k IOPS][eta 00m:00s] mglru: (groupid=0, jobs=12): err= 0: pid=3665: Fri Dec 15 00:16:06 2023 read: IOPS=1830k, BW=7148MiB/s (7495MB/s)(2094GiB/300001msec) slat (usec): min=59, max=35006, avg=198.58, stdev=201.99 clat (nsec): min=972, max=37489k, avg=630651.61, stdev=384748.50 lat (usec): min=108, max=39688, avg=829.26, stdev=461.06 clat percentiles (usec): | 1.00th=[ 355], 5.00th=[ 379], 10.00th=[ 392], 20.00th=[ 424], | 30.00th=[ 478], 40.00th=[ 510], 50.00th=[ 529], 60.00th=[ 553], | 70.00th=[ 635], 80.00th=[ 898], 90.00th=[ 1012], 95.00th=[ 1090], | 99.00th=[ 1221], 99.50th=[ 1401], 99.90th=[ 2606], 99.95th=[ 3654], | 99.99th=[18220] bw ( MiB/s): min= 4870, max= 9145, per=100.00%, avg=7157.39, stdev=81.13, samples=7176 iops : min=1246811, max=2341342, avg=1832289.80, stdev=20768.76, samples=7176 lat (nsec) : 1000=0.01% lat (usec) : 4=0.01%, 10=0.01%, 250=0.01%, 500=36.53%, 750=36.20% lat (usec) : 1000=15.90% lat (msec) : 2=11.18%, 4=0.15%, 10=0.02%, 20=0.01%, 50=0.01% cpu : usr=8.59%, sys=91.27%, ctx=512635, majf=0, minf=711 IO depths : 1=0.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=100.0% submit : 0=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=100.0%, 64=0.0%, >=64=0.0% complete : 0=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=100.0%, 64=0.0%, >=64=0.1% issued rwts: total=548956313,0,0,0 short=0,0,0,0 dropped=0,0,0,0 latency : target=0, window=0, percentile=100.00%, depth=128
Run status group 0 (all jobs): READ: bw=7148MiB/s (7495MB/s), 7148MiB/s-7148MiB/s (7495MB/s-7495MB/s), io=2094GiB (2249GB), run=300001-300001msec
workingset_refault_file 596790506
memcg 68 /benchmark node 0 122 160248 0 0x 0 0r 0e 0p 0 0 0· 1 0r 0e 0p 0 0 0· 2 0r 0e 0p 0 0 0· 3 0r 0e 0p 0 0 0· 0 0 0 0 0 0· 123 155360 0 239405· 0 0 0 0 301462r 1186271e 0p 1 0 0 0 80013r 218961e 0p 2 0 0 0 0r 0e 516139p 3 0 0 0 0r 0e 0p 0 0 0 0 0 0· 124 150495 0 516188· 0 0 0 0 0 0 0· 1 0 0 0 0 0 0· 2 0 0 0 0 0 0· 3 0 0 0 0 0 0· 0 0 0 0 0 0· 125 145582 0 1345· 0 0R 0T 0 2577270R 4518284T 0· 1 0R 0T 0 290933R 369324T 0· 2 0R 0T 0 0R 752170T 0· 3 0R 0T 0 0R 0T 0· 388483L 17226O 18419Y 95408N 1314F 578A
I think the problem might be related to this series ages faster and so have higher overhead in some case. In your test the test is large scaled so MGLRU just keep reclaiming last gen, no aging, and my RFC bring extra overhead due to workingset checking and memcg flushing (the memcg flushing patch in unstable tree may help?), and also the current refault distance checking model, simply glued to MGLRU (some known issues, the most obvious issue is that refault distance check can't prevent the file page underprotected issue at all when active is low or empty, and using active/inactive is not accurate enough for MGLRU), not performing good enough.
And for the MongoDB test, I still didn't have enough time to tidy up the setup scripts and modified repo yet, sorry about this, in past few days I only have time to check this issue at late night... but a quick test shows interesting reading too:
RFC: ================================================================== Execution Results after 902 seconds ------------------------------------------------------------------ Executed Time (µs) Rate············ STOCK_LEVEL 2544 27114484261.0 0.09 txn/s······ ------------------------------------------------------------------ TOTAL 2544 27114484261.0 0.09 txn/s······
workingset_refault_anon 10512 workingset_refault_file 22751782
memcg 44 /system.slice/docker-1313de5323016713a0efa95d3b3f1aeafc9f43df80051bd013f3d29f1e13fa58.scope node 0 12 190714 41736 640699· 0 0r 2e 0p 0r 1293703e 0p 1 0r 0e 0p 0r 0e 463477p 2 0r 0e 0p 0r 0e 5029378p 3 0r 0e 0p 0r 0e 0p 0 0 0 0 0 0· 13 139686 462351 5483828· 0 0 0 0 0 0 0· 1 0 0 0 0 0 0· 2 0 0 0 0 0 0· 3 0 0 0 0 0 0· 0 0 0 0 0 0· 14 86529 692892 3795· 0 0 0 0 0 0 0· 1 0 0 0 0 0 0· 2 0 0 0 0 0 0· 3 0 0 0 0 0 0· 0 0 0 0 0 0· 15 41548 47767 366· 0 12R 1113T 0 3497R 1857252T 0· 1 0R 0T 0 1000193R 1692818T 0· 2 0R 0T 0 0R 5422505T 0· 3 0R 0T 0 0R 0T 0· 3889671L 42917O 3674613Y 11910N 7609F 7547A
This series: ================================================================== Execution Results after 904 seconds ------------------------------------------------------------------ Executed Time (µs) Rate············ STOCK_LEVEL 1668 27108414456.6 0.06 txn/s······ ------------------------------------------------------------------ TOTAL 1668 27108414456.6 0.06 txn/s······
workingset_refault_anon 35277 workingset_refault_file 20335355
memcg 77 /system.slice/docker-731f3d33dca1dbea9d763a7a9519bb92c4ca1bbdb06c6a23d5203f8baad97f6e.scope node 0 14 218191 0x 0x 0 0 0 0 0 0 0· 1 0 0 0 0 0 0· 2 0 0 0 0 0 0· 3 0 0 0 0 0 0· 0 0 0 0 0 0· 15 170722 1923 6172558· 0 0r 0e 0p 9r 29052e 0p 1 0r 0e 0p 0r 10643e 0p 2 0r 0e 0p 0r 0e 5714p 3 0r 0e 0p 0r 0e 0p 0 0 0 0 0 0· 16 127628 1223689 10249· 0 0 0 0 0 0 0· 1 0 0 0 0 0 0· 2 0 0 0 0 0 0· 3 0 0 0 0 0 0· 0 0 0 0 0 0· 17 79949 40444 408· 0 1413R 5628T 0 352479R 1259370T 0· 1 0R 0T 0 252950R 439843T 0· 2 0R 1T 0 0R 5083446T 0· 3 0R 0T 0 0R 0T 0· 18667726L 229222O 17641112Y 40116N 36473F 35963A
And I've turned off all unrelated features off (psi, delayacct) for above tests. When PSI is on, the MongoDB test shows 70 - 100 PSI SOME, it's not using a very high performance disk. I think this could suggest some time evict of file page is not that costly. And page shadow can store fine grained data of page's access distance, so maybe I can tune the refault distance checking model for MGLRU, combine with this series, which may help to improve the protect policy to be more balanced (not too fast, and still accurate)?