Re: [PATCH 00/27] Move LRU page reclaim from zones to nodes v7

From: Balbir Singh
Date: Fri Jun 24 2016 - 02:36:13 EST



On 22/06/16 00:15, Mel Gorman wrote:
> (sorry for resend, the previous attempt didn't go through fully for
> some reason)
>
> The bulk of the updates are in response to review from Vlastimil Babka
> and received a lot more testing than v6.
>
> Changelog since v6
> o Correct reclaim_idx when direct reclaiming for memcg
> o Also account LRU pages per zone for compaction/reclaim
> o Add page_pgdat helper with more efficient lookup
> o Init pgdat LRU lock only once
> o Slight optimisation to wake_all_kswapds
> o Always wake kcompactd when kswapd is going to sleep
> o Rebase to mmotm as of June 15th, 2016
>
> Changelog since v5
> o Rebase and adjust to changes
>
> Changelog since v4
> o Rebase on top of v3 of page allocator optimisation series
>
> Changelog since v3
> o Rebase on top of the page allocator optimisation series
> o Remove RFC tag
>
> This is the latest version of a series that moves LRUs from the zones to
> the node that is based upon 4.6-rc3 plus the page allocator optimisation
> series. Conceptually, this is simple but there are a lot of details. Some
> of the broad motivations for this are;
>
> 1. The residency of a page partially depends on what zone the page was
> allocated from. This is partially combatted by the fair zone allocation
> policy but that is a partial solution that introduces overhead in the
> page allocator paths.
>
> 2. Currently, reclaim on node 0 behaves slightly different to node 1. For
> example, direct reclaim scans in zonelist order and reclaims even if
> the zone is over the high watermark regardless of the age of pages
> in that LRU. Kswapd on the other hand starts reclaim on the highest
> unbalanced zone. A difference in distribution of file/anon pages due
> to when they were allocated results can result in a difference in
> again. While the fair zone allocation policy mitigates some of the
> problems here, the page reclaim results on a multi-zone node will
> always be different to a single-zone node.
> it was scheduled on as a result.
>
> 3. kswapd and the page allocator scan zones in the opposite order to
> avoid interfering with each other but it's sensitive to timing. This
> mitigates the page allocator using pages that were allocated very recently
> in the ideal case but it's sensitive to timing. When kswapd is allocating
> from lower zones then it's great but during the rebalancing of the highest
> zone, the page allocator and kswapd interfere with each other. It's worse
> if the highest zone is small and difficult to balance.
>
> 4. slab shrinkers are node-based which makes it harder to identify the exact
> relationship between slab reclaim and LRU reclaim.
>
> The reason we have zone-based reclaim is that we used to have
> large highmem zones in common configurations and it was necessary
> to quickly find ZONE_NORMAL pages for reclaim. Today, this is much
> less of a concern as machines with lots of memory will (or should) use
> 64-bit kernels. Combinations of 32-bit hardware and 64-bit hardware are
> rare. Machines that do use highmem should have relatively low highmem:lowmem
> ratios than we worried about in the past.
>
> Conceptually, moving to node LRUs should be easier to understand. The
> page allocator plays fewer tricks to game reclaim and reclaim behaves
> similarly on all nodes.
>
> The series has been tested on a 16 core UMA machine and a 2-socket 48 core
> NUMA machine. The UMA results are presented in most cases as the NUMA machine
> behaved similarly.
>
> pagealloc
> ---------
>
> This is a microbenchmark that shows the benefit of removing the fair zone
> allocation policy. It was tested uip to order-4 but only orders 0 and 1 are
> shown as the other orders were comparable.
>
> 4.7.0-rc3 4.7.0-rc3
> mmotm-20160615 nodelru-v7r17
> Min total-odr0-1 485.00 ( 0.00%) 462.00 ( 4.74%)
> Min total-odr0-2 354.00 ( 0.00%) 341.00 ( 3.67%)
> Min total-odr0-4 285.00 ( 0.00%) 277.00 ( 2.81%)
> Min total-odr0-8 249.00 ( 0.00%) 240.00 ( 3.61%)
> Min total-odr0-16 230.00 ( 0.00%) 224.00 ( 2.61%)
> Min total-odr0-32 222.00 ( 0.00%) 215.00 ( 3.15%)
> Min total-odr0-64 216.00 ( 0.00%) 210.00 ( 2.78%)
> Min total-odr0-128 214.00 ( 0.00%) 208.00 ( 2.80%)
> Min total-odr0-256 248.00 ( 0.00%) 233.00 ( 6.05%)
> Min total-odr0-512 277.00 ( 0.00%) 270.00 ( 2.53%)
> Min total-odr0-1024 294.00 ( 0.00%) 284.00 ( 3.40%)
> Min total-odr0-2048 308.00 ( 0.00%) 298.00 ( 3.25%)
> Min total-odr0-4096 318.00 ( 0.00%) 307.00 ( 3.46%)
> Min total-odr0-8192 322.00 ( 0.00%) 308.00 ( 4.35%)
> Min total-odr0-16384 324.00 ( 0.00%) 309.00 ( 4.63%)
> Min total-odr1-1 729.00 ( 0.00%) 686.00 ( 5.90%)
> Min total-odr1-2 533.00 ( 0.00%) 520.00 ( 2.44%)
> Min total-odr1-4 434.00 ( 0.00%) 415.00 ( 4.38%)
> Min total-odr1-8 390.00 ( 0.00%) 364.00 ( 6.67%)
> Min total-odr1-16 359.00 ( 0.00%) 335.00 ( 6.69%)
> Min total-odr1-32 356.00 ( 0.00%) 327.00 ( 8.15%)
> Min total-odr1-64 356.00 ( 0.00%) 321.00 ( 9.83%)
> Min total-odr1-128 356.00 ( 0.00%) 333.00 ( 6.46%)
> Min total-odr1-256 354.00 ( 0.00%) 337.00 ( 4.80%)
> Min total-odr1-512 366.00 ( 0.00%) 340.00 ( 7.10%)
> Min total-odr1-1024 373.00 ( 0.00%) 354.00 ( 5.09%)
> Min total-odr1-2048 375.00 ( 0.00%) 354.00 ( 5.60%)
> Min total-odr1-4096 374.00 ( 0.00%) 354.00 ( 5.35%)
> Min total-odr1-8192 370.00 ( 0.00%) 355.00 ( 4.05%)
>
> This shows a steady improvement throughout. The primary benefit is from
> reduced system CPU usage which is obvious from the overall times;
>
> 4.7.0-rc3 4.7.0-rc3
> mmotm-20160615 nodelru-v7
> User 174.06 174.58
> System 2656.78 2485.21
> Elapsed 2885.07 2713.67
>
> The vmstats also showed that the fair zone allocation policy was definitely
> removed as can be seen here;
>
> 4.7.0-rc3 4.7.0-rc3
> mmotm-20160615nodelru-v7r17
> DMA32 allocs 28794408561 0
> Normal allocs 48431969998 77226313470
> Movable allocs 0 0
>
> tiobench on ext4
> ----------------
>
> tiobench is a benchmark that artifically benefits if old pages remain resident
> while new pages get reclaimed. The fair zone allocation policy mitigates this
> problem so pages age fairly. While the benchmark has problems, it is important
> that tiobench performance remains constant as it implies that page aging
> problems that the fair zone allocation policy fixes are not re-introduced.
>
> 4.7.0-rc3 4.7.0-rc3
> mmotm-20160615 nodelru-v7r17
> Min PotentialReadSpeed 90.24 ( 0.00%) 90.14 ( -0.11%)
> Min SeqRead-MB/sec-1 80.63 ( 0.00%) 83.09 ( 3.05%)
> Min SeqRead-MB/sec-2 71.91 ( 0.00%) 72.44 ( 0.74%)
> Min SeqRead-MB/sec-4 75.20 ( 0.00%) 74.32 ( -1.17%)
> Min SeqRead-MB/sec-8 65.30 ( 0.00%) 65.21 ( -0.14%)
> Min SeqRead-MB/sec-16 62.62 ( 0.00%) 62.12 ( -0.80%)
> Min RandRead-MB/sec-1 0.90 ( 0.00%) 0.94 ( 4.44%)
> Min RandRead-MB/sec-2 0.96 ( 0.00%) 0.97 ( 1.04%)
> Min RandRead-MB/sec-4 1.43 ( 0.00%) 1.41 ( -1.40%)
> Min RandRead-MB/sec-8 1.67 ( 0.00%) 1.72 ( 2.99%)
> Min RandRead-MB/sec-16 1.77 ( 0.00%) 1.86 ( 5.08%)
> Min SeqWrite-MB/sec-1 78.12 ( 0.00%) 79.78 ( 2.12%)
> Min SeqWrite-MB/sec-2 72.74 ( 0.00%) 73.23 ( 0.67%)
> Min SeqWrite-MB/sec-4 79.40 ( 0.00%) 78.32 ( -1.36%)
> Min SeqWrite-MB/sec-8 73.18 ( 0.00%) 71.40 ( -2.43%)
> Min SeqWrite-MB/sec-16 75.82 ( 0.00%) 75.24 ( -0.76%)
> Min RandWrite-MB/sec-1 1.18 ( 0.00%) 1.15 ( -2.54%)
> Min RandWrite-MB/sec-2 1.05 ( 0.00%) 0.99 ( -5.71%)
> Min RandWrite-MB/sec-4 1.00 ( 0.00%) 0.96 ( -4.00%)
> Min RandWrite-MB/sec-8 0.91 ( 0.00%) 0.92 ( 1.10%)
> Min RandWrite-MB/sec-16 0.92 ( 0.00%) 0.92 ( 0.00%)
>
> This shows that the series has little or not impact on tiobench which is
> desirable. It indicates that the fair zone allocation policy was removed
> in a manner that didn't reintroduce one class of page aging bug. There
> were only minor differences in overall reclaim activity
>
> 4.7.0-rc3 4.7.0-rc3
> mmotm-20160615nodelru-v7r17
> Minor Faults 640992 640721
> Major Faults 728 623
> Swap Ins 0 0
> Swap Outs 0 0
> DMA allocs 0 0
> DMA32 allocs 46174282 44219717
> Normal allocs 77949344 79858024
> Movable allocs 0 0
> Allocation stalls 38 76
> Direct pages scanned 17463 34865
> Kswapd pages scanned 93331163 93302388
> Kswapd pages reclaimed 93328173 93299677
> Direct pages reclaimed 17463 34865
> Kswapd efficiency 99% 99%
> Kswapd velocity 13729.855 13755.612
> Direct efficiency 100% 100%
> Direct velocity 2.569 5.140
> Percentage direct scans 0% 0%
> Page writes by reclaim 0 0
> Page writes file 0 0
> Page writes anon 0 0
> Page reclaim immediate 54 36
>
> kswapd activity was roughly comparable. There was slight differences
> in direct reclaim activity but negligible in the context of the overall
> workload (velocity of 5 pages per second with the patches applied, 2 pages
> per second in the baseline kernel).
>
> pgbench read-only large configuration on ext4
> ---------------------------------------------
>
> pgbench is a database benchmark that can be sensitive to page reclaim
> decisions. This also checks if removing the fair zone allocation policy
> is safe
>
> pgbench Transactions
> 4.7.0-rc3 4.7.0-rc3
> mmotm-20160615 nodelru-v7r17
> Hmean 1 191.00 ( 0.00%) 193.67 ( 1.40%)
> Hmean 5 338.59 ( 0.00%) 336.99 ( -0.47%)
> Hmean 12 374.03 ( 0.00%) 386.15 ( 3.24%)
> Hmean 21 372.24 ( 0.00%) 372.02 ( -0.06%)
> Hmean 30 383.98 ( 0.00%) 370.69 ( -3.46%)
> Hmean 32 431.01 ( 0.00%) 438.47 ( 1.73%)
>
> Negligible differences again. As with tiobench, overall reclaim activity
> was comparable.
>
> bonnie++ on ext4
> ----------------
>
> No interesting performance difference, negligible differences on reclaim
> stats.
>
> paralleldd on ext4
> ------------------
>
> This workload uses varying numbers of dd instances to read large amounts of
> data from disk.
>
> paralleldd
> 4.7.0-rc3 4.7.0-rc3
> mmotm-20160615 nodelru-v7r17
> Amean Elapsd-1 181.57 ( 0.00%) 179.63 ( 1.07%)
> Amean Elapsd-3 188.29 ( 0.00%) 183.68 ( 2.45%)
> Amean Elapsd-5 188.02 ( 0.00%) 181.73 ( 3.35%)
> Amean Elapsd-7 186.07 ( 0.00%) 184.11 ( 1.05%)
> Amean Elapsd-12 188.16 ( 0.00%) 183.51 ( 2.47%)
> Amean Elapsd-16 189.03 ( 0.00%) 181.27 ( 4.10%)
>
> 4.7.0-rc3 4.7.0-rc3
> mmotm-20160615nodelru-v7r17
> User 1439.23 1433.37
> System 8332.31 8216.01
> Elapsed 3619.80 3532.69
>
> There is a slight gain in performance, some of which is from the reduced system
> CPU usage. There areminor differences in reclaim activity but nothing significant
>
> 4.7.0-rc3 4.7.0-rc3
> mmotm-20160615nodelru-v7r17
> Minor Faults 362486 358215
> Major Faults 1143 1113
> Swap Ins 26 0
> Swap Outs 2920 482
> DMA allocs 0 0
> DMA32 allocs 31568814 28598887
> Normal allocs 46539922 49514444
> Movable allocs 0 0
> Allocation stalls 0 0
> Direct pages scanned 0 0
> Kswapd pages scanned 40886878 40849710
> Kswapd pages reclaimed 40869923 40835207
> Direct pages reclaimed 0 0
> Kswapd efficiency 99% 99%
> Kswapd velocity 11295.342 11563.344
> Direct efficiency 100% 100%
> Direct velocity 0.000 0.000
> Slabs scanned 131673 126099
> Direct inode steals 57 60
> Kswapd inode steals 762 18
>
> It basically shows that kswapd was active at roughly the same rate in
> both kernels. There was also comparable slab scanning activity and direct
> reclaim was avoided in both cases. There appears to be a large difference
> in numbers of inodes reclaimed but the workload has few active inodes and
> is likely a timing artifact. It's interesting to note that the node-lru
> did not swap in any pages but given the low swap activity, it's unlikely
> to be significant.
>
> stutter
> -------
>
> stutter simulates a simple workload. One part uses a lot of anonymous
> memory, a second measures mmap latency and a third copies a large file.
> The primary metric is checking for mmap latency.
>
> stutter
> 4.7.0-rc3 4.7.0-rc3
> mmotm-20160615 nodelru-v7r17
> Min mmap 16.8422 ( 0.00%) 15.9821 ( 5.11%)
> 1st-qrtle mmap 57.8709 ( 0.00%) 58.0794 ( -0.36%)
> 2nd-qrtle mmap 58.4335 ( 0.00%) 59.4286 ( -1.70%)
> 3rd-qrtle mmap 58.6957 ( 0.00%) 59.6862 ( -1.69%)
> Max-90% mmap 58.9388 ( 0.00%) 59.8759 ( -1.59%)
> Max-93% mmap 59.0505 ( 0.00%) 59.9333 ( -1.50%)
> Max-95% mmap 59.1877 ( 0.00%) 59.9844 ( -1.35%)
> Max-99% mmap 60.3237 ( 0.00%) 60.2337 ( 0.15%)
> Max mmap 285.6454 ( 0.00%) 135.6006 ( 52.53%)
> Mean mmap 57.8366 ( 0.00%) 58.4884 ( -1.13%)
>
> This shows that there is a slight impact on mmap latency but that
> the worst-case outlier is much improved. As the problem with this
> benchmark used to be that the kernel stalled for minutes, this
> difference is negligible.
>
> Some of the vmstats are interesting
>
> 4.7.0-rc3 4.7.0-rc3
> mmotm-20160615nodelru-v7r17
> Swap Ins 58 42
> Swap Outs 0 0
> Allocation stalls 16 0
> Direct pages scanned 1374 0
> Kswapd pages scanned 42454910 41782544
> Kswapd pages reclaimed 41571035 41781833
> Direct pages reclaimed 1167 0
> Kswapd efficiency 97% 99%
> Kswapd velocity 14774.479 14223.796
> Direct efficiency 84% 100%
> Direct velocity 0.478 0.000
> Percentage direct scans 0% 0%
> Page writes by reclaim 696918 0
> Page writes file 696918 0
> Page writes anon 0 0
> Page reclaim immediate 2940 137
> Sector Reads 81644424 81699544
> Sector Writes 99193620 98862160
> Page rescued immediate 0 0
> Slabs scanned 1279838 22640
>
> kswapd and direct reclaim activity are similar but the node LRU series
> did not attempt to trigger any page writes from reclaim context.
>
> This series is not without its hazards. There are at least three areas
> that I'm concerned with even though I could not reproduce any problems in
> that area.
>
> 1. Reclaim/compaction is going to be affected because the amount of reclaim is
> no longer targetted at a specific zone. Compaction works on a per-zone basis
> so there is no guarantee that reclaiming a few THP's worth page pages will
> have a positive impact on compaction success rates.
>
> 2. The Slab/LRU reclaim ratio is affected because the frequency the shrinkers
> are called is now different. This may or may not be a problem but if it
> is, it'll be because shrinkers are not called enough and some balancing
> is required.
>
> 3. The anon/file reclaim ratio may be affected. Pages about to be dirtied are
> distributed between zones and the fair zone allocation policy used to do
> something very similar for anon. The distribution is now different but not
> necessarily in any way that matters but it's still worth bearing in mind.
>


Sorry, I am late in reading the thread and the patches, but I am trying to understand
the key benefits? Is it simplification and consistency of the mm subsystem? I know that
zones have grown to be overloaded to mean many things now. What is the contention impact
of moving the LRU from zone to nodes? Your benchmark shows almost no impact with
the micro benchmarks

Balbir Singh