From 2cdb3909c9e95879b6fa63a49caaa3b8824f1cb8 Mon Sep 17 00:00:00 2001 From: Josef Bacik Date: Wed, 24 Mar 2021 09:44:21 -0400 Subject: [PATCH] btrfs: use percpu_read_positive instead of sum_positive for need_preempt Looking at perf data for a fio workload I noticed that we were spending a pretty large chunk of time (around 5%) doing percpu_counter_sum() in need_preemptive_reclaim. This is silly, as we only want to know if we have more ordered than delalloc to see if we should be counting the delayed items in our threshold calculation. Change this to percpu_read_positive() to avoid the overhead. I ran this through fsperf to validate the changes, obviously the latency numbers in dbench and fio are quite jittery, so take them as you wish, but overall the improvements on throughput, iops, and bw are all positive. Each test was run two times, the given value is the average of both runs for their respective column. btrfs ssd normal test results bufferedrandwrite16g results metric baseline current diff ========================================================== write_io_kbytes 16777216 16777216 0.00% read_clat_ns_p99 0 0 0.00% write_bw_bytes 1.04e+08 1.05e+08 1.12% read_iops 0 0 0.00% write_clat_ns_p50 13888 11840 -14.75% read_io_kbytes 0 0 0.00% read_io_bytes 0 0 0.00% write_clat_ns_p99 35008 29312 -16.27% read_bw_bytes 0 0 0.00% elapsed 170 167 -1.76% write_lat_ns_min 4221.50 3762.50 -10.87% sys_cpu 39.65 35.37 -10.79% write_lat_ns_max 2.67e+10 2.50e+10 -6.63% read_lat_ns_min 0 0 0.00% write_iops 25270.10 25553.43 1.12% read_lat_ns_max 0 0 0.00% read_clat_ns_p50 0 0 0.00% dbench60 results metric baseline current diff ================================================== qpathinfo 11.12 12.73 14.52% throughput 416.09 445.66 7.11% flush 3485.63 1887.55 -45.85% qfileinfo 0.70 1.92 173.86% ntcreatex 992.60 695.76 -29.91% qfsinfo 2.43 3.71 52.48% close 1.67 3.14 88.09% sfileinfo 66.54 105.20 58.10% rename 809.23 619.59 -23.43% find 16.88 15.46 -8.41% unlink 820.54 670.86 -18.24% writex 3375.20 2637.91 -21.84% deltree 386.33 449.98 16.48% readx 3.43 3.41 -0.60% mkdir 0.05 0.03 -38.46% lockx 0.26 0.26 -0.76% unlockx 0.81 0.32 -60.33% dio4kbs16threads results metric baseline current diff ================================================================ write_io_kbytes 5249676 3357150 -36.05% read_clat_ns_p99 0 0 0.00% write_bw_bytes 89583501.50 57291192.50 -36.05% read_iops 0 0 0.00% write_clat_ns_p50 242688 263680 8.65% read_io_kbytes 0 0 0.00% read_io_bytes 0 0 0.00% write_clat_ns_p99 15826944 36732928 132.09% read_bw_bytes 0 0 0.00% elapsed 61 61 0.00% write_lat_ns_min 42704 42095 -1.43% sys_cpu 5.27 3.45 -34.52% write_lat_ns_max 7.43e+08 9.27e+08 24.71% read_lat_ns_min 0 0 0.00% write_iops 21870.97 13987.11 -36.05% read_lat_ns_max 0 0 0.00% read_clat_ns_p50 0 0 0.00% randwrite2xram results metric baseline current diff ================================================================ write_io_kbytes 24831972 28876262 16.29% read_clat_ns_p99 0 0 0.00% write_bw_bytes 83745273.50 92182192.50 10.07% read_iops 0 0 0.00% write_clat_ns_p50 13952 11648 -16.51% read_io_kbytes 0 0 0.00% read_io_bytes 0 0 0.00% write_clat_ns_p99 50176 52992 5.61% read_bw_bytes 0 0 0.00% elapsed 314 332 5.73% write_lat_ns_min 5920.50 5127 -13.40% sys_cpu 7.82 7.35 -6.07% write_lat_ns_max 5.27e+10 3.88e+10 -26.44% read_lat_ns_min 0 0 0.00% write_iops 20445.62 22505.42 10.07% read_lat_ns_max 0 0 0.00% read_clat_ns_p50 0 0 0.00% untarfirefox results metric baseline current diff ============================================== elapsed 47.41 47.40 -0.03% Signed-off-by: Josef Bacik Signed-off-by: David Sterba --- fs/btrfs/space-info.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/fs/btrfs/space-info.c b/fs/btrfs/space-info.c index 2da6177..2dc674b 100644 --- a/fs/btrfs/space-info.c +++ b/fs/btrfs/space-info.c @@ -861,8 +861,8 @@ static bool need_preemptive_reclaim(struct btrfs_fs_info *fs_info, * of heavy DIO or ordered reservations, preemptive flushing will just * waste time and cause us to slow down. */ - ordered = percpu_counter_sum_positive(&fs_info->ordered_bytes); - delalloc = percpu_counter_sum_positive(&fs_info->delalloc_bytes); + ordered = percpu_counter_read_positive(&fs_info->ordered_bytes); + delalloc = percpu_counter_read_positive(&fs_info->delalloc_bytes); if (ordered >= delalloc) used += fs_info->delayed_refs_rsv.reserved + fs_info->delayed_block_rsv.reserved; -- 2.7.4