2015-04-23 01:36:40

by Yuanhan Liu

[permalink] [raw]
Subject: performance changes on c9dc4c65: 9.8% fsmark.files_per_sec

FYI, we found performance increasement, which is expected as commit patch says,
on `fsmark.files_per_sec' by c9dc4c6578502c2085705347375b82089aad18d0:

> commit c9dc4c6578502c2085705347375b82089aad18d0
> Author: Chris Mason <[email protected]>
> AuthorDate: Sat Apr 4 17:14:42 2015 -0700
> Commit: Chris Mason <[email protected]>
> CommitDate: Fri Apr 10 14:07:11 2015 -0700
>
> Btrfs: two stage dirty block group writeout

4c6d1d85ad89fd8e32dc9204b7f944854399bda9 c9dc4c6578502c2085705347375b82089aad18d0
---------------------------------------- ----------------------------------------
run time(m) metric_value ?stddev run time(m) metric_value ?stddev change testbox/benchmark/testcase-params
--- ------ ---------------------------- --- ------ ---------------------------- -------- ------------------------------
3 7.3 |35.267| ?0.5 5 6.6 |38.740| ?1.6 9.8% ivb44/fsmark/1x-1t-1HDD-btrfs-4M-60G-NoSync


NOTE: here are some more explanation about those test parameters for you to
know what the testcase does better:

1x: where 'x' means iterations or loop, corresponding to the 'L' option of fsmark

1t, 64t: where 't' means thread

4M: means the single file size, corresponding to the '-s' option of fsmark
60G: means the total test size


And FYI, here are more changes by the same commit:

4c6d1d85ad89fd8e c9dc4c6578502c208570534737
---------------- --------------------------
%stddev %change %stddev
\ | \
9864 ? 2% +156.9% 25345 ? 4% fsmark.time.voluntary_context_switches
9 ? 0% +17.8% 10 ? 4% fsmark.time.percent_of_cpu_this_job_got
462211 ? 1% +16.8% 539707 ? 0% fsmark.app_overhead
35.27 ? 0% +9.8% 38.74 ? 1% fsmark.files_per_sec
435 ? 0% -9.0% 396 ? 1% fsmark.time.elapsed_time.max
435 ? 0% -9.0% 396 ? 1% fsmark.time.elapsed_time
5.20 ? 2% -70.3% 1.54 ? 6% turbostat.Pkg%pc6
2447873 ? 42% -67.9% 785086 ? 33% numa-numastat.node1.numa_hit
2413662 ? 43% -68.1% 771115 ? 31% numa-numastat.node1.local_node
9864 ? 2% +156.9% 25345 ? 4% time.voluntary_context_switches
187680 ? 10% +126.8% 425676 ? 7% numa-vmstat.node1.nr_dirty
747361 ? 9% +127.8% 1702809 ? 7% numa-meminfo.node1.Dirty
1787510 ? 1% +117.0% 3878984 ? 2% meminfo.Dirty
446861 ? 1% +117.0% 969472 ? 2% proc-vmstat.nr_dirty
1655962 ? 37% -59.3% 673988 ? 29% numa-vmstat.node1.numa_local
1036191 ? 8% +110.3% 2179311 ? 3% numa-meminfo.node0.Dirty
259069 ? 8% +110.3% 544783 ? 3% numa-vmstat.node0.nr_dirty
1687987 ? 37% -58.6% 698626 ? 29% numa-vmstat.node1.numa_hit
1 ? 0% +100.0% 2 ? 0% vmstat.procs.b
0.02 ? 0% +100.0% 0.04 ? 22% turbostat.CPU%c3
6.03 ? 1% +76.9% 10.67 ? 1% turbostat.CPU%c1
5.189e+08 ? 0% +72.6% 8.956e+08 ? 1% cpuidle.C1-IVT.time
2646692 ? 7% +75.0% 4630890 ? 23% cpuidle.C3-IVT.time
5301 ? 6% -31.7% 3620 ? 3% slabinfo.btrfs_ordered_extent.active_objs
10549 ? 16% -30.3% 7349 ? 12% numa-vmstat.node1.nr_slab_reclaimable
5353 ? 6% -31.4% 3670 ? 3% slabinfo.btrfs_ordered_extent.num_objs
42169 ? 16% -30.3% 29397 ? 12% numa-meminfo.node1.SReclaimable
1619825 ? 22% +39.4% 2258188 ? 4% proc-vmstat.pgfree
4611 ? 7% -28.0% 3318 ? 1% slabinfo.btrfs_delayed_ref_head.num_objs
4471 ? 8% -27.0% 3264 ? 2% slabinfo.btrfs_delayed_ref_head.active_objs
67.93 ? 1% -24.7% 51.15 ? 4% turbostat.Pkg%pc2
2332975 ? 21% +45.6% 3396446 ? 4% numa-vmstat.node1.numa_other
2300949 ? 22% +46.5% 3371807 ? 4% numa-vmstat.node1.numa_miss
2300941 ? 22% +46.5% 3371793 ? 4% numa-vmstat.node0.numa_foreign
2952 ? 8% -23.3% 2263 ? 3% slabinfo.btrfs_delayed_data_ref.num_objs
2570716 ? 3% +25.7% 3230157 ? 2% numa-meminfo.node1.Writeback
642367 ? 3% +25.7% 807533 ? 2% numa-vmstat.node1.nr_writeback
95408 ? 13% -17.3% 78910 ? 6% numa-meminfo.node1.Slab
2803 ? 7% -21.1% 2210 ? 3% slabinfo.btrfs_delayed_data_ref.active_objs
240 ? 9% +23.1% 295 ? 16% numa-vmstat.node0.nr_page_table_pages
4626942 ? 19% +49.6% 6924087 ? 22% cpuidle.C1E-IVT.time
5585235 ? 0% +25.5% 7011242 ? 0% meminfo.Writeback
1396232 ? 0% +25.5% 1752892 ? 0% proc-vmstat.nr_writeback
962 ? 9% +23.0% 1184 ? 16% numa-meminfo.node0.PageTables
9 ? 0% +17.8% 10 ? 4% time.percent_of_cpu_this_job_got
754027 ? 2% +25.2% 944312 ? 1% numa-vmstat.node0.nr_writeback
3018674 ? 2% +25.1% 3777338 ? 1% numa-meminfo.node0.Writeback
23509 ? 1% -16.9% 19530 ? 0% slabinfo.kmalloc-1024.active_objs
2972 ? 1% +21.4% 3607 ? 0% proc-vmstat.nr_alloc_batch
13956 ? 4% -15.6% 11773 ? 8% slabinfo.kmalloc-192.active_objs
743 ? 1% -16.0% 624 ? 0% slabinfo.kmalloc-1024.active_slabs
743 ? 1% -16.0% 624 ? 0% slabinfo.kmalloc-1024.num_slabs
23790 ? 1% -16.0% 19983 ? 0% slabinfo.kmalloc-1024.num_objs
68983 ? 2% +19.1% 82190 ? 4% softirqs.RCU
222 ? 11% +47.0% 326 ? 25% cpuidle.POLL.usage
14177 ? 0% +17.8% 16702 ? 1% slabinfo.kmalloc-2048.num_objs
14045 ? 0% +18.0% 16568 ? 1% slabinfo.kmalloc-2048.active_objs
885 ? 0% +17.8% 1043 ? 1% slabinfo.kmalloc-2048.num_slabs
885 ? 0% +17.8% 1043 ? 1% slabinfo.kmalloc-2048.active_slabs
14025 ? 4% -13.3% 12157 ? 7% slabinfo.kmalloc-192.num_objs
8287205 ? 10% +16.0% 9611684 ? 0% numa-numastat.node0.numa_hit
8276795 ? 10% +15.9% 9592682 ? 0% numa-numastat.node0.local_node
2615463 ? 5% -9.6% 2365256 ? 2% numa-vmstat.node1.nr_written
1814 ? 5% -12.7% 1584 ? 11% numa-meminfo.node1.PageTables
453 ? 5% -12.6% 396 ? 11% numa-vmstat.node1.nr_page_table_pages
105943 ? 6% +13.6% 120352 ? 2% numa-meminfo.node0.SReclaimable
26492 ? 6% +13.6% 30086 ? 2% numa-vmstat.node0.nr_slab_reclaimable
0.41 ? 1% +17.1% 0.48 ? 4% time.user_time
2155 ? 4% -11.1% 1916 ? 5% slabinfo.btrfs_delayed_tree_ref.active_objs
2.028e+10 ? 0% -11.1% 1.803e+10 ? 1% cpuidle.C6-IVT.time
2155 ? 4% -10.8% 1922 ? 5% slabinfo.btrfs_delayed_tree_ref.num_objs
1202 ? 4% -11.2% 1067 ? 9% slabinfo.btrfs_trans_handle.num_objs
1202 ? 4% -11.2% 1067 ? 9% slabinfo.btrfs_trans_handle.active_objs
192641 ? 5% +9.8% 211569 ? 2% numa-meminfo.node0.Slab
268137 ? 0% +12.2% 300911 ? 2% cpuidle.C6-IVT.usage
435 ? 0% -9.0% 396 ? 1% time.elapsed_time
435 ? 0% -9.0% 396 ? 1% time.elapsed_time.max
21057 ? 0% -9.0% 19165 ? 1% uptime.idle
29.89 ? 0% +37.2% 41.01 ? 3% turbostat.CorWatt
59.95 ? 0% +19.6% 71.69 ? 2% turbostat.PkgWatt
18873 ? 0% +14.9% 21692 ? 1% vmstat.system.cs
21 ? 2% +8.8% 23 ? 3% turbostat.Avg_MHz
135 ? 0% +9.1% 147 ? 0% iostat.sda.avgqu-sz
0.69 ? 2% +7.2% 0.74 ? 3% turbostat.%Busy
7478 ? 0% +5.1% 7861 ? 0% iostat.sda.await
7478 ? 0% +5.1% 7861 ? 0% iostat.sda.w_await
239 ? 0% +3.6% 247 ? 0% iostat.sda.wrqm/s
3.54 ? 0% +3.9% 3.68 ? 0% turbostat.RAMWatt
129619 ? 0% +3.2% 133743 ? 0% vmstat.io.bo
128667 ? 0% +1.9% 131056 ? 0% iostat.sda.wkB/s


--yliu