[Bug 224080] ZFS: "arc_dnlc_evicts_thr" + "arc_reclaim_thread" - high CPU load during poudriere build...

bugzilla-noreply at freebsd.org bugzilla-noreply at freebsd.org
Thu Dec 7 05:14:32 UTC 2017


https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=224080

--- Comment #9 from Cy Schubert <cy at FreeBSD.org> ---
I can confirm that reverting r326347 resolves the issue. My zpools no longer
thrash. Performance, due to greatly reduced I/O, has vastly improved.

So far so good:

 last pid: 64954;  load averages:  1.87,  1.86,  2.05    up 0+00:58:45 
21:12:59
71 processes:  3 running, 68 sleeping
CPU:  0.0% user, 64.0% nice, 33.3% system,  0.2% interrupt,  2.6% idle
Mem: 1285M Active, 1367M Inact, 4489M Wired, 650M Buf, 800M Free
ARC: 1997M Total, 356M MFU, 1313M MRU, 8244K Anon, 23M Header, 298M Other
     1302M Compressed, 3372M Uncompressed, 2.59:1 Ratio
Swap: 20G Total, 20G Free

  PID USERNAME       THR PRI NICE   SIZE    RES STATE   C   TIME    WCPU COMMAN
64687 root             1  52   10 10604K  1996K select  1   0:00   1.63% make
60776 root             1  52   10 14460K  6092K select  1   0:00   0.41% make
33699 root             1  20    0 14312K  5308K select  0   0:00   0.22% screen
60761 root             1  46   10 10984K  2520K select  1   0:00   0.14% make
61698 root             1  20    0 13920K  4124K CPU1    1   0:00   0.11% top
33888 root             1  30   10 10196K  1500K select  0   0:00   0.11% make
64686 root             1  52   10 11692K  3180K wait    0   0:00   0.11% sh
33914 root             1  30   10 10996K  2520K select  0   0:00   0.10% make
 2712 root             1  20    0 14352K  5624K select  1   0:00   0.02% devd
 3166 uucp             1  20    0 12288K  3856K nanslp  1   0:00   0.01% upsmon
 3127 root             1  20    0 10860K  2456K select  0   0:00   0.01% powerd
 3124 root             2  20    0 19264K 19396K select  1   0:00   0.01% ntpd
 2795 root             1  20    0 11400K  2616K select  0   0:00   0.01% syslog
 3358 root             1  20    0 15748K  6640K select  1   0:00   0.00% sendma


And a different machine:

last pid: 81498;  load averages:  1.86,  1.88,  2.04    up 0+00:59:45  21:13:59
62 processes:  2 running, 60 sleeping
CPU:  0.4% user, 13.9% nice, 19.4% system,  0.6% interrupt, 65.7% idle
Mem: 1291M Active, 1381M Inact, 4490M Wired, 650M Buf, 778M Free
ARC: 2031M Total, 349M MFU, 1331M MRU, 8254K Anon, 24M Header, 319M Other
     1313M Compressed, 3410M Uncompressed, 2.60:1 Ratio
Swap: 20G Total, 20G Free

  PID USERNAME       THR PRI NICE   SIZE    RES STATE   C   TIME    WCPU COMMAN
81192 root             1  83   10 19976K 11788K CPU1    1   0:00   5.95% make
33699 root             1  20    0 14312K  5308K select  1   0:01   0.37% screen
33888 root             1  30   10 10196K  1500K select  0   0:00   0.29% make
81169 root             1  52   10 14320K  5936K select  0   0:00   0.22% make
33914 root             1  30   10 11016K  2524K select  0   0:00   0.16% make
61698 root             1  20    0 13920K  4124K CPU0    0   0:00   0.10% top
 2712 root             1  20    0 14352K  5624K select  0   0:00   0.02% devd
 3127 root             1  20    0 10860K  2456K select  1   0:00   0.01% powerd
 3166 uucp             1  20    0 12288K  3856K nanslp  1   0:00   0.01% upsmon
 3124 root             2  20    0 19264K 19396K select  0   0:00   0.01% ntpd
 2795 root             1  20    0 11400K  2616K select  0   0:00   0.01% syslog
 3358 root             1  20    0 15748K  6640K select  0   0:00   0.00% sendma
25899 root             1  20    0 12876K  3684K select  0   0:00   0.00% klogin
33698 root             1  20    0 12392K  3648K pause   1   0:00   0.00% screen

-- 
You are receiving this mail because:
You are the assignee for the bug.


More information about the freebsd-fs mailing list