Hello, thanks for your suggestion. I tried settin zfs_arc_max to 0x30000000
(768MB, out of 3GB). The system ran for almost 45 minutes before it froze.
Here's an interesting piece of arcstat.pl, which I noticed just as it was
pasing by:
Time read miss miss% dmis dm% pmis pm% mmis mm% arcsz c
15:17:41 152 152 100 152 100 0 0 152 100 2G 805M
15:17:42 139 139 100 139 100 0 0 139 100 2G 805M
<State Changed>
15:17:43 188 188 100 188 100 0 0 188 100 2G 805M
15:17:44 150 150 100 150 100 0 0 150 100 2G 805M
15:17:45 151 151 100 151 100 0 0 151 100 2G 805M
15:17:46 149 149 100 149 100 0 0 149 100 2G 805M
15:17:47 161 161 100 161 100 0 0 161 100 2G 805M
15:17:48 153 153 100 153 100 0 0 153 100 2G 219M
15:17:49 140 140 100 140 100 0 0 140 100 2G 100M
15:17:50 143 143 100 143 100 0 0 143 100 2G 100M
15:17:51 145 145 100 145 100 0 0 145 100 2G 100M
notice how it suddenly drops "c" from 805M to 100M in 2 seconds. Also "arcsz"
is 2G, which is weird because it shouldn't grow beyond 0x30000000 (768M),
right? And it's also weird to also get 100% MISS ratio
Here's "top" just before it froze:
last pid: 5253; load avg: 0.47, 0.37, 0.33; up 0+00:44:53
15:20:14
77 processes: 75 sleeping, 1 running, 1 on cpu
CPU states: 57.5% idle, 1.0% user, 41.6% kernel, 0.0% iowait, 0.0% swap
Memory: 3072M phys mem, 28M free mem, 2055M swap, 1994M free swap
PID USERNAME LWP PRI NICE SIZE RES STATE TIME CPU COMMAND
1248 root 1 59 0 5940K 2736K sleep 0:14 0.82% arcstat.pl
5206 root 9 59 0 47M 4892K sleep 0:01 0.35% java
855 root 2 59 0 5076K 1588K sleep 0:09 0.33% apcupsd
3134 root 1 59 0 5152K 1764K sleep 0:02 0.26% zpool
1261 root 1 59 0 4104K 588K cpu 0:03 0.22% top
3125 root 1 59 0 6352K 1536K sleep 0:00 0.06% sshd
1151 root 1 59 0 6352K 1504K sleep 0:00 0.05% sshd
62 root 1 59 0 1832K 540K sleep 0:01 0.05% powernowd
849 root 1 59 0 11M 1100K sleep 0:00 0.05% snmpd
465 proxy 1 59 0 15M 2196K run 0:00 0.04% squid
271 daemon 1 59 0 6652K 264K sleep 0:00 0.03% rcapd
1252 root 1 59 0 6352K 1292K sleep 0:00 0.02% sshd
7 root 14 59 0 12M 5412K sleep 0:04 0.02% svc.startd
880 root 1 59 0 6276K 2076K sleep 0:00 0.02% httpd
847 root 1 59 0 2436K 1148K sleep 0:00 0.02% dhcpagent
and finally, zpool iostat 1:
tera 1.51T 312G 207 0 1.22M 0
tera 1.51T 312G 141 0 854K 0
tera 1.51T 312G 70 0 427K 0
tera 1.51T 312G 204 0 1.20M 0
tera 1.51T 312G 187 0 1.10M 0
tera 1.51T 312G 179 0 1.05M 0
tera 1.51T 312G 120 0 743K 0
tera 1.51T 312G 94 0 580K 0
tera 1.51T 312G 77 0 471K 0
tera 1.51T 312G 115 0 696K 0
Which shows a very poor read performance, for a 4xSATA2 (this array usually
saturates my gigabit ethernet). And it's not that the kernel is processing that
much data because the CPU is 57% idle and I THINK powernowd is making it run at
900MHz.
Hernán
This message posted from opensolaris.org
_______________________________________________
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
