zfs discuss - Sep 2006 - [Blade 150] ZFS: extreme low performance

Hi forum,

I''m currently a little playing around with ZFS on my workstation.
I created a standard mirrored pool over 2 disk-slices.

# zpool status
 Pool: mypool
 Status: ONLINE
 scrub: Keine erforderlich
config:

        NAME          STATE     READ WRITE CKSUM
        mypool        ONLINE       0     0     0
          mirror        ONLINE       0     0     0
            c0t0d0s4  ONLINE       0     0     0
            c0t2d0s4  ONLINE       0     0     0

Then i created a ZFS with no extra options:

# zfs create mypool/zfs01
# zfs list
NAME                   USED  AVAIL  REFER  MOUNTPOINT
mypool                 106K  27,8G  25,5K  /mypool
mypool/zfs01          24,5K  27,8G  24,5K  /mypool/zfs01

When I now send a mkfile on the new FS, the performance of the whole system
breaks down near zero:

# mkfile 5g test

last pid: 25286;  load avg:  3.54,  2.28,  1.29;       up 0+01:44:26            
16:16:24
66 processes: 61 sleeping, 3 running, 1 zombie, 1 on cpu
CPU states:  0.0% idle,  2.1% user, 97.9% kernel,  0.0% iowait,  0.0% swap
Memory: 512M phys mem, 65M free mem, 2050M swap, 2050M free swap

   PID USERNAME LWP PRI NICE  SIZE   RES STATE    TIME    CPU COMMAND
 25285 root       1   8    4 1184K  752K run      0:09 66.28% mkfile


It seams that some kind of kernel activity while writing to ZFS blocks the
system.
Is this a known problem? Do you need additional information?

regards
Mathias
 
 
This message posted from opensolaris.org

The disks in that Blade 100, are these IDE disks?

The performance problem is probably bug 6421427:
http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6421427


A fix for the issue was integrated into the Opensolaris 20060904 source
drop (actually closed binary drop):

http://dlc.sun.com/osol/on/downloads/20060904/on-changelog-20060904.html

... but has been removed in the next update:

http://dlc.sun.com/osol/on/downloads/20060911/on-changelog-20060911.html
 
 
This message posted from opensolaris.org