I have a home email/fileserver based on a Netra X1 with two 120Gb disks and 2Gb
RAM which I updated to S10 6/06 so that I could ZFS for mirroring and
snapshotting a data slice. Both disks are partitioned identically: slice 0 is
20Gb is root and mirrored (with SVM) and UFS, slice 1 is swap, slice 3 is 90Gb
and is a ZFS mirrored pool, slice 7 is 50Mb for the SVM metadb.
So far I''ve seen [b]very[/b] high loads twice using ZFS which does not
happen when the same task is implemented with UFS:
1. My email is served to clients by the Dovecot IMAP daemon. Mailboxes are
stored in users home directories (there''s only 5 mailboxes and
there''s rarely more than one accessed at a time). As soon as /home was
moved to ZFS I noticed that as soon as the first imap client access started the
load (as measured with uptime) shot up to around 15 and stays there!
it''s normally around 0.5.
2. I created a zone with the storage path in a new ZFS volume. Installing the
zone took several hours and the load again was around 15. The first boot of this
new zone took 16 hours to complete the svcadm initialisation during which time
the load was between 12 and 15. I''ve just deleted the zone and ZFS
volume and recreated if in UFS and it took 30 minutes to install (load peaked at
5) and 25 minutes to complete the first boot.
I know it''s a low spec machine but it''s been fine running S10
with UFS. Obviously the checksumming in ZFS will add some CPU overhead
(I''m not using any compression). Are there any known problems with ZFS
causing high CPU loads??
Pete
This message posted from opensolaris.org