After migrating servers to from UFS to ZFS, we got constant file system performance degradation in couple time by week. System load has not been changed since migration to ZFS, so only reason of this problems can be ZFS itself. We had this problems on Intel server systems with 8GB and more RAM. We also tried to disable prefetch, but problem still existed. Even with latest FreeBSD 8.0 Stable we had same problem. One solution to this problem is - stop all httpd, php-cgi and mysql processes, wait 1-2 minutes and then start it again. On other servers we just migrated back to UFS and problem was gone. One strange thing we noticed is - when server have performance problem, "zpool iostat" is showing like this " pool used avail read write read write ---------- ----- ----- ----- ----- ----- ----- zroot 113G 26.1G 53 392 2.80M 20.3M " after processes restart: " pool used avail read write read write ---------- ----- ----- ----- ----- ----- ----- zroot 99.9G 39.1G 6 40 367K 1.12M " How can "used" value differ?
Responsible Changed From-To: freebsd-bugs->freebsd-fs Over to maintainer(s).
RE: "used value differs": Have you checked with "du" the status "before" and "after" to rule out that something is actually allocating a big file? There is a similar problem reported here: http://permalink.gmane.org/gmane.os.freebsd.stable/66780
Any news on this PR? Can we close it? Is the problem still present with ZFS v28 and latest 8-STABLE or 9-CURRENT? -- Martin Matuska FreeBSD committer http://blog.vx.sk
State Changed From-To: open->closed Closed on submitter request.