At 13:53, on May 6, 2005, markallen@micron.com wrote:
I'm investigating a performance problem with snapmirror and disrupting production services on my FAS980C. I've seen it impact both NFS and CIFS. The filer becomes unresponsive for 30s to 1 minute when a snapmirror process kicks off. The snapmirror process is actually on the destination end of the process so it is pulling the data from another FAS980. I'm running 7.0.0.1 on a raid 4 traditional volume with 50 spindles in the volume. My cpu's and memory are far from exceeding max capacity. Anybody have a clue as to what could be causing this?
Snapmirrors force a lot of disk write IO, and hence a lot of CPs. That's both when when a snapshot is created, and even more when a snapshot is deleted.
If your filers do a lot of writes in general, then you might be maxing out CP. When you reach the state of being in back-to-back CP's, then the filer starts not honoring Read requests, so that it can finish flushing its writes.
Run "sysstat -u 1" and watch the CP columns. I'd bet that during the problem, you'll see that you're in CP continuously, and that you end up seeing #'s or B's in the CP type column.
(I'm speaking for ONTAP 6.4.x - but I don't expect that this has changed much in 7.x.)
Davin.