Two scenarios:
1. R200, 7.2.2, two 12 TB aggregates, 85 volumes of various sizes, 70-80% full snapvault secondary, ALWAYS receiving snapvaults and nearly ALWAYS spinning to tape.
2. 6070, 7.2.4, two 8 TB aggregates, 27 volumes between 150 and 700 GB, 60-70% full recent migration (via volume snapmirrors) from 980's, serves home directories to NFS and CIFS clients.
We stopped all snapmirrors/snapvaults to the R200, terminated CIFS, and shut off NFS. CPU utilization remained 90%+ and disk reads remained high for the next 6-7-8 hours until it finally quieted down.
After volume snapmirroring the home dirs from the 980's to the 6070, we re-pointed the snapvaults (modify; start -r) and allowed NFS and CIFS access. 5 days later, file server performance still blows (worse than the 980's -- D'oh! Egg, meet Face; Face, meet Egg), with moderate CPU load 30-40-50% but high disk reads.
In both cases, we determined that lots of WAFL scans were going on:
container block reclamation volume deswizzling active bitmap rearrangement (of course...this NEVER seems to stop)
and a few other types that have already scrolled out of my buffer.
So, we figure--eventually--the WAFL scans will catch up and performance will go back to normal on the 6070. The performance of the R200 always blows.
So...here finally is the question: do toasters need a day off?
Said another way, to maintain acceptable performance and low/fast response times, do toasters need a goodly amount of (relatively) idle time in order for these WAFL scans to complete?
Could the sub-optimal performance of another group of six 6070's that we have, that we *pound on* 7x24, be explained in part by always having at least the container block reclamation scan going on?
Until next time...
The MathWorks, Inc. 508-647-7000 x7792 3 Apple Hill Drive, Natick, MA 01760-2098 508-647-7001 FAX tmerrill@mathworks.com http://www.mathworks.com ---