we have many flexvols in a large aggregate. our filer 960C is always 100% cpu and about 11,000 to 18,000 nfs ops/sec. some people have started complaining about performance.
how can we tell what flexvols may be overutilized or consuming most of the traffic so we can perhaps migrate them elsewhere?
also we only have two fibre loops with 7 shelves on each loop, if we added a third and split them down to about 4 shelves or so per loop would that improve i/o?
my favorite command for this is stats. Run
stats show -i 1 -n 30 volume:*:total_ops
That should show you the total ops of each volume on your filer, regardless of aggregate. It might also be interesting to show the average latencys as well.
stats show -i 1 -n 30 volume:*:read_latency stats show -i 1 -n 30 volume:*:write_latency
The system doing the most ops may not be the volume with the highest latency completing the request.
I usually watch sysstat -us 1 to see disk throughput. Almost never do you see the backend loop being a bottleneck for normal operations. Disk rebuilds and the like spike usage, but generally the backend loop is fine.
-Blake
On 6/30/07, Mopar Guy westlaguy@gmail.com wrote:
we have many flexvols in a large aggregate. our filer 960C is always 100% cpu and about 11,000 to 18,000 nfs ops/sec. some people have started complaining about performance.
how can we tell what flexvols may be overutilized or consuming most of the traffic so we can perhaps migrate them elsewhere?
also we only have two fibre loops with 7 shelves on each loop, if we added a third and split them down to about 4 shelves or so per loop would that improve i/o?
-
Blake Golliher -
Mopar Guy