Volume stuff like read/write_latency, read/write_data, and read/write_ops are typically good, as well as cpu load. Total transactions on an aggregate basis is useful too.Oh and free space on the aggregate and volume.
The latest nfs client package, includes nfs-iostat which can report avg round trip times which can be helpful to diagnose what's slow from a client point of view.
-Blake
On Thu, Sep 30, 2010 at 8:38 AM, Nicholas Bernstein nick@nicholasbernstein.com wrote:
I had an interesting question posed by a student yesterday - "aside from interfaces being up, and volumes being online, what do you typically monitor?" Some of my initial thoughts were:
- inodes being available - ifstat on multimode vifs to make sure interfaces are being used - io counts on luns/qtrees/vols/ - client side nfsstat or the equivalent for that protocol from the client side
Anyway, I was trying to think of some of the "non-standard" things to monitor, and thought I'd put it out to the list and see what other people are typically doing.
Cheers, Nick