On 5/21/07, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote:
> Be careful with %w... it's not that accurate. If you upgrade your > e20k to Solaris 10, you'll lose that as iowait is no longer > calculated (although the %w column is still there for output > compatibility reasons. %b (% busy) is what you should be looking at > instead.
I've found that different workloads tend to generate different graphs (should you graph these parameters). However, if under high load (%b > 80%) your service times svc_t > 30ms ... you have a potentially crap disk subsystem or a slow-down in the works. With more outstanding I/O queries (or a higher wait load) you run the risk of service times going through the roof. I've found that a jump from 30ms and up usually goes very quickly ... once you hit 50-100ms you will notice the slowdown (from an end-user perspective). Futher ... I've managed to convince most customers to generate some code and run standard response queries to see what the actual slowdown from an applications point of view is. So, if db_response.sql shows a spike of 10ms response to a 30ms response, you can corrolate the effect of that with the infrastructure... that is the best way imho. 2c _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
