[Delayed response, sorry!]
Stefan Parvu wrote:Hi David, >> The classic tools for doing capacity planning measure > operation/transaction rate and time taken, rather > > than resource usage. >> >> Once you've *done* the planning, then you correlate >> that with the amount of resources you used and use >> that to predict when you need to add more resources. >> > Isn't utilization Util=Busy/Time as part of normal Queuing theory > parlance, therefore included as an important part in any capacity > planning along with X(N) and R(N), throughput and response time ? Yes, but note they're computed values, while the time values are measured. The measured values for resource usages are traditionally in units like KB/S or IOPS, and they're hard to turn into utilization numbers. That leaves the poor sysadmin faced with coming up for a value of something that correlates to 100% utilization. Of course, if you know that for your workload, an ancient disk will always bottleneck at 40 IOPS, that's all you need. No problem. I used to do capacity planning that way, finding out the limits by creating application-specific benchmarks, then watching OS metrics to see when my leading indicators got close to the danger point. The 40 IOPS was a 3600 RPM SCSI-1 disk, discovered by benchmarking Samba with a recorded/replayed workload. That's a big whack of work: after doing it several times I bought Jain's and Gunther's books and started looking at a more scientific approach (also an easier one: I'm lazy (;-)) Conversely, Guerrilla CP is about doing a good job under nasty time constraints, and I see a huge amount of value in his analytic model of the utilization curve, not least because it predicts the drop-off after "100%" The more we get visit counts, residence and wait times out of Solaris, the better! I'm particularly impressed with the recent tools to convert lock spins int lock time, even though locks aren't something you can use queue math on. --dave -- David Collier-Brown | Always do right. This will gratify Sun Microsystems, Toronto | some people and astonish the rest dav...@sun.com | -- Mark Twain cell: (647) 833-9377, bridge: (877) 385-4099 code: 506 9191# _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
