Performance Management: Day 4
Nov. 27th, 2008 07:49 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
sawylThe day began with an attempt to solve the appalling network performance problems we encountered during yesterday's lab. Poking around we quickly noticed that although the operating system reported the adapters as being gigabit ethernet, the media settings had actually been set to 100Mbit with incorrect duplexing settings, so we dropped the interfaces, chdev'ed the devices to auto-negociate, brought the interfaces back up and found that the performance had improved by three orders of magnitude.
With that glorious triumph behind us, we looked at various IP layer tunings, how to manage fragmentation, and when to change the sizes and timeouts on the adapter queues. We then spent some time looking at NFS optimisation — surely the first optimisation is not to use NFS! — and some of the parameters that can be altered to give improved performance. After a brief lab which examined the differences in performance between NFS v2 and v3, we looked at some of the standard information gathering tools, such as PerfPMR — which, perversely, is no longer bundled into the base install — and the various system trace utilities.
All of which means that we're well on course to finish by midday tomorrow.
With that glorious triumph behind us, we looked at various IP layer tunings, how to manage fragmentation, and when to change the sizes and timeouts on the adapter queues. We then spent some time looking at NFS optimisation — surely the first optimisation is not to use NFS! — and some of the parameters that can be altered to give improved performance. After a brief lab which examined the differences in performance between NFS v2 and v3, we looked at some of the standard information gathering tools, such as PerfPMR — which, perversely, is no longer bundled into the base install — and the various system trace utilities.
All of which means that we're well on course to finish by midday tomorrow.
