Isn't there a SPECWeb test or something like that for this? I believe they cost $$$ though. You could also do a capture and replay it back to your web server.
There must be some benchmarking tools or simulation methodologies. brian On Wed, Apr 14, 2010 at 11:56:03PM -0700, Bill Broadley wrote: > On 04/14/2010 10:51 PM, Brian Lavender wrote: > > I would think that you need to tune your filesystem. How about this > > article on "On-demand readahead" > > > > http://lwn.net/Articles/235164/ > > Ah, I found this: > http://www.redhat.com/magazine/001nov04/features/vm/#tuning-vm > > What I'd recommend is pick a design point. Taking this graph: > http://broadley.org/bill/random-read.png > > Then the number of clients you plan to have simultaneously, set > min_readahead so you can have a reasonable memory footprint and good > performance. > > Say you only had a single few year old single 1TB disk (like in the > graph). If you set min_readahead to 2MB the I/O system should be able > to provide 60MB/sec. > > You should be able to manage 60MB/sec, the default on my disk was 128KB, > which would lead to a bottleneck 12MB/sec or so. > > Next time my mirror gets busy (likely around the LTS release) I'll see > if I can collect some real world numbers. > > _______________________________________________ > vox-tech mailing list > vox-tech@lists.lugod.org > http://lists.lugod.org/mailman/listinfo/vox-tech -- Brian Lavender http://www.brie.com/brian/ "Those that fail to learn from history, are doomed to repeat it." - Winston Churchill _______________________________________________ vox-tech mailing list vox-tech@lists.lugod.org http://lists.lugod.org/mailman/listinfo/vox-tech
