El Tue, 16-05-2006 a las 21:11 +0000, Olivier Samyn escribió: > Alvaro you're right, the licence do not permit you to publish to third > parties Zeus tests.
Continuing with the discussion on this topic I'm posting here some quick notes I took while talking with Alvaro in Mexico: PROPOSED SERVERS: - Apache - Apache2 - Cherokee - Lighttpd - thttpd - boa Zeus should be tested in private, others might be interesting but we don't really want those guys in the picture in order to avoid gratuitous publicity. WHAT TO TEST: - Speed at serving pages - Resource usage Most benchmark concentrate almost exclusively on speed statistics. I believe that's something really important in many cases, but in everyday life you tend to be more interested in efficiency so you don't waste resource and money to get the same job done. I'm not sure how we could measure this but perhaps something could be done considering ratios like hits served by megabyte of memory or hits served by CPU usage. Ideas here are more than welcomed. HOW TO TEST: a) Plain files: - Small (10K) - Medium (1 MB) - Big (700 MB) Ideas here? Looks like we could use an image, some tarball or PDF and an ISO (Ubuntu Dapper when released in a few days could we used I guess) I'm not sure if the exact files used is a relevant thing but i'd like to make used of well known files people might have themselves in the case they want to reproduce the benchmark themselves. b) PHP: - PHP info - Computation-intensive PHP script (ideas welcomed) - Memory-intensive PHP script (ideas welcomed) The exact scripts should perhaps written by ourselves for this specific benchmark and published as part of it. Any other ideas of how the PHP tests could be done? c) Ruby On Rails - ????? - ????? - ????? I'm a little bit clueless about how to test it but I believe it's important to test Rails and not just Ruby since most people will be using Rails apps being them their own code or public like perhaps the type blog engine. Perhaps that's what we could use in the tests, an specific release of the Typo blog engine serving a set of post made by ourselves and published as a mysql dump for people to reproduce the experiment. Let's keep in mind that Lighty has got tons of eyeballs mostly exclusively by being the webserver used by Rails hackers. If we really want Cherokee to take of in terms of usage and start earning the wide respect it deserves I believe we should try hard to provide first-class experiences to groups of users like those running Ruby On Rails or PHP. They should get even better support, info and community they get from the competition but well, that's a vision for now. Regarding hardware, I recall Alvaro mentioning (while we were having some "tortas" a.k.a sandwichs) Cherokee would benefit the most if the playfield was a huge box with many processors so it takes advantages of the SMP code where most competitors won't. What do you think? Anybody said Sun hardware? ;) Does anybody has a box like that? I guess i'll be researching a bit the web on how to do great benchmarks, any references are welcomed. Antonio Lima-Peru. _______________________________________________ Cherokee mailing list [email protected] http://www.0x50.org/cgi-bin/mailman/listinfo/cherokee
