> +1. Isn't that documented? I'm surprised! Suppose so, as a n00b though, I thought maybe cutting the default time of 15 in half would do it. It didn't. With many people on fast Internet connections now, even 1 or 2 second KeepAliveTimeout's seem to work good. > Interesting! Do you have any insights into how this varies > with filesystem and operating system/hardware behaviour? > If this comes close to the difference made by .htaccess, > I'd be looking at what lies beneath. Unfortunately the day we did this we also optimized a bunch of MySQL queries. That aside, and with the disclaimer that we're total newbies at file system optimization, we saw was about a 40% decrease in CPU time and (I have no idea why on this) about 20% more free RAM memory. We're on ext3 filesystem but it would seem like any gain you'd see from noatime would be impactful on any filesystem but I have no idea. I can say that of all the tuning we've done to the system, the noatime and the MinSpareServers produced the most immediate, "I can see a difference with my naked eye" type results. YMMV. The article below had convincing data which led us to try it but screwing with our filesystem isn't something we took lightly: http://www.philchen.com/2009/02/09/some-tuning-tips-for-apache-mod_cache-mod _disk_cache > Furthermore, what you describe appears to assume prefork MPM, > which is unlikely to be the best solution where performance > is an issue. Agreed, from what I've read worker MPM seems better but as you guys may know, if you want tech support from Red Hat on RHEL5, you're stuck with prefork. Otherwise it's recompile time (yes, all the books say everything should be recompiled) and if something goes haywire we're screwed with no tech support (albeit the tech support from Red Hat hasn't been very helpful anyway).
Attachment:
smime.p7s
Description: S/MIME cryptographic signature
|