Dont think it's related, because your problem is PHP-centric, but I think it could be worth it to somewhat add a "profiling driven optimisation" (http://jcayzac.brainlex.com/profiling.html) to Gentoo, since it's a source based distribution. It would require sysops to be able to run software inside WORKDIR and enhancing emerge --resume to actually do the second pass. Don't know if it's doable for most ebuilds... Julien.
On Sat, 05 Mar 2005 00:04:51 +0200, Jos Houtman <[EMAIL PROTECTED]> wrote: > Hello, > > Let me start with a short introduction since I am new to this list, and > you know what to expect of me. > I'am not new to linux, I have been using it as my desktop for over a > year, but I am fairly new to handling a 30+ serverpark which will grow > an estimated 3 times this year (thats if things are slow, and I am > lucky ;), any hints on keeping things manageble are appreciated but > thats maybe for a new discussion, if you have some good things please > start a new discussion to keep things clear. > > So far my introduction. > > I want to performance profile the servers, I want to know what causes > bottlenecks, which php script is taking ages, why are we waiting for IO, > which files are locked, what caused the lock, and which threads are > waiting for what lock. Are we waiting for the network, is the network > fast enough, is nfs being a pain? etc.. etc.. All questions i still > cannot answer. > > i have been working with iostat, vmstat, ps, lsof. But they all seem to > provide basic information, I am able to skim the surface of my problems, > but am unable to find the details i need to locate the problem/cause. > > An example of what happend a few weeks back. > Each webserver runs mysql, apache and php, and a couple of nfs mounts to > put uploaded and rendered images. > ones or twice a day some of these server 'locked up' unable to serve > http requests (no it was nog maxClient),but if we would give them some > peace then they recovered on there own. > After some research we found out that there was a good amount of time > spent waiting for I/O. a dozen orso apache threads had a flock status, > but for which file's? and which script was running?. All information i > was missing to pin point the problem. > After a hunch we moved the image rendering to some dedicated servers, > and this seemed to help, so we where lucky this time. > > But next time, new problems, new solution. > And I hope you guys can help me with new tools, new idea's, new > knowledge and a better understanding of whats server management/tweaking > is about and how to do it. > > With regards, > > Jos houtman > http://www.hyves.net/frm.php/58069/5mJl > Have a look, invite your friends its only fun if you have a network > > -- > [email protected] mailing list > > -- [email protected] mailing list
