begin quoting Tracy R Reed as of Fri, Feb 16, 2007 at 12:12:04AM -0800: > Stewart Stremler wrote: > > Have we run any profiling tools on the existing server to determine > > where the real bottlenecks are, rather than speculating? > > I've been running iostat and vmstat. The big problem I see is cpu. CPU > usage goes to 100% during a page load and stays there for a few seconds. > I think that is where most of our delay comes from. Occasionally we see > some swapping also. It is 289M into swap and has 512M of RAM.
For comparison, I have 1.5G of RAM and I'm 770M into swap, and I'm a relatively unloaded desktop machine. But I think what's swapped out is mostly those virtual desktops that I haven't hit in awhile... > > A new server would be okay. What are our funds like? > > As far as I know we have thousands and nothing to spend it on. Except > perhaps a server. That's what I'm thinking. > > Going uber-cheap-but-better-than-what-we-have-now is nice, and > > spending decent money for a really good machine is nice; I'm not > > so much happy with a middle-of-the-road approach, however. Unlike > > James, I might be happier with $2k spent on a server than $1k. > > I could go either way. All the box really does is serve the webpage and > mailing lists. I would definitely like reliability thus the mirrored disks. Do we have a backup scheme for sparky? Something like an rsync to various offsite machines? I'd rather not depend on mirroring if it is important enough to spend the time, money, and effort setting it up. Reliability is very good -- not just disk (redundancy makes that easy), but for everything else, too. > > [1] No idea if this is a processor, bus, disk, or software problem. > > Processor, then memory, then disk. And the software could be optimized a > bit also but not really within the current memory constraints. Excess memory is used to cache disk, so I'd think erring on the side of too much memory wouldn't be a problem. -- Unhappy that my blade 100 can only hold 2G of RAM. Stewart Stremler -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-steer
