On Fri, 2004-07-23 at 06:28, Craig Ringer wrote: > On Fri, 2004-07-23 at 20:07, sarab fadhil wrote: > > [snip detailed system and user information] > > Thanks for that info. It's really nice not to have to ask for it :-) and > I'm sure it'll help the folks here figure out your problem. > > > My problem is with the login which is slow, > > it may take 10 minutes for all 32 students to login > > together, of course some of them will login faster > > than the others. > > That's rather odd. Does a single, isolated user login on the system when > it's otherwise unloaded take a more reasonable amount of time? > > > However, once every one login the > > system speed is acceptable, of course loading Star > > Office is a slow even in a stand alone computer. > > Loading the second and subsequent OpenOffice 1.1.2 instances on my LTSP > server often takes only 2 or 3 seconds. The first one takes up to 10 > seconds, depending on whether or not the binaries are still cached in > RAM. I had to turn off the splash screen because it sometimes only got > time to half-finish drawing before the app's main window came up. So I > wouldn't think that slow loading times are something to expect. The > second and further OpenOffice/StartOffice instances should AFAIK load > quite quickly if you have enough RAM and sufficient network bandwidth. > > > I am > > using KDE as a user interface. Any advice of how to > > speed up the system is appreciated. > > The first step will probably be to start measuring what is taking the > time. I'd recommend running 'vmstat 1 | tee /tmp/vmstat_log' while the > users log in, as this provides some useful info on how hard the server > is working. Additionally, I'd fire up gkrellm, a handy system monitor > tool, to get a visual impression, and I'd have 'top' running to get > real-time summaries of what programs are using the memory and CPU time. > > It'd also be helpful to run 'free -m' before and after all the users log > in (and a few times during, if you like) to get an impression of what > sort of pressure the memory is under. > > If you could run > ps -eo pid,ppid,user,state,size,rss,wchan,cmd > before, during and after user logins (and save the output somewhere) > that'd be useful too, as that will among other things show if processes > are in uninterruptable sleep (which is usually because they're waiting > on disk I/O). > > If you upload the vmstat output, the before and after 'free' output, ps > output, and any other monitoring info you think could be handy to a web > page and post your impressions about what the server seemed to be doing > during the login process, that'd be handy. Uploading a copy of the > output of 'dmesg' and a snippet of /var/log/messages that covers the > duration > Have you checked to make sure there isn't of the user login process > wouldn't hurt, either. > > Of course, it's entirely possible that someone will be able to pop up > here and say "yeah, I've seen this...." - but in case they can't, it'd > probably be a good idea to collect this information. It's very hard to > diagnose general performance problems without at least that information. > > I should note that the use of a 7200rpm SATA disk, presumably one > designed for general PC use, will not be helping your system performance > at all. Even if each client's apps aren't trying to do much on it, those > disks just aren't designed for heavy multitasking and it'll probably be > thrashing like crazy. I run RAID 5 across 4 250GB SATA disks here, and > the performance is pretty poor (but sufficient for our purposes, as it's > mostly write-once read-occasionally archival data). AFAIK current > 'consumer' disks and SATA chipsets don't support tagged command queuing, > which means the disks have considerably less ability to reorder requests > in a way that's more efficient for their physical layout, so they'll > waste more time seeking - especially under heavy multi-user loads. There > are now 10k RPM SATA disks with TCQ support available, but controller > support for TCQ is far from universal and the disks are pricey. Still, > they're cheaper (and according to StorageReview, often faster) than many > SCSI disks. > > Are your disks connected to a SATA RAID controller like a 3ware > Escalade, to an add-in SATA card, or to SATA connectors on the > motherboard? What chipset and SATA driver are you using? > > The performance measures I've requested may help point out whether or > not the issue you're seeing is disk related, memory related, or > something else. ---- completely awesome mini-primer on system evaluation.
I am thankful to have this and will save it for the right moment. Craig ------------------------------------------------------- This SF.Net email is sponsored by BEA Weblogic Workshop FREE Java Enterprise J2EE developer tools! Get your free copy of BEA WebLogic Workshop 8.1 today. http://ads.osdn.com/?ad_id=4721&alloc_id=10040&op=click _____________________________________________________________________ Ltsp-discuss mailing list. To un-subscribe, or change prefs, goto: https://lists.sourceforge.net/lists/listinfo/ltsp-discuss For additional LTSP help, try #ltsp channel on irc.freenode.net