This is part of The Pile, a partial archive of some open source mailing lists and newsgroups.
Date: Thu, 30 Nov 2000 11:20:39 -0800 (PST) From: Rafael <raffi@linwin.com> To: mike rock <mrock@stewartsigns.com> Subject: Re: [svlug] slow system On Thu, 30 Nov 2000, mike rock wrote: > Thanks for resolving my issue with my home network,, that was a piece of > cake when compared to this next question. > > At work we have A Linux server that is primarily dedicated to being a > database server and two NT servers, one a Microsoft applications programs > server and the other an accounting dept server. These are accessed via 50 > or so NT workstations and one Linux work station(mine). > > There are times during each day that the system slows to a crawl and for > the life of me, I have been unable to determine why. I am capturing(every > 15 minutes) output from the following programs: > vmstat -n 2 6 > df > top -b -n1 > date (for reference) > > Looking at each of these reports and comparing the snapshots taken during > extreme slowdowns with other snapshots when the system was cookin,,reveals > nothing obvious except the load is very high at times,,like near 10. or > above. correspondingly, the cpu usage does not always seem to correlate > with the high load averages. The only correlation that I can see is the > elevated number of writen blocks that I see on the vmstat report. These can > go as high as 1 or 2 thousand and normally the number written out (bi) is in > the low three digits. > > Can anyone direct me to a means of isolating the source of the slowdowns? > I am beginning to suspect the RAID controller. The Linux box is an IBM > infinity 5000 withdual pentium 450's and Dual Raid 1 drives. We are > running an smp RH6.1 kernel. This is the system where we notice the > slowdown the most because 95% of all network traffic ends up there. > > If you can be of any assistance it will be appreciated, Generic Linux doesn't come with the best set of system status tools. You may want to download sysstat from Metalab in the following directory: ftp://metalab.unc.edu/pub/Linux/system/status/ sysstat home page is now at: http://perso.wanadoo.fr/sebastien.godard It compiles on RH6.2 without problems and will give you additional tools like sar, iostat, mpstat, etc. Also try procinfo. You did not tell which program takes most of the CPU time, database? I suspect problem with response on ethernet connection especialy if you have a lot of SQL, Samba or NFS requsts. Checked for network collisions, duplicate IPs? ===