I have observed the same thing on several Linux boxes. It seems to be a result of the changes made to the virtual memory manager code in the mid 2.4 kernel series when using multiple processors. 2.4.7 sounds like about the time it started, but I don't remember exactly. We actually killed kerneld on a few machines and ran without swap for quite awhile (they did have 2 - 4 gigs of ram in them, so they didn't really need to swap). The bugs were ironed out of the kernel code sometime later (I think early or mid teens) but are definitely fixed in the last few versions (2.4.18 and 2.4.19), so I would recommend installing one of those. Look in the changelogs on www.kernel.org if you need to know the exact versions to use or avoid. It wasn't a distribution- specific problem, we had both Red Hat and Debian boxes with the same problem, and a new kernel was the cure for both. Amanda seems to bring out the problem since there are multiple processes simultaneously trying to allocate rather large buffers.
Good luck, Frank --On Tuesday, October 08, 2002 02:37:49 +0000 [EMAIL PROTECTED] wrote: > My backups have recently slowed down... > > It is very peculiar and just started recently what seems to be out of > the blue. The network clients are backing up at full speed but the file > systems on the dump-host itself have slowed to about one quarter of > their normal speed. I am using dump (as opposed to tar) and what I have > observed is that a top shows kswapd and kreclaimd using 80% to 100% of > the CPU when dump is running on the dump-host. > > I am running a dual 1Ghz PIII with 1G ram on the dump-host and top shows > little (or no) swap space being used (though all of memory is used, but > Linux does that no matter what). I am running redhat linux 7.2 with kernel: > > 2.4.7-10enterprise #1 SMP Thu Sep 6 16:48:20 EDT 2001 i686 unknown > > I am running: > > amanda-2.4.2p2-4.i386.rpm > dump 0.4b31 (which is the latest dump according to: http://dump.sourceforge.net) > > I can't figure out what is causing kswapd and kreclaimd to go crazy when > the dump program runs, and it only happens on the dump-host. The other > 5 remote systems are all running Linux 7.2 or 7.3 and dump doesn't cause > them any problems. I.e. kswapd doesn't even show up in a top on the > remote machines when dump is running. > > Has anyone else ever experienced anything like this? > > Any suggestions what I can try to fix it? (Please don't suggest I use > tar.) :) > > Let me know if I can provide any other details of interest. > > Thanks, > Dick > > P.S. > > A sample of the output from top exhibits the problem: > > 8:24pm up 1:19, 2 users, load average: 3.19, 2.88, 2.46 > 110 processes: 108 sleeping, 2 running, 0 zombie, 0 stopped > CPU0 states: 0.0% user, 83.0% system, 0.0% nice, 16.0% idle > CPU1 states: 1.0% user, 92.0% system, 0.0% nice, 5.0% idle > Mem: 1028364K av, 1020680K used, 7684K free, 74916K shrd, 674280K buff > Swap: 2104464K av, 1144K used, 2103320K free 140432K cached > > PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME COMMAND > 5 root 19 0 0 0 0 SW 99.9 0.0 16:20 kswapd > 6 root 18 0 0 0 0 SW 88.4 0.0 7:45 kreclaimd > 2910 amanda 9 0 1320 1284 600 D 14.8 0.1 4:07 dump > 2970 rwk 13 0 1084 1080 832 R 4.1 0.1 0:00 top > 1 root 9 0 520 520 452 S 0.0 0.0 0:08 init > 2 root 9 0 0 0 0 SW 0.0 0.0 0:00 keventd -- Frank Smith [EMAIL PROTECTED] Systems Administrator Voice: 512-374-4673 Hoover's Online Fax: 512-374-4501
