I/O daemons hindering system performance

SssupperDo0d used Ask the Experts™
It appears that whenever an I/O intensive process kicks off, kswapd, kjournald, kreclaimd, raid5d, and bdflush start hogging CPU time and in turn slowing other active processes down.  I realize that these daemons are necessary, but they seem to be unusually active lately.

I was hoping you experts would be shed some light on my situation to gain back the performance.

- RedHat 7.3
- 2.4 SMP packaged with the install
- Dual PIII 1gHz
- 4gigs RAM
- 2gigs swap

LABEL=/      /         ext3    defaults        1 1
LABEL=/boot  /boot     ext3    defaults        1 2
none         /dev/pts  devpts  gid=5,mode=620  0 0
LABEL=/logs  /logs     ext3    defaults        1 2
LABEL=/opt   /opt      ext3    defaults        1 2
none         /proc     proc    defaults        0 0
none         /dev/shm  tmpfs   defaults        0 0
/dev/hdd1    swap      swap    defaults        0 0
/dev/hdb1    /wtback   ext2    defaults        1 2
/dev/hdc1    /archive  ext3    defaults        1 2

Let me know if more info is needed.
Watch Question

Do more with

Expert Office
EXPERT OFFICE® is a registered trademark of EXPERTS EXCHANGE®
Top Expert 2005
For any sort of IO intensive activity it would be normal to see activity from kjournald (on ext3 file systems) and bdflush. If you have any RAID5 volumes configured and used in the IO operation that daemon would be active also.

However, I wouldn't ordinarily expect to see kswapd and kreclaimd active just for IO unless the system is memory hungry. The increase in activity could be a result of increased memory use on the system (what does top have to say?). Even a moderate amount of swapping will have a noticable affect of system performance and obviously those two daemons would be active if swapping is occuring.

Is the box "up to date" w/respect to the 7.3 errata? The current kernel from the errata would be kernel-smp-2.4.18-18.7.x.
Mihai BarbosTrying to tame bits. They're nasty.
There could be also another problem. When a filesystem gets quite full, its performance under heavy I/O degrades considerably. With IDE disks and many concurent tasks this situation gets even worse.
This old question needs to be finalized -- accept an answer, split points, or get a refund.  For information on your options, please click here-> http:/help/closing.jsp#1 
Post your closing recommendations!  No comment means you don't care.

Do more with

Expert Office
Submit tech questions to Ask the Experts™ at any time to receive solutions, advice, and new ideas from leading industry professionals.

Start 7-Day Free Trial