iowait

[sherimm]

the iowait is flucutaing between 30 to 90%
below is the top output
22:18:04 up 2 days, 23:58, 10 users, load average: 39.95, 47.00, 42.19
979 processes: 978 sleeping, 1 running, 0 zombie, 0 stopped
CPU states: cpu user nice system irq softirq iowait idle
total 35.7% 0.0% 11.3% 0.2% 0.2% 50.3% 2.1%
cpu00 53.8% 0.0% 23.9% 0.0% 0.0% 22.2% 0.0%
cpu01 41.8% 0.0% 8.5% 0.0% 0.0% 49.5% 0.0%
cpu02 34.1% 0.0% 4.2% 0.8% 0.0% 52.1% 8.5%
cpu03 12.9% 0.0% 8.6% 0.0% 0.8% 77.5% 0.0%
Mem: 16418528k av, 16403864k used, 14664k free, 0k shrd, 2484k buff
2710772k actv, 404952k in_d, 158272k in_c
Swap: 20146864k av, 1157912k used, 18988952k free 2477076k cached

PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME CPU COMMAND
530 oracle 18 0 416M 415M 412M S 4.2 2.5 0:16 3 oracle
6471 oracle 17 0 410M 409M 406M S 3.4 2.5 2:40 1 oracle
14803 root 20 0 1908 1908 900 R 3.4 0.0 0:00 0 top
4697 oracle 15 0 624 624 252 D 2.9 0.0 1:58 3 compress
5846 oracle 15 0 388M 386M 384M D 2.7 2.4 131:06 2 oracle
12210 oracle 19 0 419M 418M 416M S 2.7 2.6 0:04 1 oracle
27741 oracle 17 0 410M 409M 406M S 1.7 2.5 5:38 1 oracle
5698 oracle 19 0 413M 413M 411M S 1.7 2.5 0:36 3 oracle
8590 oracle 16 0 410M 409M 406M S 1.4 2.5 8:16 3 oracle
11061 oracle 16 0 411M 409M 406M S 1.4 2.5 5:11 3 oracle
25678 oracle 16 0 410M 410M 407M S 1.4 2.5 3:06 3 oracle


It is casung the performance issues
could u pls suggest what can be done to get the iowait reduced

[Vitaly Karasik_1]

it is typical for DB servers with many disk I/O.
And, if you don't have performance issues, this situation is OK.


What can be done:
1) tune linux for Oracle using Oracle or Puschitz docswww.oracle.com/technologies/linux/index.html
www.puschitz.com/OracleOnLinux.shtml

2) tune Oracle for use more RAM

3) tune your SQL code

4) tune and/or rebuild your disk storage for performance.

[Stuart Browne]

980 processes, fully using 16GB of memory, with only 2GB as cache, using 1GB of swap..

Well, it sounds like the machine is a bit over-tuned.

Reign in the memory that Oracle is using by about 2GB, stop the machine from swapping.

Also, what IO subsystems do you have in this machine? I'm going to assume you've got RAID somehow, but are we talking software RAID over local ATA disks (I bloody well hope not!), or are we talking local-SCSI/SAS, or on NAS or SAN?

The physical delays of writing over NFS or some such is bad, so hopefully you aren't using that for the database storage. But what else?

Kernel revision will help us too.

If all else fails, I see a 'compress' process running there. Try to avoid doing things like that on the primary Database server, off load them to a secondary/tertiary server which is less mission critical.