System has 10%+ iowait

[Matt Patterson]

I'm sorry to see that no one else has replied to your question. I have 
no direct knowledge of the problem, but I find myself asking similiarly 
unanswered questions all the time, so I figured I would throw out an 
idea or two. Have you looked at the performance of the system if it is 
writing/reading to a usb/firewire disk? Have you tried booting the 
single processor kernel? Also, have you tried the napic, nolapic, nacpi, 
acpi=off options for your kernel. Possibly you have some sort of irq 
controller conflict or issue, thats why I point to the kernel options. I 
would also look at the bios options and try any sata controller options 
and also look if the smp table version can be modified. I would also use 
hdparm to check out whether dma is enabled on the drives.

Also, just for kicks toss in a Knoppix cd and see if hard disk access 
works from there. If it does, copy the kernel!

I hope you find a solution!

Matt


James Andrewartha wrote:

> I've got a recently installed hoary (5.04) system that spends a lot of 
> its time in iowait. The specs are: HP Proliant DL 360, dual 3.4GHz 
> xeon with HT, 1GB ram, two Seagate 7200.7 160GB SATA drives in RAID 1. 
> It's running in EM64T mode, and is a general 
> web/mail/dns/spam/lists/gateway server. The workload of this server 
> was adequately handled by the dual p3 733 it replaced, so it's not 
> like I'm thrashing it.
>
> Basically, anything that touches the disk ends up in the D state for 
> several seconds, including things like kjournald, syslog and 
> postmaster (for postgres, for dspam). I've tried the default 2.6.10 
> kernel, and 2.6.11 from universe, and booting 2.6.10 with irqpoll, and 
> breaking the RAID, but none of this has an effect. The only thing I 
> can thing of is it's using ata_piix for the "Intel Corp. 6300ESB SATA" 
> controller, and that using ahci.ko instead might fix it, but I really 
> don't know.
>
> Some stats:
> vmstat 1:
>  0  2   4952  53992  25948 546584    0    0     0  1404 1377   120  0  
> 1 72 28
>  0  1   4952  54024  25948 546584    0    0    44   220 1382   171  0  
> 1 70 30
>  0  3   4952  53896  25960 546640    0    0    80  4900 1317   111  1  
> 1 63 35
> procs -----------memory---------- ---swap-- -----io---- --system-- 
> ----cpu----
>  r  b   swpd   free   buff  cache   si   so    bi    bo   in    cs us 
> sy id wa
>  2  3   4952  53896  25964 546636    0    0     4  1432 1419    75  0  
> 1 37 63
>  0  2   4952  53896  25972 546696    0    0     0   740 1415    64  0  
> 1 33 67
>  0  1   4952  54112  25980 546688    0    0    16   508 1311   178  1  
> 1 74 25
>  0  0   4952  55728  26132 547012    0    0   380   456 1178   373  0  
> 1 83 16
>  0  0   4952  56312  26132 547012    0    0     0     0 1062    53  0  
> 0 100  0
>  0  0   4952  56312  26132 547012    0    0     4     0 1061    63  0  
> 0 100  1
>
> And a graph of cpu usage (user,system,nice) and iowait:
> http://quoll.daa.com.au/stats/cpuusage.png
>
> Thanks for any suggestions,
>