CentOS - Jun 2006 - SAR

Folks,

    At what point in iowait should I start to worry about having a 
bottleneck, or is this something that can't be answerd with a single 
integer?  According to sar, after my last reboot to turn off 
hyperthreading as a test, at one time, I see 4.9% iowait, but then one 
minute later, it droped back to 0.01%, and rarely even gets to 1.0%, at 
least what I remember from yesterday.  ?

thanks....

Sam

On Wed, 2006-06-07 at 13:07 -0400, Sam Drinkard wrote:
> Folks,
> 
>     At what point in iowait should I start to worry about having a 
> bottleneck, or is this something that can't be answerd with a single 
> integer?  According to sar, after my last reboot to turn off 
> hyperthreading as a test, at one time, I see 4.9% iowait, but then one 
> minute later, it droped back to 0.01%, and rarely even gets to 1.0%, at 
> least what I remember from yesterday.  ?
> 
> thanks....
> 
> Sam
He-he! The age old conundrum of tuners everywhere. What's appropriate
for observed performance X? Of course, first query leaves untold all the
things that are needed to get a SWAG (Scientific Wild-Ass Guess).

I/O wait is, of course, a result of many things: HD speed, cache size,
bus speed, latency, I/O request rate *and* duration of "flooding" I/O
requests, CPU time available to process it..

Given what you mentioned in the "I can't reconcile..." and
"swap"
threads, I would guess (not knowing your hardware or anymore about your
load) that you could sleep well if your peak I/O wait is 4.9% for short
periods. If it is usually down at 1% or less, smoke a joint: nothing to
worry about!

IMO, in complete ignorance.

Look at other SAR areas to see the things that affect that and you will
know if it's reasonable, as in "I really can't do much to improve
it and
it's too small to justify the effort anyway" or ... not.

Let's see. If you stayed consistently at 4% for 24 hours, should cost
you about 57 minutes. From there you can figure if it's worth any
effort.
> <snippo>
HTH
-- 
Bill
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL:
<http://lists.centos.org/pipermail/centos/attachments/20060607/8af5a847/attachment-0002.sig>

On Wed, 2006-06-07 at 13:07 -0400, Sam Drinkard wrote:
> Folks,
> 
>     At what point in iowait should I start to worry about having a 
> bottleneck, or is this something that can't be answerd with a single 
> integer?  According to sar, after my last reboot to turn off 
> hyperthreading as a test, at one time, I see 4.9% iowait, but then one 
> minute later, it droped back to 0.01%, and rarely even gets to 1.0%, at 
> least what I remember from yesterday.  ?
Generally iowait should be your bottleneck since it is the most
expensive thing to fix.  Except for special cases like number
crunching in RAM and sometimes graphics, your disk and network
i/o are the slow parts. 

-- 
  Les Mikesell
    lesmikesell at gmail.com