Hi,
On Thu, Feb 24, 2005 at 06:58:31PM -0800, Brian Watters wrote:
Hello all,
Just wanting to check in and see if CentOS v3.4 final has found the fix to the issues that RedHat v3.0 ES has been having with very high CPU use on IOWAIT ??
There is lot of confusion around the whole IOWAIT-thingie. It's perfectly OK to have machine running 100% (or very close) I/O wait. These is nothing wrong in that if the disk I/O is still running as expected.
It's just CPU-time that is accounted as process waiting for I/O to finnish (it's seen as idle time on vanilla 2.4).
So do you actually have 'sluggish' I/O like speeds dropping cloose to nothing or just this I/O wait?
IF so what was the fix and or work around? .. We have lots of RH 3.0 servers that we are in the process of converting over to CentOS.
There is no extra patches on CentOS-3 kernels which would make it any better than the upsource kernel.
For the record, i've been iterating this 'iowait' a lot. My final closure for it was that ext3 just is broken with 3w-9xxx controller and was unbareable, so i pathed the XFS in to the kernel tree and has been much happier since that.
Another thing is that i was easily triggering this 'close to nothing for I/O speed' with ext3 on vanilla 2.4-seris too. Never with UP-kernel tho.
The kernels has been available like ages on
http://core.upi.iki.fi/out/kernel/
if someone (few has) would find those usable. This does account to the '3ware 9500S' thread too seen here lately - why do you fight with that when someone has patched it in already and even tested it for you? :P
The problem being tho that it's not available while installing .... It could be, but all the cloners seem to be so uptight to 'just pure clone' and not to make any enhancements over upstream source provider.
Anyway. I mostly do this kernel enhancing for parts I do need and share the results with whom might find those useable.