I'm experimenting with image conversion on one of our ESXi hosts. When I do
a run, the I/O wait jumps to 99% and stays there. CPU etc stays
around 0%.
This seems really odd, and makes me think it's caused by vmware. Is
there anything you can do with the vmware configuration to improve this?
You can replicate the problem I'm seeing by doing this:
(using ImageMagick)
1. mkdir /tmp/delete_me/
2. convert -scale 640x480 ~home/12390-20090921.pdf /tmp/delete_me/
12390-scale.jpg
3. wait 5 seconds.
3. Run top. Look at the "wa" (IO wait) percentage, and compare it to
the other numbers, like Cpu.
The ESXi host is ESXi 3.5, HP DL360 G5, with 6 146gb 10k RPM SAS drives in a RAID5 configuration, 32GB of RAM.
It finishes fine. It just takes a long time. Seeoimg CPU at 0.1% and I/O wait at 100% is bizarre.
I've seen stuff like that when NFS goes down.
I suspect the RAID5 config is part of the problem.
I'm going to run a couple of tests and see if the ESXi host is being taxed. I doubt it is, even though we have a few hosts running on it that are mostly idle.
Thanks.