Dear colleagues,
I've found why guest's storage performance was much less than host's
performance (the mistake was too stupid to tell about it, really).
But I'd like to ask one more question if you don't mind. :) I played with
various KVM options (cache, io and so on...) and now I can say that I got most
IOpS with cache=writeback. Tests are being performed with the fio utility, here
is the jobfile contents:
[readtest]
blocksize=4k
filename=/dev/vdb
rw=randread
direct=1
buffered=0
ioengine=libaio
iodepth=32
size=20%
[writetest]
blocksize=4k
filename=/dev/vdb
rw=randwrite
direct=1
buffered=0
ioengine=libaio
iodepth=32
size=20%
Is there any way to make ACS run create KVM domains with cache=writeback
instead of cache=none without patching the source code?
Thank you!
--
V.Melnik
