Good day.
During 'playing' with XCP in test environment, found very severe issues
with IO - during random writing constant flow or read requests are
issued, halving performance of block device.
Condiditions:
LVMoISCSI SR, disk array exported in blockio mode (no caching).
PV guest, VDI is type=vhd.
Before do any tests disk is fully initialized by dd if=/dev/zero
of=/dev/xvdb bs=2M
VDI size is 64G. Test area is about 1G size. Block size for random IO is
4k. Test done by fio, but same problems was with any type of writing.
FIO config is at the end of the message.
During IO on some areas of VDI (f.e. offset=16G) load monitor on storage
shows read requests with almost same frequency as write requests. Size
of read request is 512 bytes, so it looks like bitmaps (which resides in
512 area at the every 2M block in VHD format).
Some offsets show that read requests (repeatedly), some don't.
Question is: why VHD is rereading those data again and again? Why this
happens not at every offset?
fio config:
[test]
blocksize=4k
filename=/dev/xvdb
size=1G
rw=randwrite
direct=1
buffered=0
ioengine=libaio
iodepth=32
offset=16G
XCP is 1.1; storage is simple linux box with iet (tested with stst too)
and software raid10.
_______________________________________________
Xen-api mailing list
[email protected]
http://lists.xen.org/cgi-bin/mailman/listinfo/xen-api
