qemu-discuss
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

virtio-scsi ~10times slower on windows then IDE


From: Nikola Ciprich
Subject: virtio-scsi ~10times slower on windows then IDE
Date: Fri, 3 Jul 2020 11:15:15 +0200
User-agent: Mutt/1.5.21 (2010-09-15)

Hello dear QEMU users and developers.

While investigating very poor write performance on one of our customers
cluster, I've discovered strange issue.

They have a bunch of qemu-kvm virtual machines on top of ceph storage (NVME
and SATA pools). NVME write performance is OK, but SATA write speed is way
slower then it should be (~10MB/s seq write). I've done a lot of tests
and discovered following:

Windows (2019) write performance using virtio-scsi is very poor (~10MB/s)
Windows (2019) write performance on same volume, but using IDE is OK (>100MB/s)

Linux write performance is OK using both virtio-scsi and IDE.

machines are started using same setup, for virtio-scsi adapter, we use
same # of queues as #CPUS, io=native, discard enabled, caching disabled
(but trying different values of those didn't have almost any impact on
windows virtio-scsi performance)

Also tried different versions of virtio-scsi drivers, to no avail.

I've tried switching between virtio-scsi and IDE many times, and it's always
OK for IDE, always bad for virtio. But only for SATA pool volumes.

Host OS is centos 7, we're using ceph-14.2.8, I have this problem on both
qemu 2.12.1 and 4.2.1

I was able to reproduce the issue on one of our ceph clusters (while other one
with very similar setup performs well), so I'm free to play and investigate,
however I'm running out of ideas already..

anyone else seen this issue?

I'd be very grateful for any hints

BR

nik


-- 
-------------------------------------
Ing. Nikola CIPRICH
LinuxBox.cz, s.r.o.
28.rijna 168, 709 00 Ostrava

tel.:   +420 591 166 214
fax:    +420 596 621 273
mobil:  +420 777 093 799
www.linuxbox.cz

mobil servis: +420 737 238 656
email servis: servis@linuxbox.cz
-------------------------------------



reply via email to

[Prev in Thread] Current Thread [Next in Thread]