Bug 893327

Summary:	virtio_scsi performance 20%+ worse than virtio_blk in some scenarios
Product:	Red Hat Enterprise Linux 6	Reporter:	Xiaomei Gao <xigao>
Component:	qemu-kvm	Assignee:	Fam Zheng <famz>
Status:	CLOSED DEFERRED	QA Contact:	Virtualization Bugs <virt-bugs>
Severity:	medium	Docs Contact:
Priority:	high
Version:	6.4	CC:	akong, areis, bmcclain, bsarathy, chayang, famz, juzhang, kwolf, lyarwood, michen, mkenneth, pablo.iranzo, pbonzini, qzhang, rbalakri, tcarlin, tvvcox, virt-maint, wquan, xigao
Target Milestone:	rc
Target Release:	---
Hardware:	x86_64
OS:	Linux
Whiteboard:
Fixed In Version:		Doc Type:	Bug Fix
Doc Text:		Story Points:	---
Clone Of:
Clones:	999304 (view as bug list)		Environment:
Last Closed:	2014-12-15 09:42:18 UTC	Type:	Bug
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:
Bug Depends On:	1106420
Bug Blocks:	999304, 1002699

Comment 4 Paolo Bonzini 2013-10-24 15:26:26 UTC

The main difference here is that vcpu0 is always at 100% in the virtio-scsi tests.  The load is much more balanced between vcpu0 and vcpu1 for virtio-blk.

This applies to both qcow2 and raw actually, but we only see worse performance from it in qcow2.  I think we should first analyze/fix this fairness issue to see whether it affects performance, because it's "weird".

Comment 5 Amos Kong 2013-10-29 03:18:29 UTC

Hi xgao,

can you list the host numanode info and the pin setup info?

Comment 6 Xiaomei Gao 2013-10-30 01:56:18 UTC

(In reply to Amos Kong from comment #5)
> Hi xgao,
> 
> can you list the host numanode info and the pin setup info?

Irqblance is running on both host and guest, we didn't do any pin setup on host and guest.

[root@hp-z800-06 ~]# numactl --hardware
available: 2 nodes (0-1)
node 0 cpus: 0 1 2 3
node 0 size: 8175 MB
node 0 free: 7287 MB
node 1 cpus: 4 5 6 7
node 1 size: 8192 MB
node 1 free: 7906 MB
node distances:
node   0   1 
  0:  10  20 
  1:  20  10

Comment 9 Xiaomei Gao 2014-04-03 02:42:39 UTC

(In reply to Fam Zheng from comment #8)
> Xiaomei,
> 
> This bug has been around for a while and I know we are switching to fio.
> Does latest performance tests have such vcpu fairness measurement (e.g. on
> 6.5, 6.6 and new qemu-kvm)? Can you help confirm if the unfairness is still
> observed?

Okay, We will test the latest 6.6 qemu-kvm and see if the issue still happened. We will update the comment once fresh results are on hand.

Comment 10 Ademar Reis 2014-05-28 11:52:15 UTC

(In reply to Xiaomei Gao from comment #9)
> (In reply to Fam Zheng from comment #8)
> > Xiaomei,
> > 
> > This bug has been around for a while and I know we are switching to fio.
> > Does latest performance tests have such vcpu fairness measurement (e.g. on
> > 6.5, 6.6 and new qemu-kvm)? Can you help confirm if the unfairness is still
> > observed?
> 
> Okay, We will test the latest 6.6 qemu-kvm and see if the issue still
> happened. We will update the comment once fresh results are on hand.

Keeping the needinfo until you have the test results. Thanks.