Bug 1038528

Summary:	IOPS is poor with ahci driver
Product:	Red Hat Enterprise Linux 7	Reporter:	Xiaomei Gao <xigao>
Component:	qemu-kvm-rhev	Assignee:	John Snow <jsnow>
Status:	CLOSED WONTFIX	QA Contact:	Yanhui Ma <yama>
Severity:	medium	Docs Contact:
Priority:	low
Version:	7.0	CC:	hhuang, jsnow, juzhang, knoel, kwolf, michen, qzhang, rbalakri, virt-bugs, virt-maint, wquan, yama, yuhuang
Target Milestone:	rc
Target Release:	---
Hardware:	x86_64
OS:	Linux
Whiteboard:
Fixed In Version:		Doc Type:	If docs needed, set a value
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2017-11-16 22:14:26 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:

Comment 3 John Snow 2014-07-17 18:43:45 UTC

AHCI support in QEMU is still tech-preview and is not feature-complete ... There are still several issues to solve prior to worrying about IO performance; in particular at least #887844 and #953062 and #901631.

There are several other features and bugs not yet filed that contribute to AHCI not being able to move beyond a tech preview status and as such any optimizations would be premature since a major refactor might be in the horizon; but keeping this issue in mind for those refactors will be helpful.

This is probably 7.2 timeframe at least, though.

Comment 5 John Snow 2015-06-25 16:20:19 UTC

One more time: AHCI support was recently finalized upstream and there's not enough time to do performance analysis for 7.2, kicking the can for hopefully the last time.

Comment 9 juzhang 2016-05-17 01:10:52 UTC

Hi Yanhui,

Could you handle this issue?

Best Regards,
Junyi

Comment 15 John Snow 2017-11-16 22:14:26 UTC

AHCI has never really been prioritized for performance improvements. Our official recommend has been that users seeking performance should use virtio-blk-scsi or virtio-blk-pci; that AHCI is perhaps slower than IDE is not of tremendous concern.

That said, if *severe* performance impediments can be observed, I'd like to fix them -- but I should stop pretending it's a priority for the enterprise product as *ATA devices are generally used only to install the OS, and then most management solutions clone their VMs from there.

ATA interfaces are hardly used for our downstream product, and that's the way it ought to be.

As such, I'm going to close this bug for now as "WONTFIX," but if you find anything obviously abhorrently wrong or prohibitively slow, please reopen and I will investigate it.

Further, if anyone manages to find this bug via google and you have good evidence for exactly _what_ is slow, or have new information to contribute, or would like this fixed upstream, please email me and cc the qemu-block list upstream and we will work on resolving this in the FOSS version of this package.

Thanks,
John