Bug 1186914

Summary: RHEL6 qemu-kvm: backport cache=directsync
Product: Red Hat Enterprise Linux 6 Reporter: Ademar Reis <areis>
Component: qemu-kvmAssignee: Stefan Hajnoczi <stefanha>
Status: CLOSED ERRATA QA Contact: Virtualization Bugs <virt-bugs>
Severity: unspecified Docs Contact: Dayle Parker <dayleparker>
Priority: high    
Version: 6.6CC: chayang, jentrena, jherrman, juzhang, kwolf, mkenneth, pbonzini, pdwyer, qzhang, rbalakri, rpacheco, salmy, virt-maint, wquan, xigao
Target Milestone: rcKeywords: FutureFeature, Performance
Target Release: ---   
Hardware: Unspecified   
OS: Unspecified   
Whiteboard:
Fixed In Version: qemu-kvm-0.12.1.2-2.453.el6 Doc Type: Release Note
Doc Text:
qemu-kvm supports directsync cache mode on virtual disks With this update, qemu-kvm supports the "cache=directsync" option in the host file, which enables the use of the directsync cache mode on virtual disks. When "cache=directsync" is set on the virtual disk (configured in the guest XML or the virt-manager application), write operations on the virtual machine are only completed when data is safely on the disk. This increases data security during file transactions between virtual machines, and also improves performance by allowing I/O from the guest to bypass the host page cache.
Story Points: ---
Clone Of: Environment:
Last Closed: 2015-07-22 06:08:50 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:    
Bug Blocks: 1185250, 1198205    

Description Ademar Reis 2015-01-28 21:01:23 UTC
We want cache=directsync to improve performance in some customer scenarios. For more details, see https://bugzilla.redhat.com/show_bug.cgi?id=1185250#c20

Comment 1 Jeff Nelson 2015-02-18 17:54:00 UTC
Fix included in qemu-kvm-0.12.1.2-2.453.el6

Comment 3 Xiaomei Gao 2015-03-24 08:39:51 UTC
Verify the bug on kernel-2.6.32-545.el6.x86_64 and qemu-kvm-0.12.1.2-2.458.el6.x86_64, and here is the comparison of directsync and writethrough when performing multiple small writes on a file opened with O_DSYNC flag.

1. # for i in `seq 1 5`; do fio --ioengine=sync --readwrite=write --bs=1k --size=50m --fsync=1 --name=test --filename=/dev/vda | grep iops; done

cache=writethrough
  write: io=51200KB, bw=4288.5KB/s, iops=4288, runt= 11939msec
  write: io=51200KB, bw=4812.3KB/s, iops=4812, runt= 10640msec
  write: io=51200KB, bw=4853.9KB/s, iops=4853, runt= 10550msec
  write: io=51200KB, bw=4856.4KB/s, iops=4856, runt= 10543msec
  write: io=51200KB, bw=4855.4KB/s, iops=4855, runt= 10545msec

cache=directsync
  write: io=51200KB, bw=5734.8KB/s, iops=5734, runt=  8928msec
  write: io=51200KB, bw=5472.5KB/s, iops=5472, runt=  9356msec
  write: io=51200KB, bw=5449.8KB/s, iops=5449, runt=  9395msec
  write: io=51200KB, bw=4796.3KB/s, iops=4796, runt= 10675msec
  write: io=51200KB, bw=5573.1KB/s, iops=5573, runt=  9187msec

2. # for i in `seq 1 5`; do java Java2Disk /dev/vda 1000 100 rws; done

cache=writethrough
  Sync Test rws - msec to perform 1000= 212 SyncsPerSec= 4716.981
  Sync Test rws - msec to perform 1000= 213 SyncsPerSec= 4694.8354
  Sync Test rws - msec to perform 1000= 206 SyncsPerSec= 4854.369
  Sync Test rws - msec to perform 1000= 204 SyncsPerSec= 4901.961
  Sync Test rws - msec to perform 1000= 206 SyncsPerSec= 4854.369

cache=directsync
  Sync Test rws - msec to perform 1000= 159 SyncsPerSec= 6289.3086
  Sync Test rws - msec to perform 1000= 157 SyncsPerSec= 6369.427
  Sync Test rws - msec to perform 1000= 159 SyncsPerSec= 6289.3086
  Sync Test rws - msec to perform 1000= 155 SyncsPerSec= 6451.613
  Sync Test rws - msec to perform 1000= 158 SyncsPerSec= 6329.114


There is ~14.2% IOPS boost in scenario 1 and ~32.08% performance improvement in scenario 2 when performing multiple small writes on a file opened with O_DSYNC flag, so set the bug verified status. If there is something wrong, please correct it.

Comment 5 errata-xmlrpc 2015-07-22 06:08:50 UTC
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://rhn.redhat.com/errata/RHBA-2015-1275.html