Bug 508109 - qemu-kvm cpu usage spikes on F11 host with idle centos 5.3 guest
qemu-kvm cpu usage spikes on F11 host with idle centos 5.3 guest
Product: Fedora
Classification: Fedora
Component: qemu (Show other bugs)
All Linux
low Severity medium
: ---
: ---
Assigned To: Glauber Costa
Fedora Extras Quality Assurance
Depends On:
  Show dependency treegraph
Reported: 2009-06-25 12:49 EDT by Dennis Jacobfeuerborn
Modified: 2009-08-07 06:46 EDT (History)
6 users (show)

See Also:
Fixed In Version:
Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of:
Last Closed: 2009-08-07 06:46:10 EDT
Type: ---
Regression: ---
Mount Type: ---
Documentation: ---
Verified Versions:
Category: ---
oVirt Team: ---
RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: ---

Attachments (Terms of Use)
VM performance details screenshot (45.42 KB, image/png)
2009-06-25 12:49 EDT, Dennis Jacobfeuerborn
no flags Details
oprofile data (2.98 KB, text/plain)
2009-06-25 18:44 EDT, Dennis Jacobfeuerborn
no flags Details

  None (edit)
Description Dennis Jacobfeuerborn 2009-06-25 12:49:02 EDT
Created attachment 349426 [details]
VM performance details screenshot

When I run a centos 5.3 x86_64 guest on my F11 x86_64 system with kvm I see the qemu-kvm process cpu usage jump to ~20% (in top, ~50% in virtual-machine-manager) every few seconds even though the guest is completely idle (I even disabled the HAL polling for /dev/hdc). Running a "top" using the console the only process that show up as using cpu time is top itself.

I attached a screenshot which show the behaviour pretty well.

This is the command line the process is running with (started from virtual-machine-manager):

/usr/bin/qemu-kvm -S -M pc -m 512 -smp 1 -name centos5 -uuid 2754a928-6d0c-7833-e0e6-c717de291258 -monitor pty -pidfile /var/run/libvirt/qemu//centos5.pid -boot c -drive file=/var/lib/libvirt/images/centos5.img,if=ide,index=0,boot=on -net nic,macaddr=54:52:00:27:40:99,vlan=0 -net tap,fd=16,script=,vlan=0,ifname=vnet0 -serial pty -parallel none -usb -vnc

The VM is configured with 1 VCPU and 512mb ram. I also shut down the network interface in the guest using "ifdown eth0" but that didn't change anything.

The host system is a fresh F11 x86_64 install running on a system with a "AMD Athlon(tm) 64 X2 Dual Core Processor 4200+" CPU and 4gb of ram.

The kernel and qemu versions are:

Comment 1 Dennis Jacobfeuerborn 2009-06-25 18:44:15 EDT
Created attachment 349483 [details]
oprofile data

I attached some oprofile profiling data. I'm not very familiar with the proper usage of oprofile but maybe that data can help shed some light on the source of the problem.
Comment 2 Mark McLoughlin 2009-07-03 03:30:33 EDT
Hard to tell what's going on here

Does booting the guest with divider=10 help at all?
Comment 3 Dennis Jacobfeuerborn 2009-07-04 18:31:59 EDT
After firing up the VM today the major peaks on the performance graph no longer appeared, only the smaller ones. I updated the machine in the mean time so that might have something to do with it.

Adding "divider=10" further reduces both the value and the frequency of the minor peaks.
Comment 4 Mark McLoughlin 2009-08-07 06:46:10 EDT
Thanks for the update

This just sounds like normal behaviour - the guest may be idle, but it is still handling timer interrupts etc.

Closing, but do re-open if you think there's a genuine qemu/kvm bug here

Note You need to log in before you can comment on or make changes to this bug.