Random high latency with KSM enabled and multiple CPU sockets

[Lawrence Dudley]

If running a box with multiple CPU sockets and Ubuntu 14.04 Server 64-bit qemu-kvm with KSM enabled as by default, random high latency occurs on network connections for VMs, as high as 4000ms - 5000ms. The host also reports high IO wait times.

Disabling KSM fixed this for me.

I run the same setup on boxes with only 1 CPU socket and it's not an issue.

qemu-kvm version: Installed: 2.0.0+dfsg-2ubuntu1.3

[Serge Hallyn (serge-hallyn)]

 affects: linux

[Lawrence Dudley (m-lawrence)]

I should add that to fix this while running you need to do, as root:

echo 2 > /sys/kernel/mm/ksm/run

This makes the pages duplicate, might take a second or two and disables KSM.

[Brad Figg (brad-figg)]

This bug is missing log files that will aid in diagnosing the problem. From a terminal window please run:

apport-collect 1368439

and then change the status of the bug to 'Confirmed'.

If, due to the nature of the issue you have encountered, you are unable to run this command, please add a comment stating that fact and change the bug status to 'Confirmed'.

This change has been made by an automated script, maintained by the Ubuntu Kernel Team.

[Lawrence Dudley (m-lawrence)]

I've looked through all the logs that are relevant and can't find any specific output that's relevant - marking as confirmed.

If you've got any logs in specific you need me to pull out I can do so.

[Lawrence Dudley (m-lawrence)]

Wait, someone else needs to mark it as confirmed.

[Serge Hallyn (serge-hallyn)]

Is this still an issue for you?

There were some actual KSM corruption bugs which were somewhat recently fixed.

KSM in general wil use more CPU in order to reduce memory contention, so your results are not necessarily unexpected. Aside from turning KSM off, you can also reduce the KSM frequency.

There are some other tuning hints at http://www.linux-kvm.org/page/Tuning_Kernel

[Launchpad Janitor (janitor)]

[Expired for qemu-kvm (Ubuntu) because there has been no activity for 60 days.]

[Launchpad Janitor (janitor)]

[Expired for linux (Ubuntu) because there has been no activity for 60 days.]