Optimizing the QEMU/KVM Storage Stack

Session information has not yet been published for this event.

Scheduled: Friday, November 5, 2010 from 2:20 – 3:05pm in President's Ballroom

One Line Summary

Tools and recent progress in improving QEMU/KVM storage performance.


Virtualization has become a widespread tool to flexibly provision and consolidate machine instances. In the past, these benefits were traded off against lower performance than when running physical machines. This gap has been significantly narrowed in virtualized CPU and networking performance, but storage performance remains a relatively unexplored area.

This talk shows the performance landscape of QEMU/KVM storage and how its architecture handles disk I/O-bound workloads. QEMU supports a range of storage options that have very different characteristics. As a result, users are often surprised by the small configuration changes that can degrade or improve disk I/O throughput.

The results of recent work in improving QEMU/KVM storage performance are discussed. These optimizations provide insights into both the current status of the virtualized storage code path and where future bottlenecks lie.

Techniques and tools are presented which can be used to gather a picture of the entire virtualization stack and then zoom in on performance bottlenecks. These approaches can be applied to performance analysis beyond QEMU/KVM and are valuable to audiences both in the virtualization and other communities.


KVM, QEMU, performance, I/O virtualization, virtio

Presentation Materials



  • Stefanha

    Stefan Hajnoczi

    IBM Linux Technology Center


    Stefan Hajnoczi joined the IBM Linux Technology Center in 2010 where he works on QEMU/KVM and Linux virtualization. He previously worked on cross-platform virtualization at Transitive before it was acquired in 2008.

    Stefan is also involved in the Etherboot Project and works on the gPXE network bootloader, which provides an open source PXE implementation and takes network booting a step further with HTTP, iSCSI, and ATA-over-Ethernet support.

Leave a private comment to organizers about this proposal