KVM I/O & Converged QEMU/KVM Gluster

A few articles have come out in the last few days RE: KVM I/O performance, and converged QEMU/KVM Gluster implementations.

IBM & Red Hat were able to achieve 1+ Million DB2 transactions per second, with a cost per transaction of $0.51/tpmC.  Accordingly, this is 88% of bare metal performance. This was done using what is currently a tech preview I/O vitalization technology (virtio-blk-data-plane) in RHEL 6.4.  More details on the test rig are in The Register article.  I was honestly a little surprised to see this was performed on a mix of legacy HDD & SSD’s.

Additionally, on the bleeding edge Gluster front, there have been implementations of Gluster FS calls directly from QEMU/KVM guests to Gluster storage.  This eliminates the use of FUSE as an abstraction layer, which should be quite a performance boost.  It will be interesting to see if this functionality makes its way into Red Hat Storage & the RHEV product, to provide higher performance access to commodity storage exposed via Red Hat Storage.  No timeline on that – though a tech preview will probably come first.

As mentioned, these are currently community efforts &/or tech previews, but I think they give a good indication of what Red Hat is working on and how (along with partners, community & customers) Red Hat is leading innovation in the space.