On Tue, 2012-09-11 at 12:36 +0800, Asias He wrote: > Hello Nicholas, >
Hello Asias! > On 09/07/2012 02:48 PM, Nicholas A. Bellinger wrote: > > From: Nicholas Bellinger <n...@linux-iscsi.org> > > > > Hello Anthony & Co, > > > > This is the fourth installment to add host virtualized target support for > > the mainline tcm_vhost fabric driver using Linux v3.6-rc into QEMU 1.3.0-rc. > > > > The series is available directly from the following git branch: > > > > git://git.kernel.org/pub/scm/virt/kvm/nab/qemu-kvm.git vhost-scsi-for-1.3 > > > > Note the code is cut against yesterday's QEMU head, and dispite the name > > of the tree is based upon mainline qemu.org git code + has thus far been > > running overnight with > 100K IOPs small block 4k workloads using v3.6-rc2+ > > based target code with RAMDISK_DR backstores. > > Are you still seeing the performance degradation discussed in the thread > > "vhost-scsi port to v1.1.0 + MSI-X performance regression" > So the performance regression reported here with QEMU v1.2-rc + virtio-scsi ended up being related to virtio interrupts being delivered across multiple CPUs. After explicitly setting the IRQ affinity of the virtio0-request MSI-X vector to a specific CPU, the small block (4k) mixed random I/O performance jumped back up to the expected ~100K IOPs for a single LUN. FYI, I just tried this again with the most recent QEMU v1.2.50 (v1.3-rc) code, and both cases appear to be performing as expected once again regardless of the explicit IRQ affinity setting. --nab
