Hi folks, On Wed, 2012-07-04 at 18:52 -0700, Nicholas A. Bellinger wrote: > > To give an idea of how things are looking on the performance side, here > some initial numbers for small block (4k) mixed random IOPs using the > following fio test setup:
<SNIP> > fio randrw workload | virtio-scsi-raw | virtio-scsi+tcm_vhost | bare-metal > raw block > ------------------------------------------------------------------------------------ > 25 Write / 75 Read | ~15K | ~45K | ~70K > 75 Write / 25 Read | ~20K | ~55K | ~60K > > After checking the original benchmarks here again, I realized that for virtio-scsi+tcm_vhost the results where actually switched.. So this should have been: heavier READ case (25 / 75) == 55K, and heavier WRITE case (75 / 25) == 45K. > In the first case, virtio-scsi+tcm_vhost is out performing by 3x > compared to virtio-scsi-raw using QEMU SCSI emulation with the same raw > flash backend device. For the second case heavier WRITE case, tcm_vhost > is nearing full bare-metal utilization (~55K vs. ~60K). > > Also converting tcm_vhost to use proper cmwq process context I/O > submission will help to get even closer to bare metal speeds for both > work-loads. > Here are initial follow-up virtio-scsi randrw 4k benchmarks with tcm_vhost recently converted to run backend I/O dispatch via modern cmwq primitives (kworkerd). fio randrw 4k workload | virtio-scsi+tcm_vhost+cmwq --------------------------------------------------- 25 Write / 75 Read | ~60K 75 Write / 25 Read | ~45K So aside from the minor performance improvement for the 25 / 75 workload, the other main improvement is lower CPU usage using the iomemory_vsl backends. This is attributed to cmwq providing process context on the same core as the vhost thread pulling items off vq, which ends up being on the order of 1/3 less host CPU usage (for both workloads) primarly from positive cache effects. This patch is now available in target-pending/tcm_vhost, and I'll be respinning the initial merge series into for-next-merge over the next days + another round of list review. Please let us know if you have any concerns. Thanks! --nab -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
