Re: virtio-scsi <-> vhost multi lun/adapter performance results with 3.6-rc0

[Stefan Hajnoczi <stefanha@xxxxxxxxx>]

On Sat, Aug 11, 2012 at 12:23 AM, Nicholas A. Bellinger
<nab@xxxxxxxxxxxxxxx> wrote:
> Using a KVM guest with 32x vCPUs and 4G memory, the results for 4x
> random I/O now look like:
>
> workload         | jobs | 25% write / 75% read | 75% write / 25% read
> -----------------|------|----------------------|---------------------
> 1x rd_mcp LUN    |   8  |     ~155K IOPs       |      ~145K IOPs
> 16x rd_mcp LUNs  |  16  |     ~315K IOPs       |      ~305K IOPs
> 32x rd_mcp LUNs  |  16  |     ~425K IOPs       |      ~410K IOPs
>
> The full fio randrw results for the six test cases are attached below.
> Also, using a workload of fio numjobs > 16 currently makes performance
> start to fall off pretty sharply regardless of the number of vCPUs..
>
> So running a similar workload with loopback SCSI ports on bare-metal
> produces ~1M random IOPs with 12x LUNs + numjobs=32.  At numjobs=16 here
> with vhost the 16x LUN configuration ends up being in the range of ~310K
> IOPs for the current sweet spot..

This makes me wonder what a comparison against baremetal looks like
and the perf top, mpstat, and kvm_stat results on the host.

Stefan
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html