Re: [PATCH v4 3/5] virtio-scsi: default num_queues to -smp N

[Stefan Hajnoczi <stefanha@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1321 bytes --]

On Wed, May 27, 2020 at 11:38:17AM +0100, Daniel P. Berrangé wrote:
> On Wed, May 27, 2020 at 11:29:23AM +0100, Stefan Hajnoczi wrote:
> > Automatically size the number of virtio-scsi-pci, vhost-scsi-pci, and
> > vhost-user-scsi-pci request virtqueues to match the number of vCPUs.
> > Other transports continue to default to 1 request virtqueue.
> 
> IIRC this was raised on earlier versions of the series, but i don't
> recall the outcome and no caveats are mentioned here...
> 
> Is this default still valid for very large $vCPUs. eg if I run QEMU
> with "-smp 512" or even larger (I've seen people discussing 1000's of
> CPUs), is this going to cause problems with the virtio-scsi default
> queue counts ? Is there such a thing as "too large" for the num
> of queues setting ?   num vectors defaults to a value derived from
> num queues, so is there concept of "too large" for num of vectors
> setting ?
> 
> Ideally the commit message would answer these questions for future
> reference.  Same for the next patch to virtio-blk

Good point. Actually this patch and the virtio-blk ones no longer
contain the queue number policy. The new virtio_pci_optimal_num_queues()
function encapsulates the policy to avoid duplication. I'll resend and
update that patch with the full rationale.

Thanks!

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]