On 11/15/23 6:57 AM, Stefan Hajnoczi wrote: > On Wed, Nov 15, 2023 at 12:43:02PM +0100, Stefano Garzarella wrote: >> On Mon, Nov 13, 2023 at 06:36:44PM -0600, Mike Christie wrote: >>> This adds support for vhost-scsi to be able to create a worker thread >>> per virtqueue. Right now for vhost-net we get a worker thread per >>> tx/rx virtqueue pair which scales nicely as we add more virtqueues and >>> CPUs, but for scsi we get the single worker thread that's shared by all >>> virtqueues. When trying to send IO to more than 2 virtqueues the single >>> thread becomes a bottlneck. >>> >>> This patch adds a new setting, virtqueue_workers, which can be set to: >>> >>> 1: Existing behavior whre we get the single thread. >>> -1: Create a worker per IO virtqueue. >> >> I find this setting a bit odd. What about a boolean instead? >> >> `per_virtqueue_workers`: >> false: Existing behavior whre we get the single thread. >> true: Create a worker per IO virtqueue. > > Me too, I thought there would be round-robin assignment for 1 < > worker_cnt < (dev->nvqs - VHOST_SCSI_VQ_NUM_FIXED) but instead only 1 > and -1 have any meaning. > > Do you want to implement round-robin assignment? >
It was an int because I originally did round robin but at some point dropped it. I found that our users at least: 1. Are used to configuring number of virtqueues. 2. In the userspace guest OS are used to checking the queue to CPU mappings to figure out how their app should optimize itself. So users would just do a virtqueue per vCPU or if trying to reduce mem usage would do N virtqueues < vCPUs. For both cases they just did the worker per virtqueue. However, I left it an int in case in the future someone wanted the future.
