On Thu, Aug 15, 2019 at 09:42:40AM +0000, ? jiang wrote:
> This change lowers ring buffer reclaim threshold from 1/2*queue to budget
> for better performance. According to our test with qemu + dpdk, packet
> dropping happens when the guest is not able to provide free buffer in
> avail ring timely with default 1/2*queue. The value in the patch has been
> tested and does show better performance.
>
> Test setup: iperf3 to generate packets to guest (total 30mins, pps 400k, UDP)
> avg packets drop before: 2842
> avg packets drop after: 360(-87.3%)
>
> Signed-off-by: jiangkidd <jiangkidd@xxxxxxxxxxx>
To add to that:
Further, current code suffers from a starvation problem: the amount of
work done by try_fill_recv is not bounded by the budget parameter, thus
(with large queues) once in a while userspace gets blocked for a long
time while queue is being refilled. Trigger refills earlier to make sure
the amount of work to do is limited.
With this addition to the log:
Acked-by: Michael S. Tsirkin <mst@xxxxxxxxxx>
> ---
> drivers/net/virtio_net.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index 0d4115c9e20b..bc08be7925eb 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -1331,7 +1331,7 @@ static int virtnet_receive(struct receive_queue *rq, int budget,
> }
> }
>
> - if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
> + if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) {
> if (!try_fill_recv(vi, rq, GFP_ATOMIC))
> schedule_delayed_work(&vi->refill, 0);
> }
> --
> 2.11.0
