Re: [dpdk-users] VIRTIO for containers

["Tan, Jianfeng" <jianfeng.tan@intel.com>]

Hi Zhike,


On 10/26/2017 8:53 PM, 王志克 wrote:
> Hi,
>
> Thanks for reply.
>
> To put tcp/ip rx into app thread, actually, might avoid that with a 
> little change on tap driver. Currently, we use 
> netif_rx/netif_receive_skb() to rx in tap, which could result in going 
> up to the tcp/ip stack in the vhost kthread. Instead, we could backlog 
> the packets into other cpu (application thread's cpu?).
>
> [Wang Zhike] Then in this case, another kthread like ksoftirq will be 
> kicked, right?
>
> In my understanding, the advantage is that the rx performance can be 
> even improvement, while disadvantage is that more cpu resource is used 
> and another queue is needed. If that can be done in a smart way, like 
> system has idle CPUs, we can use this way, else fall back to only use 
> one kernel thread. Just my 2 cents.

Yes, make sense. We need a smart mechanism to decide if it is handled in 
vhost kthread or ksoftirqd kthread. And also, we could even avoid 
forking a vhost kthread, to avoid too many context switches.

Thanks,
Jianfeng