[pve-devel] [PATCH] add virtio-vhost-user nic support

[Alexandre DERUMIER]

>>But the OVS bridge still pass all traffic to the kernel, because we use 
>>in-kernel network card drivers?

It's more complex than that, but basicly, dpdk can talk directly with nic, bypassing the kernel.
http://www.intel.com/content/dam/www/public/us/en/documents/presentation/dpdk-packet-processing-ia-overview-presentation.pdf (page 15)

That's also why dpdk don't work with all network vendor nic
http://dpdk.org/doc/nics.


>>Anyways, I guess you have already done some performance benchmarks? Is 
>>it really faster? 

I'm going to do benchmark next week.



----- Mail original -----
De: "dietmar" <dietmar at proxmox.com>
À: "aderumier" <aderumier at odiso.com>
Cc: "pve-devel" <pve-devel at pve.proxmox.com>
Envoyé: Mercredi 4 Mai 2016 06:03:06
Objet: Re: [pve-devel] [PATCH] add virtio-vhost-user nic support

> Why bypassing the kernel ?: 
> Current kernel stack have some limitations, around 1mpps (millions packets/s) 
> vs 10mpps for dpdk by cpu core, so for virtual routers and firewall 
> (synflood,udpflood of 64bytes), this is really needed) 

But the OVS bridge still pass all traffic to the kernel, because we use 
in-kernel 
network card drivers? 

Anyways, I guess you have already done some performance benchmarks? Is 
it really faster?