[dpdk-dev] [RFC] examples: remove l3fwd-vf example

[Ananyev, Konstantin]

Hi Thomas,

> -----Original Message-----
> From: Thomas Monjalon [mailto:thomas.monjalon at 6wind.com]
> Sent: Wednesday, April 20, 2016 11:47 AM
> To: Ananyev, Konstantin
> Cc: Qiu, Michael; Zhang, Helin; Liu, Yong; Cao, Waterman; dev at dpdk.org
> Subject: Re: [dpdk-dev] [RFC] examples: remove l3fwd-vf example
> 
> 2015-08-04 17:12, Ananyev, Konstantin:
> > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Qiu, Michael
> > > Actually, l3fwd works fine with fm10k vf.
> > >
> > > I don't know what's the exact reason of l3fwd-vf still in DPDK,
> > > at least we could make full support for vf in l3fwd instead of another
> > > sample with most code are the same compare with l3fwd.
> >
> > Right now, l3fwd is not able to work properly for cases where number of forwarding  lcores
> > exceeds number of tx queues on any active port.
> > As an example:
> > 2 forwarding lcores and a port with just 1 TX queue (e1000 legacy device).
> >
> >  To make l3fwd work for such cases you need to add some sort of synchronisation on TX path.
> >  Which means one of 2 ways:
> >  either introduce different TX path into l3fwd (one with sync if legacy/virual device is used, another without)
> >  and select it on process startup/config phase,
> >  or sync overhead for fastpath.
> 
> Any news about removing l3fwd-vf example?
> 
> l3fwd has been reworked but l3fwd-power, l3fwd-vf and l3fwd-thread are
> still based on the old l3fwd with APP_LOOKUP_METHOD compile-time flag.

As far as I know, no-one from Intel side is working on it right now.
Konstantin