[ovs-discuss] coexistence of ofproto-dpif and customized ofproto provider

[Ben Pfaff]

On Wed, Mar 19, 2014 at 04:03:23AM +0000, Wuhongning wrote:
> On Wed, Mar 19, 2014 at 03:52:42AM +0000, Wuhongning wrote:
> >> On Wed, Mar 19, 2014 at 01:21:53AM +0000, Wuhongning wrote:
> >> >> On Mon, Mar 17, 2014 at 12:41:13PM +0000, Wuhongning wrote:
> >> >> >> if we have a hardware chip with rich networking feature, and want to
> >> >> >> expose as a enhanced flow switch. at the same time, we also want to
> >> >> >> keep the standard ovs, so what's the best practice to let them work
> >> >> >> together without bother each other?
> >> >> >
> >> >> >Write a second ofproto provider, then configure ovs-vswitchd to use
> >> >> >both.
> >> >>
> >> >> how to "configure", can we do it on-the-fly? restart daemon process
> >> >> is acceptable, but any form of re-package(compile, link...) is difficulty.
> >>
> >> > It's not necessary to recompile or restart.  Just create a bridge with
> >> > "datapath_type" set to the name of the correct datapath provider.
> >>
> >> then where shall I put the newly written ofproto module, and what's
> >> the package style of it? (a .so lib, or a standalone process). we want
> >> to keep all files in the ovs package distributed without any change,
> >> except .ini and .conf files.
> 
> > You're talking about actually running newly written code that didn't
> > exist before?  I can't imagine how you'd do that without compiling or
> > linking.
> 
> yes we want to write new code to cope with some new flow capability of
> hardware, the new ofproto surely needs to be compiled. but we don't
> want to bother the original package, we only want to add something to
> a running system. this is the reason why we are considering develop a
> forked branch of vswitchd, and let it to run with original vswitchd
> concurrently.

I don't think there's a need to fork anything to run one vswitchd with a
given datapath provider and a second vswitchd with a different one.