[Xorp-hackers] XORP and/or Click with an overlay protocol

[Pavlin Radoslavov]

Victor Faion <vfaion at gmail.com> wrote:

> On Mon, Dec 1, 2008 at 18:46, Pavlin Radoslavov <pavlin at icsi.berkeley.edu>wrote:
> 
> > Victor Faion <vfaion at gmail.com> wrote:
> >
> > > Hello,
> > >
> > > I wanted to use my forwarding engine (which has its own forwarding table)
> > > together with my own protocol (at the application layer, using sockets)
> > with
> > > XORP. I wasn't sure if it's better to implement a separate process that
> > > interacts with XORP's FEA (this would be the forwarding engine) and
> > another
> > > process that represents the protocol or if I should implement all of this
> > > using Click and then plug it into XORP (or just use it only with Click).
> > In
> > > other words, how much of XORP's code I would need to change to do this
> > and
> > > would it be easier to do it in Click or to use both?
> >
> > Without additional info about your protocol it is difficult to give
> > you advice that will fit best your specific needs.
> >
> > If I make the assumption that your control protocol is similar to, say,
> > OSPF or RIP, my generalized advice would be to implement your
> > control protocol as a separate process that interacts with the XORP
> > FEA. If you don't have any specific requirements, you shouldn't need
> > any additional changes to XORP.
> >
> >
> > Re. your question of XORP vs. Click.
> > From XORP's perspective, Click is an IPv4/IPv6 data plane,
> > though Click itself is much more than that.
> > Hence, if you implement your protocol in XORP, the "shall I use
> > XORP+Click" question becomes a question of whether you want to use
> > Click as the IPv4/IPv6 data plane.
> > On the other hand, if you have a relatively simple protocol with
> > some unusual requirements (say, it requires tight integration with
> > the data plane), and the existing UNIX kernel API is not sufficient,
> > you might be able to save time getting the initial prototype working
> > if you use only Click.
> >
> > Hope that helps,
> > Pavlin
> >
> >
> >
> > > Victor
> > > _______________________________________________
> > > Xorp-hackers mailing list
> > > Xorp-hackers at icir.org
> > > http://mailman.ICSI.Berkeley.EDU/mailman/listinfo/xorp-hackers
> >
> 
> 
> 
> Hello,
> 
> Thank you for the response, the control protocol is a link-state routing
> protocol. It uses LSR but also needs to associate additional information
> with hosts and this is why I think I might need to make another XORP process
> for this protocol, and I think its easier to plug in a new protocol into
> XORP rather than Click.
> 
> As for using Click as the data plane, I could make my forwarding table as a
> Click element, or would it be simpler to do it as a separate XORP process
> without relying on Click?

If it is just a routing protocol you don't need Click. It will be
simpler if you use the existing UNIX kernel forwarding, and
implement LSR as a separate XORP process.

Pavlin


> Victor
> _______________________________________________
> Xorp-hackers mailing list
> Xorp-hackers at icir.org
> http://mailman.ICSI.Berkeley.EDU/mailman/listinfo/xorp-hackers