[Vini-users] A few Vini questions
acb at CS.Princeton.EDU
Tue Jun 19 16:55:53 EDT 2007
I don't see any reason why you couldn't use PL-VINI to set up
something like this on a cluster of machines. You might also check
out the Virtual Node support in Emulab:
I haven't tried it but it sounds close to what you want. Perhaps you
could work up a hybrid approach, for instance using vnodes on Emulab,
and running the PL-VINI scripts to generate the hundreds of XORP
configurations you'd require.
On 6/16/07, Brighten Godfrey <pbg at cs.berkeley.edu> wrote:
> Hi Andy,
> Thanks for your reply.
> >> I need to use a very Vini-like setup for a project in which I'm
> >> testing modifications to BGP. I may want to use Vini rather than
> >> constructing my own arrangement, but I have a few questions first...
> >> - Is it possible to run Vini myself, rather than using the PlanetLab
> >> instantiation? How hard would this be? (I'd like to scale up to
> >> very large networks on a cluster.)
> > You could use MyPLC to bring up your own "private" PlanetLab on your
> > local cluster, running identical software to the public PlanetLab.
> > The PL-VINI tools should just work in this environment. It might also
> > be possible to run the PL-VINI tools in Emulab but I haven't tried it;
> > for example, Emulab will let you instantiate a MyPLC on Emulab nodes.
> >> - How much overhead is there in an IIAS router? Probably memory
> >> would be the limiting factor for me. If we assume that I'm running
> >> BGP in Xorp with small routing tables, about how many megabytes can I
> >> expect the whole router to use?
> > We haven't measured memory overhead, but I'd guess around 100MB should
> > be sufficient to run each instance of UML+XORP for small routing
> > tables.
> >> - What size of virtual topologies is it possible to use on PL-VINI
> >> today?
> > I think that depends on what you're trying to do. It sounds like
> > you're thinking of running multiple virtual routers on each physical
> > machine.
> Yes, that's right.
> > Then I suppose the size of your topology will be limited by
> > memory as you pointed out. If you give me more details on what you're
> > thinking about, maybe I can tell you whether or not using PL-VINI is
> > the way to go.
> I'd like to be able to scale up to at least hundreds, ideally
> thousands, of instances of Xorp, each running a BGP process which
> peers with (on average) about 4 other Xorp instances. The idea is to
> use the Internet topology at the AS level, or at least as large of a
> subgraph thereof as is feasible. I'd be sending occasional probe
> packets, but I have some flexibility in the rate, so that wouldn't be
> a problem.
More information about the Vini-users