The Routing Over Large Clouds Mailing List Archive by date

[Sam Wilson <ercm20@tattoo.ed.ac.uk>]

> > > But, as ATM large clouds do not and, in short-term, will not, exist,
> > > NHRP is not a short- or long- term answer.
> > 
> > I don't believe this is true for all values of 'large' and 'short-term'. 
> > In Scotland we have 4 Academic MANs based on ATM or mixed ATM/FDDI
> > technology.
> 
> If your network is really a large cloud, that is, not necessarily
> geographically large, but contains a large number of non-IP switches,
> then, that is a bad news. It will be an administrative nightmare.

Of the order of a couple of dozen ATM switches across the 4 MANs;
similar numbers of IP routers each with a single ATM interface -
routers-on-sticks like I said. 

> > NHRP MAY help us to avoid either the
> > obvious router bottlenecks
> 
> "obvious router bottlenecks"? What are they?

Routers on sticks - we can't afford to install physically multihomed IP
routers in parallel with ATM switches.

> Isn't it obvious that cell-switching IP routers, whose cell switching
> fabrics are setup by RSVP to have VCs for each flow, are just as fast
> as legacy ATM switches?

Of course <wry smile> but such devices are not readily and affordably
available in the UK.  Yet.

> > or the administrative complexity of
> > overlaying a mesh of IP connections over a simple ATM star topology. 
> 
> If the mesh of IP connections are necessary only within a subnet,
> a small cloud between the cell-switching IP routers, it's administration
> is only as complex as the current configuration.
> 
> And, are you aware that if you have a large cloud with N nodes and
> a mesh of IP connections over a simple ATM star topology, most
> links will have O(N^2) VCs? That is, if you use CBR, each VC can't
> have high bandwidth.

I'll cover the when I reply to Scott Marcus' message.

> > Cell switching routers clearly aren't that
> > in the short term,
> 
> What is clear?

What I wrote was "...each time I read one of the NHRP drafts I keep
thinking there MUST be a better way.  Cell switching routers clearly
aren't that in the short term...".  If I want to build a large-ish
IP-over-ATM network with kit I could spec and order 3-6 months ago and
install today CSRs are not the better way than NHRP.  If we were buying
in a year's time they might be.  We need native ATM too to pure IP
routers with ATM i/fs as the network nodes is not an option. 

> You can order Ipsilon swiches today. You can make a more general
> CSR from legacy PVC/SVC ATM switches controlled by a usual UNIX WS
> through RS 232c.
> 
> It is true that RSVP spec may not fix in the short term, in which
> case, we can use ST2 or draft RSVP, can't we?

Just as the whole world is not ATM it is also not IP.  

> > and aren't currently the IETF consensus about the way
> > to go.
> 
> What consensus, do you think, we need?

How about the ION chairs or Joel saying "it's clear that the NHRP effort
is a waste of time - let's drop it".   :-)

> To let CSR work, we need no new protocol.
> 
> When we discussed on the issue of WG creation, Joel Halpern, the
> IESG routing Area Director, stated that we can't create a WG only
> to develop a router that needs no protocol.

I can't answer that, but I might ask why the IP over ATM WG didn't
specify CSRs as the way to do tha job.

> So, the IETF consensus is, seemingly, that to go for CSRs
> we need no further IETF consensus.
> 
> Anyway, to abandon NHRP, we only need a consensus that it is
> proven to not to work.

Exactly - I don't see such a consensus in the IP-over-ATM-related areas
of the IETF - do you?



Sam Wilson
Network Services Division
Computing Services, The University of Edinburgh
Edinburgh, Scotland, UK