Re: Use of SAP in a PIM SSM Network

[zaid <zalbanna@MCI.NET>]

Just wondering ........

Instead of having a SAP like mechanism , does it make sense to
have a [Si,Gw] for every [Si,Gx] (x=1....n). Where Gw is a well known
group
with a well known port number ??




-----Original Message-----
From: Toerless Eckert <eckert@cisco.com>
To: Jon Crowcroft <J.Crowcroft@cs.ucl.ac.uk>
Cc: Toerless Eckert <eckert@cisco.com>; Leonard Giuliano
<lenny@juniper.net>; Hugh LaMaster <lamaster@nren.nasa.gov>;
ssm-interest@external.cisco.com <ssm-interest@external.cisco.com>;
J.Crowcroft@cs.ucl.ac.uk <J.Crowcroft@cs.ucl.ac.uk>
Date: Thursday, November 16, 2000 1:43 PM
Subject: Re: Use of SAP in a PIM SSM Network


> not clear - the GLOP and ipv6 ways of allocating multicast introduce
> _some_ hierarchiy - in fact its relatively easy to impose all sorts
if
> you have _enough_ address space.....

Isn't the right phrase for IPv4 "If you had _enough_ address space..."
?

> if yo uhave applications which genuijley have significant number of
> sources  with dynamics, then the server solution does not scale -
you
> have just pushed the management problem to that of
> 1/ locating one of a set of servers (how)
> 2/ making servers fault tolerant (anyone for anycast? :-)
> 3/ managing the application state in an unnatural way
> 4/ putting performance (delay and procerssign) limits on the scale
of
> the application

Well, /1 i think is not siginificantly more expensive if done outside
the
network layer than inside. The way we have it now is to have one
common
"server" (RP) infrastructure for all applications together, and isn't
it exactly this one size fits all" that limits flexibility introduces
unnecessary complexity and can not be individually be optimized for
particular suites of applications ?

/2 Sure, i love anycast ;-)

/3 I do not get it.
"Biking is an unnatural way of transportation (TM) automotive society"
?

/4 Only if we can assume that we can not build scaling server
infrastructures
for large scale applications. I think we can. There must be something
all
those content server boxes will be good for. How about this ? Wasn't
there
sufficient research in distributed systems in the last 20 years to
come up
with appropriate solutions for larger scale applications, as long as
you
stick to some scenarios in the first place (instead of trying to solve
it generically for all of them together in the network layer) ?

> ok so you could have a hierarchy of redundant servers....,and what
> standard internet protocoos are you going to use to manage those?

Why does it have to be standardised in the IETF ? Why not some other
place ?
Why are things like CORBA not standardised in the IETF ? Don't those
distributed applications relying on it run over the Internet also ?

> no, i think we need to retain the deering multicast model and just
> keep up the pressure on Large Router Vendors (this means you:-)
> to be smart about scaling the routers up to meet the users
> requirements rather than scaling the users expectations down to what
> you want to do:-)

We won't take the holy grail away, Lancelot ;-)
I guess this router vendor would just love to see some return of
investment
on multicast, and as far as i am concerned this in the first place
means
to get ip packet replication in routers used and payed for. The
control plane
operations will just follow the users application demands, and SSM is
just
another alternative. If there were more interesting Internet-wide
applications
that can not work well with SSM, then certainly the focus would be
different.

Cheers
Toerless