[AG-TECH] Multicast beacon questions

[Bill Nickless]

At 02:37 PM 10/24/2001 -0400, David E. Bernholdt wrote:
>As my networking friends point out, the current beacon approach isn't
>terribly scalable as the number of AG nodes increases (presuming a 1:1
>association with beacons). Its already enough to trigger concern with
>folks here, though maybe its just that they're not used to seeing
>any mcast traffic at all.  Has anyone thought about this?

My experience with multicast deployment is that a given network is at one 
of two levels of multicast deployment (when multicast is deployed at all!):

  - Initial deployment: uses software forwarding, dense-mode
    flood-and-prune routing protocols, and so forth.  Generally this
    works "correctly" for things like the NASA Select TV one-to-many
    desktop applications.  Overall packet rates are in the 10-250
    packets/second range.

  - Solid Deployment: uses hardware forwarding, sparse-mode
    forwarding protocols, IGMP snooping on switched networks, and
    so forth.  This level of deployment supports packet rates exceeding
    1000 packets/second, and is absolutely required for Access Grid.

If a network is at the "Initial Deployment" level, then the beacon group 
traffic levels will cause problems right away.  I believe this is a good 
thing, because it highlights the need to make infrastructure upgrades to 
the "Solid Deployment" level.

In other words, if today's beacon group is causing problems with a given 
network, that network is not yet properly configured to support an Access 
Grid node.

The good news is that experimental network hardware and software is no 
longer required to get to the "Solid Deployment" level.  The required 
multicast routing protocols are in Cisco GA code, for example.

===
Bill Nickless    http://www.mcs.anl.gov/people/nickless      +1 630 252 7390
PGP:0E 0F 16 80 C5 B1 69 52 E1 44 1A A5 0E 1B 74 F7     nickless at mcs.anl.gov