[c-nsp] PGM, SPM packets and high CPU utilisation

[Ras]

We make heavy use of Microsoft's MSMQ libraries to implement our PGM
messaging bus. Unfortunately this implementation of PGM likes to send
an awful lot of SPM messages within the PGM streams, each of which
have the Router Alert IP option set.

At the moment we're seeing ~10kpps of these packets. Our distribution
layer (Cat4500, SupV) is melting under the pressure (90%+ CPU) and
Cisco don't seem to be able to give any clear indication as to whether
upgrading hardware would help (as these packets all get software
routed, simply moving to standard 'faster switching hardware' doesn't
help much).

Has anyone else run into this problem? If so any ideas on how to
mitigate or resolve the issue? At the moment we're potentially looking
at having to move all our PGM speakers into a single gigantic
world-spanning VLAN to avoid having to route it, but that obviously
isn't an ideal solution.

Thanks in advance,
Ras