[c-nsp] MSMQ, IP Router Alert and melting switches

Ras jeekay at gmail.com
Tue Jun 19 10:25:19 EDT 2007


Greetings,

We've recently run into a problem with a home-grown application that
uses MSMQ 'Reliable Multicast' (ie PGM) to communicate.

Long story cut short, PGM relies on a set of packets called SPMs all
of which have the IP Option 'Router Alert' set. This is one of those
things that seem like a fantastic idea on paper, but...

During busy periods we're seeing thousands of these packets per
second. As each individual packet has to be inspected by the CPU (this
isn't just software forwarding, this is full-on
to-the-CPU-for-full-examination punting) they melt under the pressure
(90%+ CPU and OSPF/HSRP/BGP starts flapping). We're running with
Cat4500s, with a combination of Sup4/5s. Speaking to Cisco this is
essentially expected behaviour as the forwarding capacity for packets
needing CPU examination is in the very low thousands.

Has anyone run into this problem before? If so, how did you solve it?
Our options seem to be one of a) Make MSMQ not set the alert bit as we
don't require it anyway (doesn't seem to be possible), b) Make the
Ciscos ignore router alert bits (not terrible, but might affect
non-PGM traffic and I'm not even sure this is easily achievable) and
c) Stick in a platform that can handle this better than the Ciscos
(not convinced this is easy either as it's a CPU thing not a
forwarding plane thing).

Thanks,
Ras


More information about the cisco-nsp mailing list