From: Howard C. Berkowitz (hcb@gettcomm.com)
Date: Mon Jun 21 2004 - 16:26:04 GMT-3
While I can't speak if IOU is the tool used in Cisco's scalability
lab and other non-training internal functions (e.g., simulating a
large system bug), software tools have long been in use within Cisco
development. There is a very specific need for them that has nothing
to do with training, and I doubt they were developed with training in
mind.
Cisco developers, at the IETF or other industry groups, often speak
of simulation results as an input to general work on the scalability
of protocols. IIRC, it was Dino Farinacci that did some work
suggesting that EIGRP really shouldn't be used for domains of more
than 2000-3000 routers (not that any reasonable network designers,
other than perhaps some very, very special situations, would build a
domain this large). There are ISPs running flat ISIS networks with
1500 routers, and I suspect they didn't just build this and see if it
worked -- they probably got Cisco to benchmark a simulation.
In some of my own router development work, we would put Zebra or
GateD on racks and racks of UNIX processors running routing
protocols, but this was not at all for training. It was done either
for load generation to benchmark a hardware router, or to explore
possibility scalability problems with routing protocols that only
happened in very large networks.
Incidentally, for the very highest speed tests, one tends to capture
a router protocol stream and then put it on a high-end hardware
protocol simulator, as from HP (or Agilent; I never remember who
makes what), Ixia, or Tektronix. These devices don't really
understand the routing protocol, but can do simple
send-this-if-I-receive-that. The important thing is they can do this
at 10 Gigabits, either OC-192 or 10GE. In the absence of commercial
products to do this, I'm aware of some custom hardware built, solely
for developers, but router manufacturers.
This archive was generated by hypermail 2.1.4 : Sat Jul 03 2004 - 19:40:46 GMT-3