[comp.protocols.tcp-ip] question about architecture of AGS+

STC@BROWNVM.BROWN.EDU (Steve Carmody) (12/05/90)

Here at Brown, we've been having fun this fall trying to select a
vendor for high speed, multi-protocol routers. At this point,
we've got a few questions which we haven't been to resolve.
In most cases, we have multiple answers to the questions, but
haven't been able to determine which is the correct one. If
someone can answer one or more of these, we'd appreciate the
help. All of the questions are about the Cisco AGS+.

1.  If a box is configured with MEC ethernet cards and an
FDDI interface, packets usually seem to stay within the c-
bus. When, if ever, must a network interface card send the
packet back to the main processor card (located on the VME
bus)?

2. In the most recent Bradner tests, the performance of the
AGS+ fell off in the Dual-stream
between interface cards test when filters were turned on. The box
ran well with dual streams without filters, but throughput
fell as soon as one filter was enabled. Does anyone know why?

3. The numbers in the Bradner tests seem to indicate that the
AGS+ performs well. We have heard rumors (but have seen no
numbers) that claim that performance of the AGS+ falls off if
it must route mutiple streams containing multiple protocols.
Does anyone have any measured numbers to support this rumor?

4. One of the software improvements Cisco has introduced to
boost performance is called "fast switching". This seems to
allow the network interface board to make more of the
decision about which interface to send the packet to. That's
good news. We've heard (again, no more than a rumor) that the
fast switch code uses a seven entry cache on the network
interface card. If thats true, a synthetic benchmark could
run well (or could be tuned to run poorly), depending on the
contents of the packet stream directed at the router being
tested. Does anyone know how the fast switch code really
works?

once again, thanks for any help.