Cell Relay Archive

[mdoar@amaryllis (Matthew Doar)]

Phoebe Li (pyl@acpub.duke.edu) wrote:
>From March 1995 Data Communications:

>14 vendors were invited to participate in its ATM Stress test and seven
>did participate.

As someone else has already pointed out, the Data Comm. tests had a
few peculiarities.

(1) The DEC traffic management was just turned off, because the test
equipment couldn't handle it. Hardly a fair comparison.

(2) There appeared to be some confusion in the article about ``cell
delay variation'' (as described in the ATM Forum UNI spec) and ``cell
interarrival time''. Then elsewhere in the article, it was claimed
that the standard deviation of the cells' delays was used as the
jitter measurement. No details of the distributions of delays were
given, just mean values.

(3) Multicast traffic, control traffic, NMS traffic: all were not
considered. The usefulness of a switch is surely more than just its
ability to send packets where you want them to go, at a certain rate?

(4) The latencies quoted to show the effects of bursty IP data streams
on a CBR stream didn't make it clear whether just the latencies of the
*CBR* cells were being considered, as would seem sensible.

(5) Finally, latencies of tens of microseconds are of dubious value as 
significant performance measurements in networks which may involve WAN
connections with delays of milliseconds.

I have no doubt that useful data exists in the article in a general
sense, but it seems we have a way to go in really testing ATM
switches. 

Is anyone repeating the tests or doing similar evaluations?



--
[Not necessarily the opinions of N.E.T]
.......................................................................
Matthew Doar		   N.E.T. (Network Equipment Technologies, Inc.)
mdoar@net.com	 	 800, Saginaw Drive, Redwood City, CA 94063 USA.