Cell Relay Archive

[Berry Kercheval <berry@join.com>]

ag@eden-gardens.ee.duke.edu (Anoop Ghanwani) writes:
> I am trying to find out if the copy function in multicast
> ATM switches requires a non-neglible amount of time
> relative to the service time of a cell.  I'd appreciate
> any insights that anyone can offer.

The answer is, it depends.  While I was at Xerox PARC we built two
experimental ATM switches with multicast features.

 One used a crossbar to replicate the multicast cells and send them
back to the input port for another trip through the fabric; Obviously
in this case it took at least twice as long for a multicast cell to
get through the switch as it did for a unicast cell.

The other switch used an interesting shared memory approach, and
tagged each cell with the destination ports in a bit mask.  A unicast
cell was tagged with one bit, a cell to be discarded with no bits, and
multicast cells with more than one bit.  In this switch it took no
more time to process a multicast cell than it did a unicast one.

Now, both of these were experimental switches (there were three of the
first, and perhaps eight of the second built) but illustrate my point:
there is no hard and fast rule for this.  

Ask your switch vendor.

  --berry