The IP over ATM Mailing List Archive by date

[bansal@ccrl.nj.nec.com (Vivek Bansal)]

Alan and Craig,

I think the idea of having one VC per transport connection is a neat idea given
the way the protocols (like TCP/UDP) are configured in present systems. If we
want to use any of the features that ATM provides in terms of QoS guarantees and
traffic specification at the time of connection setup then we have to have a
VC per connection where ATM can provide guarantees at connection level rather
than protocol level. 
We have been doing this on out testbed here by modifying our ATM device drivers,
and it is really helpful. We do suffer a hit per packet as now the driver has
to look up the port number but we have to stay with that for now (unless we
change the TCP/IP implementation). We are implementing a new transport protocol
(with new APIs) with ATM specific intelligent modules in which we have
incorporated this idea too. 

Vivek...