The IP over ATM Mailing List Archive by date

[c.oneill@trl.oz.au (Chris O'Neill)]

>There is a "solution" other than ABR - switches / concentrators with
>a decent amount of buffering which implement frame drop and EPD, and
>possibly other congestion control disciplines. These switches would
>have to treat UBR and CBR/VBR differently so that the large buffering
>for UBR didn't impact the latencies for CBR traffic. Hey - if that worked
>we wouldn't need ABR :-)

This seems like a good time to ask: Is there any reason why UBR with EPD
doesn't work well enough?

By well enough I mean with good enough "goodput" and I'm suggesting that
instantaneous fairness is a secondary concern.  I'd like to know if anyone
has simulated a situation where TCP used over UBR with EPD and adequate
buffering does not give a goodput close to 100%.

I find it a little curious that there is such a perceived need for ABR in
ATM when services such as SMDS and frame relay already provide a service
with similar characteristics to UBR with EPD.  No-one seems to be
suggesting that there are fundamental defects in the quality of service
provided by SMDS and frame relay to data applications.

>
>Bryan Gleeson
>Adaptec.

Chris O'Neill
Telstra Research Labs
Melbourne Australia