Cell Relay Archive

[albert.e.manfredi@boeing.com]

In article <8olu4v$2m1$1@bcarh8ab.ca.nortel.com>,
  "Dominic Richens" <dominicr@nospamnortelnetworks.com> wrote:
>
> Thomas Frieling <thomas.frieling@experteach.de> wrote in message
> 8okv48$i9o$1@nnrp1.deja.com">news:8okv48$i9o$1@nnrp1.deja.com...
> > In article <8oj7bc$fo9$1@nnrp1.deja.com>,
> >   albert.e.manfredi@boeing.com wrote:
> > >(...)
> > > So RED permits this flow regulation to be done
> > > independently by each router along the path.
> > > (...)
> >
> > Only by the router? Isn't it possible that an ATM-switch - aware of
the
> > fact that certain VCs carry TCP/IP-traffic - randomly discards
packets
> > (AAL5-PDUs) and thus resets the TCP-window size of only one
connection at
> > a time before its buffer situation forces to drop packets massively
and
> > thus resets the window sizes of all connections?
>
> Yeah, that is kinda what I was getting at when I said that W-RED
> implementation in an ATM switch was getting a little out of hand
(IMHO).
> PPD and EPD are generic AAL5 techniques, useful for FrATM trunks,
X.25,
> MPLS, IP..whatever. Whereas W-RED in specifically TCP/IP...and
requires that
> one layer (ATM) knows specific stuff about another layer (TCP) that
is 3
> layers away!

Okay, I understand where everyone's coming from.

EPD discards cells of a given VC based on something it checks out at
the AAL5 layer. With WRED, since multiple IP flows could exist in that
VC, an ATM switch would have to go at least up to Layer 3, and perhaps
Layer 4, to see whether the cells should be discarded.

Not sure whether this is much more work, actually, is it? The cells
you'll be discarding are going to be consecutive within that VC, just
as they are for EPD. And IP is probably the only Layer 3 you care to
worry about. Dunno. Sounds like it might not be such a big deal to do
this within the switches.

--
Bert
albert.e.manfredi@boeing.com


Sent via Deja.com http://www.deja.com/
Before you buy.