[e2e] ATM

[Alex Cannara]

Being somewhat familiar with current network-system developments, using
present and upcoming network processors, fabrics and physical
interfaces, I can testify that ATM is on it way to oblivion, even in
backbone systems.  The current demise of the 'new economy', at Internet
speed, however, has saved much existing ATM equipment, for a little
while, simply due to new-buildout, new-purchase cost cutting.  This will
not last long.  The committee-compromise-created overhead in ATM (not
just the 10% byte penalty, but the lose-a-cell-lose-a-packet folly,
cost...) is seen by system engineers as folly, especially in a commodity
business having access to new, inexpensive Gb/s and upcoming 10Gb/s
alternatives.

Alex

Craig Partridge wrote:
> 
> In message <5.0.2.1.2.20010417131553.03059950 at mail.reed.com>, "David P. Reed" w
> rites:
> 
> >The function is important, I agree, and ATM hardware was an off-the-shelf
> >solution you can kludge to perform it (though why frame relay wasn't good
> >enough is beyond me - maybe all those companies with ATM switches needed
> >some market after the original one died?)
> 
> >From what I can tell, the answer is number of circuits one can provision.
> Remember the circuits are over-provisioned and that's a *feature* in
> the provisioning world.  So out of the DSLAM comes a T1, T3, OC3 (name
> your speed circuit) over which you stuff as many subscriber lines as
> you can.  So, for instance, some of the more risk loving DSL providers only
> guarantee around 10Kbps per subscriber (with peak rate of, say, 1.5 or
> 10 Mbps).  And it turns out that your typical ATM equipment is (slightly)
> better at supporting vast numbers of tiny circuits (ATM equiv of tinygrams)
> with large peak rates than frame relay equipment...
> 
> Craig