James G. Sack (jim) wrote:
Van Jacobson dropped a disparaging remark about ATM being notable for not correctly grasping the concepts.
Huh? Now that's just patently absurd. It also shows a lack of understanding about transport/network layer vs. data link/physical layer. I think I need to read this, did anybody find a transcript?
It's not TCP vs. ATM. It's Ethernet vs. ATM. Note: The wikipedia article about ATM seems to have an axe to grind against ATM.
The problem is that Ethernet uses CSMA/CD (carrier sense multiple access/collision detect) and cannot, by its nature, provide transmission guarantees.
Apparently, there is a form of "industrial ethernet" which relies on a star topology. Presumably, the main controller has to rely on a round-robin/token to arbitrate between the arms of the star and we're back to an ATM/token system.
QoS requires bandwidth reservation along the entire length of the communication chain aka a virtual circuit. There's just no way around that.
Now, unlike a *physical* circuit which has to be set up and stay connected the entire time, a *virtual* circuit only has to arbitrate between packet classes and can swap stuff around as long as it fullfills its guarantees. So, as long as I can dump enough non-reserved traffic to fullfill my guarantees, bandwidth for a virtual circuit doesn't have to be reserved but unused. It just has to be reserved.
-a -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
