Paul,
Even if QoS/CoS is desirable, is its widescale deployment possible
given the current Internet? Here are some of the issues stopping QoS:
1. No billing available: If there is no billing system which can
determine when people use QoS capabilities, than how can it be
deployed widely and effectively.
2. No support on routing infrastructure: Routers today have
several problems:
* No policing at ingress: You can't have QoS unless
you can limit how much traffic enters the network,
and discard, or at least mark the excess traffic.
* No effective BW reservation mechanism: RSVP may
solve this in the future, but it's not deployed now.
* No effective Class of Service mechanism:
* Packet/sec processing limits: Even if the Internet
infrastructure knows how to do all of the above,
(reserve, police, mark, and queue according to
QoS parameters), it will still fail if the box
cannot satisfy the packet rates coming to it.
Both your company and mine are working on ways to fix these
limitations. But widescale QoS capabilities requires that every
router between a source and destination has these capabilities.
QoS guarantees maybe possible within a single AS that uses a
router infrastructure with Frame Relay or ATM switching
capabilities (which tend to have implicit support for all of the
above listed features). Of course, this leads to the next problem,
namely:
3. No support across ASes. First of all, BGP provides no QoS
metrics. So there is no way to determine if a particular AS
should even be considered in setting up a QoS path. Second, while
a single AS could be upgraded to QoS-capable equipment, a forklift
upgrade across the Internet to QoS-capable equipment won't happen
anytime soon.
I think the Internet will stay at best-effort for quite a while, with
small pockets of QoS in special situations.
Prabhu
______________________________ Reply Separator _________________________________