Gary,
Thank you very much for your post. Your views concerning QoS are held by the majority of my colleagues and are typical of a Cisco employee. We buy 100% Cisco routers, and definitely have Cisco's influence inbred into our networking culture.
However, IP is a higher layer
protocol and therefore can discriminate among different types of traffic far better
than ATM.
After re-reading my last statement, I realize there is a slight inaccuracy. You are right that IP can identify different types of traffic. But you need a connection-oriented mechanism to provide end-to-end prioritization of that traffic. Routers use TCP (layer 4) to provide connection-oriented services. TCP's weakness is that it's built on top of a connectionless protocol (IP).
Correct that IP can not, without modifications, support real time traffic. However IP
switching, RSVP, IP precedence bit setting, etc. etc. are gunning to be just this type
of modification.
RSVP, IP precedence, class-based queueing, etc... will not scale in ISP/carrier-class networks. Cisco needs to listen hard to this message from ISP's. I don't think they get it yet. These technologies Cisco is pushing require an awful lot of signaling overhead, so much so that they will cause a medium to large-sized network to collapse if implemented throughout the network. This is the problem with handling QoS issues at layer 4, the TCP layer.
The alternative that Ascend is pushing for is to handle QoS issues at the ATM layer. As you mentioned, ATM has some excellent advantages including low latency and high throughput, and in addition it scales very well to large networks. Quality of Service is currently handled with different adaptation layers and connection admission control, and there are improvements being worked on to provide per-VC absolute QoS. Identifying types of IP traffic is a relatively simple task to conquer--IMO much easier than for TCP/IP to scalably prioritize traffic. And the IETF is working on a standard to do label switching (MPLS). Once this is accomplished, we'll have standards-based ATM switches that can read IP headers. Take the best of both worlds (ATM & IP), and you have a router/switch that is ideal for the core of large ISP networks.
I'd like to add that the Ascend 500 switches already can do this, although the implementation is proprietary. And I know that AT&T is using this feature, so it looks like at least one carrier (if not many more) really likes it.
Cisco's approach has been to try to make routers faster. The "tag switching" method is one example of this. But there are weaknesses to this approach in that you hold on to the shortcomings of IP. Tag switching does not provide absolute QoS.
the most cost effective way to
deploy multiservice real time environments is to adapt the technology that's already
there rather than ripping out an installed infrastructure.
Except for some overhead, IP maps quite well to ATM.
Also, I do not think LU or NT need an IP-centric company like Bay. In my projected scenario, routers will be prevalent at the edge, and hybrid ATM/IP switches will dominate the core. But edge routers are commodity products even now. There are a buttload of companies that make them, and there's very little profit in that business anymore (except for Cisco who can charge outrageous prices). It would be a mistake similar to 3Com buying US Robotics.
Thank you for your comments. I am enjoying this conversation.
Bucky89 |
