To: Tunica Albuginea who wrote (12505 ) 8/2/1999 2:08:00 PM From: Elmer Flugum Respond to of 18016
PERSPECTIVES August 2, 1999 From Telephony Magazine:internettelephony.com The MPLS raison d'être JOAN ENGEBRETSON Quick networking quiz. The purpose of multiprotocol layer switching is: a) to eliminate ATM switches in IP networks b) to bring better-than-Sonet protection to networks based on packet-over-wave division multiplexing c) to enable IP networks to support quality of service-based offerings such as virtual private networks d) it depends on whom you talk to. Answer: If your reply was d, you have correctly detected the lack of consensus about MPLS, the heavily hyped networking protocol. Networking quiz, Part 2. Match responses a, b and c above with the IP backbone operator that provided each one: Frontier Global Center, UUNet, Williams Communications. Answers: a) UUNet, which has begun to implement MPLS in Europe, is one of the earliest adopters. But the ISP's plans for the protocol are not particularly ambitious. Initially, MPLS will operate parallel to UUNet's existing IP-over-ATM network. The goal is to phase out ATM worldwide, says Mike Odell, UUNet's senior vice president and chief scientist. Like many ISPs, UUNet originally may have turned to ATM switches because at the time, they offered higher port density or higher-speed interfaces than routers could provide. But sooner or later, the company found itself using ATM to overcome a different weakness of routers: Typical routing protocols look for the shortest open path between two points--but in some cases, a longer path may be less congested. ATM lets ISPs shift traffic to that alternate path, making more efficient use of network resources. The downside is that the cumbersome conversion from IP to ATM can prevent a backbone from operating at its full line rate--a phenomenon that becomes more pronounced at higher speeds, Odell says. UUNet hopes that MPLS will provide the benefits of ATM without the penalties. Like ATM, MPLS can establish alternate paths between city pairs, and by eliminating IP-to-ATM conversion, Odell expects to upgrade to an OC-192 backbone that achieves the full 10 Gb/s line rate. b) Frontier Global Center also has begun to implement MPLS and expects to complete the project this month. The pioneering ISP seems to have a penchant for deploying evolving technology--some might say the company works without a safety net (no pun intended). But that doesn't phase Alan Hannan, Frontier's director of network architecture. While many competitors plan to move to packet-over-WDM in the future, Hannan says Frontier already has deployed a network based on that architecture. By eliminating the equipment associated with extra Sonet and ATM layers, he says the design minimizes cost and simplifies network management. Packet-over-WDM uses Sonet framing but not Sonet protection. Currently, that means restoration time "rivals Sonet," Hannan says. Soon MPLS restoration time will beat Sonet, he says, because routes will be restored locally: By eliminating the need to communicate with a path's endpoint during restoration, MPLS will eliminate speed of light latency. Like UUNet, Frontier also is using MPLS for traffic engineering by establishing virtual paths between city pairs. c) MPLS virtual paths also can be used to support differentiated quality of service--a capability both Frontier Global Center and Williams plan to tap. By combining MPLS with DiffServ, another emerging standard that enables carriers to assign priorities to different data streams, Williams eventually plans to establish several parallel virtual networks, says Mark Allen, Williams' director of optical networking. A high-priority virtual network could be reserved for real-time, delay-sensitive traffic, while a lower-priority network could carry less critical data. The company currently uses ATM-over-WDM but is transitioning its IP network to operate directly over WDM, says Allen, who does not envision implementing MPLS for a year or so. He feels network management tools have not evolved far enough to support the volume of virtual paths that would have to be established customer by customer. Frontier's Hannan argues that such tools are not difficult to develop. It's worth noting, though, that Frontier's initial QOS plans, targeted for a few months from now, call for differential services on an aggregate, rather than a customer-by-customer, basis. One last quiz: MPLS is a) ready to deploy b) not ready to deploy. The answer, it seems, depends on what a service provider hopes to accomplish--and how badly.