Print 

JUNIPER M40 OUTPERFORMS CISCO AS NEW TERABIT ROUTERS START TO COME ON STREAM

-- TONY LI ON JUNIPER DESIGN & ISSUES OF INTELLIGENT TRAFFIC MANAGEMENT AT CORE OF STUPID INTERNET,

pp. 1- 8

Tony Li, one of the designers of the new Juniper Networks terabit routers, in his first interview, explains the major issues and innovations in the design process and discusses the problems of intelligent traffic management at the core of a "stupid" Internet.

After its initial stage of VC funding, Juniper partnered both with several carrier and equipment companies: Ericsson, Lucent, Nortel, 3Com, and Siemens/Newbridge, and with end-user investors, such as AT&T Ventures, the Anschutz Family, and UUnet. The partnering gave them early feedback during both the design and implementation phases of development. The partner's investment in Juniper also helped to cement the partnership, ensuring a mutual commitment to the process of bringing the M40 to market.

Juniper has built a router that separates routing computation from packet forwarding by carrying on the two functions in parallel. It has also taken advantage of custom made ASIC (Application Specific Integrated Circuit) Design to build a router that achieves a look up rate of 40 million packets per second with a single chip (ASIC). Juniper's innovation in its switching fabric is in an increased efficiency and ability to use less sophisticated technology to implement the same size fabric as compared to other, traditional crossbar designs. It gained an efficiency that translates into fewer parts and a more efficient utilization of the parts that are used. The result is less costly less to the end-users because it's less costly to produce. The use of a single system wide buffer also holds down costs.

The M40 makes significant use of MPLS which provides a hybrid network architecture, where one can support, at the same time, both datagram mode forwarding and a connection-oriented service. It offers the hybrid strength of being able to route packets through the network based on something other than the destination address. This gives the network an major degree of increased flexibility, which makes it possible to provide new services such as voice over IP and VPN services on an integrated Internet backbone.

Unfortunately, there's a conflict between the amount of separately defined QoS levels one wants for customers at the edge of the network and what one can actually aggregate and deal with by routing on the backbone. This is being dealt with in two ways. First is the application of MPLS to aggregate flows with QoS and routing properties. The second approach is the work that is being done in the Diff-Serv working group in the IETF. Diff-Serv provides a convenient and easy way of "coloring" packets with particular levels of QoS requirements. These are bulk, generic kinds of "coloring" that are not flow specific and thus require minimal amounts of state within the router forwarding function.

The current Diff-Serv proposal supports up to 64 different "colors", or more precisely, 'Per-Hop Behaviors' (PHBs). Some of these are defined to be global, some are reserved for local use. It's left to each provider to determine the PHBs that are applicable for the services that they wish to deliver. However, because the definition of a PHB can in fact be wholly local to a provider, it raises an interesting question about the ability to define and deploy interprovider Diff-Serv functionality.

Differentiated services within a single provider's network is of interest to the provider, but to the end user, who wants to use the Internet backbone as a global facility, differentiation with a limited scope isn't a practical solution. Some providers are expecting to differentiate themselves based on their domain-specific services and are not interested in supporting globally defined services. This would present a drag on the deployment of global services that could be overcome only when there is a de-facto standard and the non participants are the exception.

It is possible that these scenarios can be avoided, but it will take initial experimentation with interprovider QoS agreements, in both bilateral and multilateral agreements. An appropriate forum where such issues could be aired and global service definitions discussed without antitrust problems is needed.

SCOPE OF IP TELEPHONY PROTOCOL DEVELOPMENT EXPANDING BUT SPEED SLOWING

pp. 8

MCI'S VBNS SERVICE EXPLORES QoS

-- RICK WILDER REPORTS ON TESTING OF JUNIPER M40S AND ON FUTURE OF BANDWIDTH INTENSIVE APPLICATIONS

pp. 9 - 15, 24

In 1995 the NSF funded MCI to provide a very high speed backbone service to connect the national supercomputer centers. Given the administration's interest in promoting Internet2 and the NGI, the vBNS effort has now moved to a focus on prototyping the next generation internet backbone. We interview Rick Wilder, Director of Engineering for the vBNS.

On part of the vBNS Wilder is testing RSVP sessions mapped onto ATM swithed virtual circuits. Wilder is also using RSVP in setting up label-switched paths through the backbone. The application here is traffic engineering. For, as traffic conditions in the network change, it's possible to easily change the loading of paths on the fly.

Wilder expects to see decreasing use of ATM on major internet backbone not just because of the well known cell tax but also because the amount of complexity in the router interfaces needed to do the segmentation and re-assembly, i.e. the conversion of packets into cells and back again. The complexity is actually in the hardware and/or firmware and this, in turn, requires that the interfaces themselves be more complex and more expensive. When MCI went to OC3 routers based on ATM, it took longer to get reliable router interfaces than it did with DS3 interfaces or with Packet over SONET (PoS) interfaces which is what it is currently running.. With PoS they will use MPLS, at least initially, to replace the traffic engineering capability they currently have with ATM.

Having begun tests on the Juniper M40 in their labs last summer, they started to use them to run production traffic on the vBNS in early February. They have not noticed any problems arrise from the trial and are generally happy with the way things have gone. Wilder points out that DWDM could be used to provide QoS. He adds that ISIS (Intermediate System to Intermediate System) routing protocol is probably the protocol where the most QoS routing work is now being done in the IP world at the IETF. That's because its being used by most of the very large ISPs for use in their backbones today. The IETF working group chaired by Tony Li is doing a lot of work in the area of adding QoS differentiation to the routing.

He finds that QoS in a connectionless Internet packet network can be achieved. It's just that conceptualizing the problem so far has been a daunting task. And for very good reasons. Breaking it down into its component pieces and getting enough different efforts working together to build the whole architectural solution for a problem as complex as this one doesn't come quick and easy.

TWO VIEWS OF ELECTRONIC COMMERCE

-- EYEBALLS VERSUS THE CLUETRAIN MANIFESTO TV WITH A "BUY" BUTTON OR SELF ORGANIZING MARKETS?

pp. 16 - 22

We review a study on e-commerce "Portals To Profit: E-commerce Business Models and Enabling Technologies", published on April 16 and available from www.datacommresearch.com.

According to one of the co-authors: "Our main thesis is that e-commerce competition will transform all commerce, destroying many if not most traditional business models, and forcing companies to invent new ways to make money. Thus, it is imperative that businesses understand the new models, many of which involve cost-based pricing, below-cost pricing, auctions, reverse auctions, ad-targeting, etc."

So far so good. But the report goes on to make its point through the eyes of the Fortune 500. E-commerce is primarily about size and, as such, is a race to see how the large corporations can translate their standard views of the world into this new medium. The views are basically the old industrial age mechanics of economies of scale using, this time, the latest digital technologies and suites of software agents behind the scenes to manipulate and shape the thoughts and response of the customers. The customers are portrayed mechanistically as sets of "eyeballs" on which the new technologies act.

It certainly represents a legitimate point of view and its focus on automated mechanistic ways of dealing with the masses describe a modus operandi that may be all that is needed for success. After all 90% of the people on the net have been there less than 2 years. They may never be able to see the Internet as much else besides TV with a buy button. However another view exists. The Cluetrain manifesto suggests that the Internet makes a whole series of special relationships between customer and businesses possible. It is the view of Chris Locke. Someone who has, as have we, been on the net for nearly 20 years. We believe that, the longer one is on the net, the more the experience broadens one's horizons.

In this world view the web is driven by the corporation's customers who can join together to route around companies that can't or won't meet their needs. Web enabled consumers, according to Cluetrain, are there not to be manipulated but to join together in efforts that will replace those companies that: "don't get it." Cluetrain represents both a culture and ways of looking at Internet commerce that business would be very much ill-advised to ignore. Locke had the following to say in a lengthy interview with the COOK Report. "My experience with MecklerWeb was five years ago. A lot has happened since then. The same dynamics are still there. Only today they are orders of magnitude more powerful. But no one who really understands these dynamics is talking about them ‹ except for little e-zines out on the fringes of the net. "