A Practical Navigator for the Internet Economy

Packet Design Has Unique Research Role

Seeks Improved Routing to Offset Problems Inherent in MPLS

Judy Estrin, Kathie Nichols, Van Jacobson Want Cost Benefits of Convergence By Leveraging Strengths of TCP/IP Architecture

pp. 1-7

Packet Design is a venture headed by Judy Estrin, Kathie Nichols, and Van Jacobson. It is the fourth company founded by Estrin and her husband Bill Caricco. Buoyed by their first three successes the founders have the ability to run Packet Design with a business model of a "perpetual startup." As explained by Judy and Kathie in an interview, while Packet Design does anticipate making money by licensing and spinning off startups of its own, it has a much more unique purpose to its existence. Namely it has been formed to do research into improvements of routing. Judy explains that the frantic growth of the commercial Internet from 1995 onward meant that short term quick fixes were taken to the evolution and scaling of routing and backbones. To her dismay as the path to the convergence of voice and data networks, the PSTN and the Internet is now clearly defined and momentum of convergence is increasing, she sees a possibility that we could wind up with the worst of both worlds in the final architecture.

In her own words: "The convergence of data, voice and video we are seeing today is driven by the dramatic increase in data traffic now being pushed across the PSTN infrastructure. The traffic patterns associated with new data applications are very different from those of phone conversations. Yes, the Internet needs to maintain the manageability of the telephony world -but not at the expense of scalability. MPLS, a string-oriented technology, was developed to solve a point problem: integrating local IP and ATM environments. It works well for that use, but its proponents have positioned it as a panacea for all sorts of other problems. At first glance, MPLS seems like the perfect answer to a converged Internet. But it's really just a quick fix. Because its architecture is based on strings rather than clouds, it has all the disadvantages of strings and, in the long run, it creates more problems than it solves."

Judy answered affirmatively our questions as to whether a part of her message was that she believed that an Internet where the circuit-switched orientation of MPLS (strings) played the major architectural role would be more costly to operate than an Internet that was faithful to its founding connectionless philosophy (clouds).

"The reason that we feel so strongly about this is that we believe that it is the properties of IP which can take us directly to the type of Internet that yields the best scaling characteristics at the most favorable cost from a manageability perspective. Now what the telephony world does very well is to offer us some best practices in the areas of manageability and billing and accountability. We need to map these onto the Internet world. But you don't achieve this by making the Internet infrastructure look like the telephony infrastructure. You need to figure out how to do all those things within the confines of clouds."

While Judy, Kathie and Van believe that they have definite ideas that when implemented will produce answers to MPLS and improvements in routing,, as Judy phrased it; we are not ready to disclose them in this conversation. She did however point us to a very significant paper given at NANOG 20 by Van Jacobson and two colleagues.

The NANOG paper is titled 'Toward Milisecond IGP Convergence." See It is by Cengiz Alaettinoglu, Van Jacobson, and Haobo Yu. It suggests that sub-second reroute times would give increased network reliability; support for multi-service traffic (e.g., VoIP), and lower cost/complexity when compared to layer two protection schemes like SONET. Since current IP re-route times are typically in the tens of seconds, the industry should want to do better. There are two choices: replace IP routing with something else like MPLS fast failure recovery or figure out what's wrong with IP routing and fix it.

They prefer to fix the problems of routing. What they are proposing in this paper is a way to decrease the time that it takes changes in outing announcements to propagate within a network from tens of seconds to a few thousandths of a second. If this can be done the operation performance of IP networks will increase enormously. The following paragraph describes some of the paper's conclusions. Because we are summarizing, the arguments that follow are not fully stated. Readers are encouraged to read the entire slide set at the Packet Design URL above.

The Dijkstra SPF algorithm used to compute changes to SPF trees (route forwarding tables) is almost 40 years old. More recent algorithms can compute changes to SPF trees in time proportional to log n rather than n log n. This allows a net to scale up to virtually any size while bringing the calculation time down from seconds to microseconds. "Consequently stable, robust IP re-routing that works at the network's propagation rate (the theoretical maximum for any re-routing scheme) is both possible and achievable. To get there we have to (in rough priority order): (1.) switch to a modern algorithm for SPF calculation, (2) make the granularity of the hello timer milliseconds rather than seconds, and (3) allow different detection filter constants for link up and down events.'

Scaling the Internet via Exchange Points

Many Players Jumping into Rapidly Expanding Global Market

Technology and Business Model Issues as Seen By Equinix

pp. 8 - 15

We examine the world of exchange points which are undergoing dramatic increases in numbers as propelled by the need of ISPs to peer in as many places as possible to reduce transit costs and by the business opportunities opened by the neutral business model exchange where fiber providers, carriers, ISPs, web hosting companies, ASPs, storage networks, SMTP specialists and others can gather together under one roof to do business with each other in an extremely cost effective manner. Well known players in the neutral exchange point market are PAIX,and AboveNet both owned by Metropolitan Fiber Networks (MFN) and Equinix. MFN's AboveNet has six facilities - three in the US and three in Europe. PAIX expects to have six exchanges in the US open by year's end. Calling their Exchanges Internet business Exchanges (IBX), only Equinix caters to the full range of clientele listed above.

An article in the November issue of Telecommunications points out that "companies such as COLO.COM, CoreLocation, Equinix, Eureka, MFN's PAIX.net, and Switch & Data are expanding into major telecom hubs. Unlike the traditional telecom hotel model where carriers lease physical space from a building owner, neutral COs provide air conditioning, backup DC power, HVAC, dust control and high-level security in addition to real estate." In this respect an interesting new player is the NAP of the Americas. Located initially in a Miami. Florida Switch and Data Facility building and managed by Telcordia, the NAP aggregates traffic for a group of carriers which service Latin America.

The Telecommunications article concludes that "while demand will not end soon, there is concern that the colocation industry is heading toward a pseudo space glut with plenty of space but nowhere to connect to the backbones. It is estimated there are 42 national CO providers, with more than 25 million square feet coming on-line next year, a 50 percent increase over current availability."

Of the half dozen or so major players Equinix remains the most interesting. It seems to have concluded that it can achieve economies of scale and enough profits to conclude a very ambitions global build out of 30 very large facilities by building new state of the art centers that are on the order of ten to even twenty times larger than those run by its competitors. Equinix is very wisely taking great care not to run into the trap of insufficient backbone connections but building only in locations where fiber from at least five different providers is available. It has raised over $600 million in capital. It has negotiated a build out schedule with Bechtel that will depend on its ability to finish new exchanges on schedule, populate them quickly with customers and gain adequate cash flow to finance the next buildings on its schedule.

Given the scope of its ambitions it is fortunate for Equinix that it has the greatest technical depth of all the players having been started by the founders of PAIX and having acquired the services of respected infrastructure architects Bill Norton and Sean Donelan. We interview Equinix CTO Jay Adelson and his colleague Lane Patterson. The result is an in depth look at issues such as peering, by pass of ILECs in urban areas, the business models of metro area fiber providers and fabrics for interconnection. Reviewing the myriad of details involved in Equinix's understanding of its industry, the attention that has been given to developing an understanding of the kinds of assistance useful to its customers is quite impressive. An understanding of the economics of interconnection for players of widely differing size assures that a range of options from direct interconnects to access to a variety of switching fabrics is available.

Given the pace of change there are many options open to Equinix customers. The Equinix staff is there to help ensure that the customers derives the greatest synergies possible for his business by making the wisest choices. As Adelson says: " If you are shipping a lot of bits you basically have an economic and technological justification for participating at an exchange."

ITU and IETF in Agreement on ENUM Administration

Letter from ITU to ICANN Blocks .tel gTLD Applications As Competition to ENUM

Administration Modeled on Neutral Tier 1 Database Holder of Pointers to Records of Provisioned Services

pp. 16 -17

The IETF and ITU concludes successfully a set of meetings in Berlin at the end of October. The meetings resulted in a set of understandings that will keep the ENUM domain out of the hands of ICANN and that pave the way for deployment of ENUM provisioning under the e164.arpa domain with each national e164 numbering administration authorized to choose the distribution entity within that countries boundaries. The likely division of ENUM administration in the US into a single neutral Tier one records keeping entity and multiple tier two provisioning entities is explained. Finally abstracts and URLs of the internet drafts relevant to administration and the Berlin meetings are listed. Meanwhile Network Solutions driven by its business interests ignored its earlier chastisement for the premature start of ENUM trials and announced that it was trailing non ascii character domain names before the IETF completed its standardization work. This prompted a public rebuke form ISOC warning that NSI's action will harm the stability of the Internet Domain Name System. See: http://www.infoworld.com/articles/hn/xml/00/11/08/001108hnmultilingual.xml

IS-IS Bug Causes UUNET Route Flap

pp. 18 -19

NANOG discussion of how UUNET's architecture proved susceptible.

ICANN Having No Authority to Create New gTLDs Lacks Legitimacy in US and Is Increasingly Rejected in Europe

Dixon Explains How .eu Has Been Kept from Icann Control

pp. 19 - 21

On November 13 ICANN posted to its web site a response to the law suit against it by the countrycode administator from Belize stating that it had no authority to add new names to the root after all. Such authority, it admitted belonged to the Department of Commerce. See http://www.icann.org/tlds/correspondence/esi-v-icann-13nov00.htm

Thus continued the shell game between the Ameican government and its illegitimate, illegal offsping known as ICANN. We republish a conversation on from the Domain policy list where Jim Dixon shows how Euopeans are becomingless willing to accept ICANN authority. Dixon also explains how .eu was created as a country code domain to keep it free of ICANN's control.

DNRC Letter Documents ICANN Past Testimony to Show Duplicity Behind So Called Clean Sheet Study of Public Board Members

pp. 21 - 22

The Letter calls on ICANN to drop its clean slate Member at Large Study showing that the "study" breaks a two year long series of public promises by ICANN leaders.