A Practical Navigator for the Internet Economy

Provider Based CIDR Likely To Impede Progress Of Smaller Players, pp. 1-8

When it looked like the availability of Class B addresses would soon be exhausted, CIDR was developed several years ago to permit fine-grained allocation of network addresses so that more networks could be supported globally. Although advocates point out that it was always intended to slow the growth in size of the Internet's routing tables, it is really only in 1995 that CIDR has also become the principal means of slowing down and indeed nearly halting the runaway growth in number of routes advertised to the backbone routers of the global Internet. It had become necessary to do this because the total number of routes advertised was doubling in less than a year while the doubling of hardware software capability needed to handle route doubling takes between two and three years.

While current routers can handle upwards of 60,000 routes, and the current number of routes advertised is between 30 and 35 thousand, the consensus of the technical community is that further growth in number of advertised routes should be severely constrained because the current buffer in number of routes advertised and equipment capability is necessary to dampen route flaps. Flaps can occur when changes in network routing occur. If a major backbone router goes off line, or especially if a peering session is suddenly dropped, the other routers have to reconfigure alternative paths for traffic directed at the router in question and propagate the reconfiguration to the farthest corners of the network. Think of the ripples caused by a stone thrown into a pond. Should that router come back on line or other changes in the dynamic configuration of the network occur before the first changes have propagated to the corners of the network, we will have a new set of ripples colliding with the previous set and routers "flapping" instead of routing. Increased computing power in routers is seen to be the best way for dampening the impact of such flaps. New router code from Cisco apparently is also helping.

Provider based CIDR is seeking to prevent routing table growth by insisting that users and downstream ISPs get network numbers from their larger upstream suppliers and aggregating their routes so that smaller set of routes needs to be announced to the entire Internet for their hosts to be globally reachable. If this technique is to be maximally effective, suppliers are finding that they must insist that anyone who changes providers must accept IP numbers from the CIDR blocs of the new supplier rather than continue to use the numbers initially assigned. If these numbers from the previous provider were used, then the new provider would have to announce them globally as a series of new routes adding to the burden of routing in already overburdened backbones. Consequently, the customer is forced to renumber his hosts -- an act that is time consuming and costly -- and in extreme cases so burdensome as to be prohibitive.

Such policy tends to favor the larger providers who can give the best reassurances to customers that they will never have to move to another supplier and would be able to survive fees based on the number of routes they ask those with whom they interconnect to carry should this practice begin to occur as an additional means of holding down the growth of routing tables.

Dave Crocker has been a severe critic of provider based CIDR - on the CIDRD mail list and elsewhere -- suggesting that policies being followed would put the squeeze on small ISPs and advocating investigation of geographic based rather than provider based CIDR. In an interview with us he extensively describes the problem and advocates the testing of some geography based alternatives. While the chances of developing the geography based alternatives do not look to be good, he also points out how the current practice of multi-homing, that is an essential part in the growth of small ISPs, causes new routes to be propagated limiting thereby the benefits of provider based CIDR. As a result multi-homing could become more difficult and/or more costly to do in future. Noel Chiappa, Dave's most consistent critic, has provided us with a critique of Dave's position. The article concludes with Noel's response to Dave, and Dave's comments on Noel's response.

We publish this debate less to try to ascertain who is correct than to highlight the warning signs of the impact of network growth on smaller providers. Those who survive will be those who most promptly factor correctly the economic and operational implications of these developments into their business plans.

Asked To Comment By Maine Public Advocate We Find NYNEX Plan Not In Public Interest pp. 9 - 14

This summer NYNEX presented the Maine PUC with a plan to attach all Maine public schools and libraries to the Internet during the next five years - spending $20 million to do so instead of returning the $20 million in overcharges to rate payers.

The NYNEX plan proposes a centralized, one-size-fits-all 56 kbs solution for the entire state. NYNEX would install and control all parts of a frame relay cloud and the Internet links there to. Schools and libraries would be hung as "dumb" leaves on the branches of the network. The result is a poor one for the citizens of Maine. But, unlike Indiana where state government has assisted Ameritech in implementing a solution with similar intent, citizen groups in Maine are petitioning the PUC to set up an independent oversight board to assist independent ISPs in working with NYNEX and local communities to establish a decentralized network owned and controlled by local communities. If Maine successfully pursues its current direction, we believe it should become a model for the rest of the nation.

Discussion Of Transient InternetMCI Routing Problem pp. 15-18

When an SSE synchronization bug caused, from the perspective of Westnet, a several hour partition between MCI and Sprint on October 4th, we posted notice of the difficulty to the North American Network Operators list and asked what had happened. We received a torrent of discussion that showed concern with the way the majors are interconnecting at the NAPs. However it also showed that the Internet ethic of problem solving cooperation among service providers is alive and well.

NTIA Funds MercerNet, pp. 19-20

In the spring of 1995 MercerNet started of as a continued attempt to get a full motion multipoint video network into Mercer County schools and libraries in order for Princeton and other wealthy bedroom communities in the county to be able to continue to offer advanced placement courses to their students in the face of cutbacks in state aid. The plans did not include Internet and seemed have little to do with education or community development except in the most elitist way.

At the beginning of the summer Comcast became involved with the program and decided to invest in it as a testbed for a product that it would seek to market to its other franchises. As a result the project was broadened and Comcast made a substantial investment in bringing Internet to all 21 participating locations. It obligates Comcast to build a dedicated fiber based video network and a coax based Internet at Ethernet speeds network to 12 schools and 9 branches of the county library. Comcast will pay all installation costs, the Feds will pick up the video classrooms, and Comcast will pay the monthly connect fees for the first year. The project now can do some good things at the county level, if the planners open their activities via the Internet to county residents. In the three weeks since the award of the grant there has been, unfortunately, no public communication from the awardees.

Internet Society - Tony Rutkowski, Mike Roberts And Rick Adams Discuss The Role Of The Charter Members, pp. 20-21

The Charter Members seem to have been an after thought as Harms, Kahn and King acted as individuals to incorporate the society. Privileges granted CNRI, Educom and RARE in June 1992 are source of discord.

MCI's Internet Business Model As Depicted In Internet Week Interview Differs From Vint Cerf's Position In Sept. Cook Report, pp. 21 - 22

According to a summary posted from Japan's Glocom Institute on Dave Farber's IP list on Oct 14 and 16th, Stephen VonRump, MCI's VP of Data Marketing, in an interview with Internet Week, expressed the view that MCI was working to introduce settlements similar to those among telcos with those with whom it peered. And that MCI was "convinced that such agreements and settlements systems are necessary for the continued growth of the Internet" because "for the Internet to sustain its growth, commerce and business really need to trust the Internet to carry traffic." According to Stephen Anderson of Glocom the article further stated: "MCI is likely to go to the FCC for regulation to seek improved Internet quality and standards. MCI will use standards issues to wage competitive war against smaller ISP companies that may not be able to invest in the large capital commitments needed to meet high standards." VonRump has been in touch with us and we are asking him to explain, the differences in these positions from the more reasoned ones taken by Vint Cerf in his late August interview with us. Expect a follow up next month.