| robertshaw.info Telecoms, Internet and Convergence Fred Goldstein's Comments on the End-to-End Stupid Network Arguments From: Fred Goldstein <fgoldstein@WN.NET> He makes a big deal about the "end to end" argument, and how the intelligence is outside the network, rather than inside as he thinks it is in the phone network. In that regard, he's taking a line from Isenberg, who's made a business for himself promoting his "stupid network" slogan. The trouble is, Isenberg has it backwards. Isenberg's current "stupid" career began after he left AT&T, where they were working on the "Intelligent Network" (IN). As I explained to this very list on 2/28/98, IN is about moving the intelligence out of the network switches and into external processors, so that network operators can upgrade the network without new switch generics. It's orthogonal to "end to end" intelligence in that it's used for control of call setup, rather akin to the DNS and the like. The telephone network has even more content-neutrality than the Internet, because as a circuit-switched network, it has zero visibility of the bearer channels. Once the call's set up, bits is bits. No firewalls, censorware, caches, or other content-invasive intermediaries a la the Internet as people tend to see it nowadays. And since it's connection-oriented, it doesn't waste its time setting up a new path every time a packet arrives. Thus the amount of intelligence in the Internet's switches (routers) is many orders of magnitude above what goes into a telephone switch, even a huge one. (The biggest Nortel DMS-100/250/500 has only a Motorola 88000 microprocessor to handle all common control functions, with some slower chips embedded for real-time peripheral handling. The DMS-10 is upgrading to a PowerPC. The 5ESS has, I think, a couple of homegrown CPUs, but the 5E-VCDX replaces the central one with a Sun SPARC workstation.) The "end to end" phrase came about a couple of decades ago when TCP/IP was being contrasted with the PTTs' own favorite data networking protocol, X.25. An X.25-based network has hop-by-hop *and* edge-to-edge "reliability" built into the network, with numbered packets and retransmissions. This probably sounded like a good idea in 1976 when 33ASR teletypes were the main users, but it rapidly proved to be inefficient. TCP does end-to-end retransmissions; the IP routers in the middle don't guarantee deliver and don't care. In general that's a good idea. But "end to end" only concerned the reliability of transmission. The phrase has lately become magical, but that's based on ascribing it meanings that it never really had. No, but there was one single most important regulatory act that made the Internet possible. Again from the mid-1970s, it was the FCC's revolutionary "sharing and resale" order. That overturned AT&T's requirement that "private lines" only serve a single customer. Common carriage was thus extended to inter-company data circuits. The early packet networks had to be FCC-tariffed common carriers in order to take advantage of AT&T's "service to other common carriers" tariff. But with the sharing-and-resale prohibition blocked, a private sector untariffed ISP (or other ESP, before the Internet itself was public) could order up a line from its premises to a subscriber's, or to a peer's. Without that, no Internet, period. I think that restriction still exists in a few backward countries, where ISPs thus need to be licensed and regulated. <quote of my post from 1998> My beef against Isenberg? He is not making a fair or accurate comparison of the Internet and telco-switching What is IN used for? Primarily it's used for directory-like services. This is where you have to realize that the E.164 number space serves three separate functions: Name, address, and route. The first one to get big was 800 numbers. Dial an 800 number (really a kind of name) and a LEC switch does an IN database dip to find out what IXC owns that 800 number. That database is open to all carriers. (This provided 800 number portability.) Then the IXC, getting the call, does its own database dip (private database), which tells them what to do with the call. It might translate to another E.164 number (address). Or it might translate to a route and/or a non-E.164 address (switch port). That translation might be dynamic, based on things like time of day, current level of load on different potential destinations, where the caller is, etc. The big 800 providers (mainly AT&T and MCI) make tundishloads of money selling these features to big call receivers. They're regular services now, in the price books. A second use of IN technology is local number portability (LNP). That takes a dialed E.164 number (name) and looks up what actual CO it belongs to, returning an E.164 address (route) for that CO. For instance if I port a number from Bell's Back Bay CO (617-266) to Global NAPs' CO, I can keep a 617-266 number, but the LNP database returns a "location routing number" (again, in E.164 format, say 617-207-xxxx) which identifies the switch. The dialed number is now resolved to an E.164 address in that switch. This method was mandated by the FCC to enable users to change local telcos without changing numbers, just as happened with 800 several years ago. (BTW this was hardly done voluntarily by the ILECs!) I hardly think that this is the sort of thing that Internet fans should be pooh-poohing. Indeed, it sounds a little better than the "dumb-ain name system", no? Nothing in Isenberg's "stupid" network does away with the need for name and address resolution. His other key point is that the phone network is only good at providing ones service, circuit-switched 64k/voice calls. True, but the Internet is only good at providing one service, connectionless "send and pray" packets. They're both useful services, but are optimized for different applications. His postulated "stupid" network would, I supsect, require far more complexity and intelligence than the phone network, Maybe the telco-dominated definition of ATM didn't fill the bill quite right, but nothing else does either. <quote of my post from 1998> |