Aswath Weblog

Intelligence at the End

Generally it goes unchallenged if anything derisive is stated about PSTN and anything that glorifies IP. It is almost, nay, exactly like the electoral behavior that is happening in US now. In this rant, I am taking on a big challenge of confronting myths that are well established and people who are “rock stars” for having established them. In my opinion this all started with the “Stupid Network” paper by David Isenberg. As you probably know that paper very well. That paper argues the fatal flaws of PSTN network architecture. Even though it has become part of the knowledgebase of every communications engineer, its major claims need to be diluted or its scope narrowed. But the misunderstanding does not stop there; subsequently, many others have invoked this paper and the “end-to-end” paper to state certain behaviors for PSTN with out proper analysis and justification while hiding behind these two papers' authority. The purpose of this long essay is to critically analyze these two papers and bring out certain falsely but commonly held views.

First let us look at the “end-to-end” paper. A common and widely held interpretation of this paper is that this paper advocates that most of the data transfer functions must be done at the end-points. At least that is how this paper is quoted most of the times. But a careful reading of this paper and analysis of contemporaneous papers will suggest a more nuanced claims. A major architectural debate at that time was where to do error correction. The ISO crowd, as evidenced in X.25, had concluded that link by link error correction is the preferred approach. The Arpa crowd had concluded that it has to be end-to-end, but there was disagreement to what is meant by the “end”. You see, in those days, the network interface was an outboard card and if the error correction is done by that card the data could still be lost before it is delivered to the application in the main processor. So it is mandatory that some form verification must be done at the “real end”. Given that it has to be done, so the reasoning went that full verification be done at that end. The “end-to-end” paper makes it abundantly explicit. But if you need further elaboration, I will recommend the RFC authored by Padlipsky. The paper extends this line of reasoning to other functions as well; but it does not suggest that all the functions be done only at the end. It grants that depending on network characteristics and application requirements, it might be optimal to do some of the functions at the intermediate points. But, the paper points out, it should be determined by the applications and the network architecture must allow for a dynamic decision. As such, the paper is more reasoned and less dogmatic, even though lately the paper is used for dogmatic claims. It is more like how Darwin was invoked by “social Darwinists”.

At this time, I want to suggest to those that claim that IP built on ISO model that many of these early architects of IP networking will find it amusing, if not offensive. IP is most decidedly does not conform to ISO reference model, which Padlipsky calls it ISORM (to be pronounced eye-sore-m). So please stop equating the IP protocol stack to ISO stack. The IP architects may get allergic reactions.

Next let us consider some of the points made by the “Stupid Network” paper. To begin with, the implication of the title as to the nature of PSTN is erroneous and the associated claim that in PSTN end points are not intelligent are false. PBX is an example of an intelligent device that is outside the control of the carriers. If you think PBX is an exception, don't you think Fax and data modems are intelligent devices. Here are two examples that destroy the myth PSTN is designed (sometimes this claim is diluted to “optimized”) for voice. It is legitimate to debate on the merits of statistical multiplexing and compare and contrast it to time division multiplexing. But if you grant that transport costs have come down close to zero, the differences between them are not that critical.

The second point made by the paper that I take issue with is its agreement with PSTN carriers that only the carriers are in a position to realize services on behalf of the users. I wish that the paper had taken issue with that claim rather than agreeing to it and conclude that it is an architectural flaw in PSTN. Even when the paper was written, the end had intelligence and TAPI and TSAPI were developed to take advantage of them. Both of them were results of efforts that were independent of the carriers.

The next claim the paper makes that is not totally accurate is that unlike in IP, the end points do not have freedom in selecting the codec. STU III is an example where the end points decide which codec to use for a given call and the Network is not consulted or even notified. Let us even take the “sand that gave rise to the pearl” of the paper – AT&T True Voice. Of course two compatible end-points could have realized the improved voice quality without requiring any further assistance from the network. But that was not the objective of AT&T at that time. They wanted to realize demonstrably improved voice quality between ANY two end points and that too within a short span of time. An implication of intelligent at the end is that new features can be realized at the rate the end devices are adopted by the user community. In short PSTN didn't stand in TV's way; conversely, there is no hope for IP network to deliver on the objectives of TV.

In my opinion, the sad aspect of the paper is that a generation of engineers didn't pay attention to developing intelligent end points for PSTN. Most of the computers as early as early 90's were bundled with telephone applications with address book, logs and voice mail. Now these features are available in VoIP world, but they are services offered by VOIP service providers and not realized at the ATA (for the most part). So where are the intelligent end devices? Where is the angst when one reads about services in the middle when they could be realized at the end? Shouldn't we at least attempt to realize them at the end?