The Evolution of the Internet - Part 1: Introduction
Just as everyone is recognizing the inevitability of convergence of the world's communications to Internet Protocol packets, DARPA, the agency that started the Internet, is trying to supercede packets with the next generation of smart packets. Smart packets, or so-called Capsules, could make a wholesale change in the way information floats around cyberspace. If DARPA's ideas bear fruit, the Internet will become smarter and more efficient. Investors need to understand the likelihood and implications of changes of this magnitude, since companies like WIND thrive on technologies that underpin the Internet. If upheaval is coming, investors need to know how their companies stand to benefit or suffer.
This post defines and introduces the notion of Capsules, and explains why the concept is appealing to DARPA. Part 2: Conflicting Theories, discusses fundamental conflicts between theories that seem to imply diametrically opposed tradeoffs between processing and bandwidth. That part ends with a guess about how contrasting views will be accommodated without compromise. Part 3: Implementation, presents a roadmap for implementing DARPA's revolutionary idea in the civilian world, with a focus on the likely impact on the subject of this thread, WIND.
An Internet packet is a chunk of bits wrapped in a blanket of Internet Protocol, consisting of a few housekeeping items including the address of where the packet must be delivered. If a client with a low-resolution screen requests a download of a finely digitized Picasso painting, many packets of parts of the painting will be transmitted and ultimately assembled on the client. The client must somehow filter the image before finally displaying a reduced, possibly gray-scale version of the painting. Why did the Internet have to transmit so much data when so little was used? Why did the user have to wait while the full image downloaded at a snail's pace? If appropriate filtering to reduce resolution could be moved upstream, a great deal of Internet bandwidth would not have been wasted and time lost.
One approach to resolving problems like this is to have the client make its characteristics known to the server spawning the picture. This would work very well if this kind of problem always took the simple form of, say, high-resolution versus low-resolution. Then clients always make their resolution type known, and servers dispatch the appropriate image. This is similar to what occurred in the early days of line printers. Fortran programmers lamented the difficulty of figuring the correct number of blank lines needed to advance the paper to the beginning of the next page. They prevailed on printer makers to add intelligence to printers by reserving the first character of each line as a control character, not to be printed. As I recall, a value of one for the character told the printer to advance the page, a blank meant advance one line, while a zero meant do not advance the line (or some such thing).
Once advanced text processing took hold on general processing computers, the Fortran control character proved laughably inadequate for controlling page layout. The obvious solution, of course, was to convert everything to bit maps, and download images to printers, but that solution put too much responsibility on the computer to know all the idiosyncrasies of the printer. It also required excessive communication between the computer and the printer to account for every bit to be printed, especially with high-resolution color printing. Necessity being the mother of invention, a better idea was born at Adobe, namely, the PostScript page layout language. Adobe envisioned a world in which printers are really computers that smartly interpret page layouts to form printed output, rather than print from simple character files like in the old days, or even from bit-mapped files. But page layouts themselves can be so varied, that a language was needed to describe fully all the minutia. As a result, PostScript evolved into a LISP-like language, syntactically complex and complete. Few people realize that modern printers are flexible computers interpreting sophisticated computer programs generated automatically by word processors. This complexity, magnified by the need to support a variety of user options and means of communication, explains why so many of the world's printers require a full-fledged operating system, every bit the equivalent of general purpose computers. It also explains why so many rely on VxWorks. Modern printers are smart, and they implement their intelligence by executing computer programs exactly like regular computers.
DARPA envisions Internet packets changing from being merely buckets of bits with an address to being a computer program containing instructions of almost unlimited variety and scope, along the lines of PostScript. Smart packets, or Capsules, would have numerous uses enhancing military communications. Since U.S. military intends to co-exist with civilian communications, this means that military requirements will help shape Internet architecture whether you like it or not. In civilian application, if nothing else, Capsules promise to reduce bandwidth requirements, albeit at the cost of additional processing.
Smart packets will be constructed on-the-fly by the communications module in tomorrow's computers exactly like PostScript files are generated automatically by today's word processors. The smart packet (program) will be sent to the next node upstream where it will be executed similarly to how Java Applets are executed through interpretation or just-in-time compiling and execution. Bandwidth will be better utilized by eliminating unnecessary, or unfiltered, transmissions. Excessive transmission of non-data bearing programs will be limited by making packet programs persistent, or sticky. Once a program has been transmitted up the line, follow-on packets need only contain data and a reference to the program. Only when the reference cannot be located in the local node is it necessary to request a program refreshment. With caching, there is no reason why bandwidth gains from upstream filtering and transformation cannot be almost perfectly realized in practice.
Smart packets could solve the problem posed by handheld devices limited by small, gray-scale screens and low-bandwidth connectivity to the Internet. The smart packet solution promises a more robust and sophisticated solution than the sort of tricks envisioned today by companies like Spyglass or Unwired Planet, in which special servers interface between formats required by handheld devices, suitably standardized, and regular Internet services. The standardized-preparation solution proved inadequate for printers and is arguably inadequate for communications.
More to come.
Allen |
