SCHMIDT@SUMEX-AIM.STANFORD.EDU (Christopher Schmidt) (09/07/88)
Herewith are some of my observations on the commercial exhibits at AAAI in Saint Paul. Most people I talked to felt that this wasn't as exciting a show as in some years previous. Eg. I didn't notice any new entries in the workstation arena. Software vendors concentrated on porting their wares to new platforms and implementing marginal improvements. For me, no exhibits aroused the visceral excitement of the dead fish tracking system in the TI exhibit last year. One firm did show a system that visually tracked a radio-controlled toy tank in the area. I was pleased that vendors were allowed to actually sell on the floor and didn't have to be coy about prices as in previous years. A number of vendors showed board-level co-processors to support compute-intensive embedded AI products. Eg. two vendors (Topologix, Definicon) sold boards containing 4 Inmos transputers, and encouraged the buyer to put up to eight of these in a workstation. (The Definicon, starting at $25,000 per board can run into some big bucks, but they claim you get 80 MIPS per board. (!)) Sequent was at the show with a machine containing 2-30 80386's. Neither BBN nor Intel showed their multiprocessor platforms this year. TI showed a four-explorer multiprocessor hack that they put together to suit the needs of a small group of customers. The processors share a NUBUS hence memory, but the structure of the bus is not a bottleneck as long as each processor sticks primarily to an associated memory board. The burden of inter-processor communication rests on the programmer's management of memory he declares to be shared. TI also showed the Explorer II Plus, a re-engineered Explorer II processor board clocked at 40 MHz--roughly twice the speed of the Explorer II. Symbolics showed the MacIvory board for the Macintosh. Symbolics and TI seemed to have agreed not to quibble about speed claims for their macintosh board products, describing performance as roughly equal. The current Ivory chip is in 2-micron technology. The sales rep held out the hope that if Symbolics could get a Silicon foundry with the same fabrication technology as TI used in their lisp chip, they could get a four-fold increase in speed with the existing design. In the meantime, they want to complete with TI on the basis of their software. Implementation point: unlike the TI MicroExplorer, which uses private memory, the MacIvory uses the mac's nubus memory. Mitigating the obvious performance penalty is a memory cache on the Symbolics board (on the Ivory chip??). They also showed a VMEbus workstation based on the Ivory chip. After the show, I read that Symbolics has a run-time package for the IBM PC. I didn't see it. Hewlett-Packard and Symbolics both showed synthetic animation systems implemented in lisp. So who said lisp isn't a systems language! Coral showed their standalone application generator and foreign- function interface; otherwise the same product as last year. Lightweight processes will have to wait for the next major release--in 9 months or so. Gold Hill showed their expert system shell ("Goldworks II") running on Suns and Mac II's (on top of Coral lisp in the latter case!). Lucid publicized a service they now provide (called Distill) to analyze a system running in one of their lisps, and generate a specially compacted run-time core image with code and data specially grouped to minimize paging of irrelevant data. A new software company (new to me at least) from Cambridge, England called Harlequin, showed a Common Lisp implementation and what they called a state-of-the-art programming environment. (This meant Masterscope.) They had also designed and implemented their own PostScript-based window system ala NeWS (which they were also selling independently of their lisp). My personal reading of the market is that managers want "standard" window systems rather than better window systems, so their effort is doomed to be even less successful than Sun's. On the other hand, other attendees at the show seemed less excited by X than they were last year, so maybe the pendulum is about to swing the other way?? (I hope!) Maybe then managers will become interested in better programming environments (as opposed to "standard" programming environments)... Speaking of better programming environments... As everyone on this list should know by now, Xerox AIS has reorganized as the employee-owned `Envos', carrying pretty much the entire AIS product line. The big introduction this year was the Medley release which was shown running on Sun-3's and Sun-4's as well as on doves and dandelions. The Sun-based product was achieved by implementing the same virtual machine as used on the d-machines, but in C instead of microcode. This approach provides an extraordinary degree of compatibitily, as you might expect. It also means that Medley doesn't require extraordinary amounts of real memory (typically associated with other lisps on the Sun), because of the compact instruction set and garbage collection strategy. Speed on the high-end configuration demonstrated (a Sun 4/260) was described as a little over twice that of a dorado. Envos also introduced Rooms as a commercial product. Both Rooms and LOOPS were announced to be shipable at the same time as the rest of Medley (October 1st). (!) The booth was the most visually dramatic on the show floor--sort of a giant life-sized version of the magenta and teal announcements sent in the mail. The price of Medley (for Sun or d-machine) is $10,000 per copy ($7000 for educational customers). LOOPS is $7000 ($4900 educational). Rooms is $2995 ($2100 educational). Special pricing is available for government and quantity customers. Owing to some confusion, there were two different users group meetings. I attended the Monday meeting. Most of the meeting covered the makeup of Envos and products (summarized above). People from a couple of sites who had beta-tested Medley on the Sun talked about the experience. Both said it was not much different from moving between earlier releases. I asked about source availability. The answer was that they didn't plan to make sources available, at least not for the old price. There was no discussion of RS232 or FX80 problems!! Advice on what Envos should do next was solicited. I asked for a Mac II product (and wasn't the only academic with this wish). Some suits voted for an 80386 version. As usual, I invite other readers to comment on the show and/or make corrections/additions to what I have said above. --Christopher -------