jml@druny.UUCP (LeonJM) (03/03/84)
- First, a disclaimer. The opinions expressed herein are my own and do not necessarily reflect those of AT&T. (Love those watchdogs) Normally I just sit back and watch as everyone rants and raves about this or that. However, this time I'm in the mood to foam at the mouth myself. This is in response to Gary Little's remarks about his favorite microprocessor. Gary's remarks were from a programmers point of view. I would like to respond from a different (emphasis on different) point of view. That being a user of general purpose UNIX based microcomputers. First a comment about that "varmint" Zilog Z8000 family of microprocessors. The first commercial microcomputer to run UNIX System III was the C8002 from Onyx and uses the Z8002. We have one and it was the first of many UNIX microcomputers we have purchased or used on loan for the past 18 months. Onyx has sold a bunch of them. As of June 1983, my numbers indicate that Onyx had sold more UNIX microcomputers than any other manufacturer. I don't have current numbers. Our location has done an extensive study of UNIX microcomputers since 1982. This study included, running benchmarks, porting various special programs, testing hardware, examining architecture, and having friendly users bang away at the machines in their normal work routine. The users, by the way, were all software or hardware engineers, most with extensive UNIX backgrounds. This study was conducted with more than 20 of the popular UNIX machines available and is still ongoing. One of the machines tested was the C8002. It was purchased in 1982 and still compares favorably to other similar micros, (Similar meaning: single cpu, 1MB main memory, at least 40 MB winchester, > 8 serial ports, etc). Since Gary uses an elegant instruction set (debatable) as the basis for touting the MC68000, I feel free to talk about benchmarks. When you start talking about individual instructions then cpu benchmarks become relevant. (Of course the compiler is what is really important. Unless all you ever do is program at the assembler level. A bad compiler can destroy a good microprocessor's abilities.) Of the approx 20 UNIX machines we tested, the three fastest micros were all based on a different processor. Namely the 8086, 68000, and the Z8000. All three machines were virtually identical as far as the benchmarks were concerned. Why is it then, that the lowly 8086 and Z8000 based microcomputers compare favorably with the mighty 68000? The reason, I believe, is implementation. How the different manufacturers' chose to use the respective microprocessors was the telling feature. Our study of UNIX micros has come up with many conclusions. The study has also produced enough memos to make a book. This news item is necessarily a small reflection of our findings and therefore is filled with generalizations. So, please don't assume that we didn't think of some obvious counter arguments just because it wasn't mentioned here. One of my conclusions was that as far as the users were concerned, it didn't matter what microprocessor was used. Moreover, it as far as throughput was concerned, it didn't matter which microprocessor was used. In fact, which winchester disk used was more important! The disk subsystem was always the bottleneck with a UNIX system. I'm sorry Gary, but in reality, system response on a UNIX system has little to do with which microprocessor it has in it. Also, some of the 68000 based micros were buggy and users generally didn't particularly want to use them. Now it may be that the people that implemented UNIX on the 68000 machines had an easier time, however, the end result is what is important to a user. I realize that Gary was talking as a programmer and about which chip he would rather program for and on. I do believe that the 68000 does have a lot of potential. In fact, the most sophisticated microprocessor based computer I use is based on the 68000. However, my point is this: (1) Potential is nice, but achievement is better and (2) The microprocessor is only as good as its supporting cast. It's the whole I'm concerned with, not the parts. By the way, the memos I mentioned above, any specifics about particular micros, specific numbers (such as benchmark results) and conclusions are not available to anyone outside of AT&T and are even limited within AT&T. (They paid big bucks for the answers and therefore want to keep them.) John Leon AT&T Information Systems Laboratories 11900 N. Pecos St. 30E28 Denver, Co. 80234 (303) 538-3501 druny!jml
jml@druny.UUCP (LeonJM) (03/03/84)
- The Subject line of my last article should read: re: My favorite chip instead of re: fav John Leon AT&TISL