Talk:DEC PRISM
![]() | This article is rated Start-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects: | ||||||||||
|
Comments
[edit]The narrative blurs significantly beginning at the third paragraph.
The PRISM specification was developed over a period of many months by a five person team: Dave Cutler, Dave Orbits, Rich Witek, Dileep Bhandarkar, and Wayne Cardoza. Existing RISC architectures influenced this team, as did the Cray instruction set, MIPS not more to my knowledge than any other. This work was 98% done 1985-1986 and was heavily supported by simulations (written and run by Pete Benoit) on a large VAXcluster at DECwest.
The MIPS R3000 came out three years later, in 1988. Sample silicon was likely not available until late 1987, if then. The Advanced Computing Environment consortium was two to three years after that (1990-91?).
DEC struggled throughout the mid-to-late 1980s with how best to respond to RISC, UNIX,and workstations. One group advocated sticking with the VAX, and taking it into mainframe OLTP; others to adopt something standard (and generic) like MIPS for short-term sales, others to develop a proprietary RISC-Y VAX successor, like PRISM or Alpha. The R3000 contingent had their own "next big thing" effort -- when it did not save the company (due largely to DECnet imposed incompatibilities with non-DEC MIPS-based systems) and the VAX 9000 was underwhelming, PRISM morphed into Alpha.
The PRISM team was pulled this way and that throughout 1985 to 1988 by the debate within the company. From one week to another PRISM was ECL or CMOS, 32-bit or 64-bit, a vector-oriented system, a cluster-centric database accelerator, or something else as the factions within the company tried to make it fit into a common world view (as well as existing VAXclusters.)
I believe Dave's frustration with this period at DEC led to his decision to go to Microsoft.
The seriousness of DEC's situation later in the 1980s created the concensus that allowed Alpha to emerge.
John Coombs
PRISM Hardware Product Manager
DECwest / Digital Equipment Corporation 1985-1988
— Preceding unsigned comment added by Jrcoombs (talk • contribs)
- In retrospect I see how the wording is unclear in terms of the "MIPS influence", which seems to be what I was saying. However it does seem that some sort of comparison between PRISM and MIPS is needed, because the two archetechtures are very similar -- and contrast, for instance, with the other well known system of the era, RISC/SPARC.
- By the way, does the Alpha Implementations book have GOOD details on the history of PRISM and earlier attempts? I'd love to flesh this article out a lot more, but the book is about $120 so... Maury 23:18, 12 July 2006 (UTC)
- Maury, I believe the Alpha Implementation book (we're both referring to Dileep's book right?) offers a good balanced history as well as some good comparisons of Alpha vs. other RISC architectures. Perhaps you could get it through your local library or interlibrary loan? John
- Gentlemen. I just (2014-05-24) checked www.bookfinder.com and discovered numerous reasonably-priced copies of "Alpha Implementations and Architecture" and "Alpha Architecture Reference Manual" (I have no direct or indirect relationship with this website) Neilrieck (talk) 12:04, 24 May 2014 (UTC)
Follow-up
[edit]Too bad this seems to have been ignored for five years. This is an important chapter in history that should be told accurately. The usual problem is that internal corporate projects often do not have reliable sources we can verify, so this seems heavy on personal oral history. It might clarify things if we make explicit what was done where. As I recall there were four DEC "west" labs. Cutler's was commonly called "decwet" since it was in Seatle (that was its uucp name at any rate). DEC SRC was the group headed by Bob Taylor with other former Xerox people. DEC WRL was headed by Forest Baskett formerly of Stanford, and they did Titan and multi-Titan, I think. Steve Borne (one of them) headed the DEC WSE group, which I think did the intirim MIPS DECstations? The others I think were done on the east coast, and am not sure of those (my COI was that I knew some of the west coast people). And should certainly mention the connection between Mica and NT, at least notable people who worked on both.
Given the extra embolding in the article, is really abot more than just PRISM, but the other RISC projects as well. Not sure if Titan is RISC, so it is more like "1980s DEC research architectures"? Can keep the current title for now, maybe have some redirects or at least references for the others. W Nowicki (talk) 23:07, 6 April 2011 (UTC)
The article is wrong about how MIPS got chosen
[edit]I was there. I did the MIPS SW port originally. This paragraph needs to be fixed:
Frustrated with the growing number of losses to cheaper faster competitive machines, independently, a small skunkworks group in Palo Alto, outside of Central Engineering, focused on workstations and UNIX/Ultrix, entertained the idea of using an off-the-shelf RISC processor to build a new family of workstations. The group carried out due diligence, eventually choosing the MIPS R2000. This group acquired a development machine and prototyped a port of Ultrix to the system. From the initial meetings with MIPS to a prototype machine took only 90 days. Full production of a DEC version could begin as early as January 1989, whereas it would be at least another year before a PRISM based machine would be ready.[7]
It should be something like this:
Frustrated with the growing number of losses to cheaper faster competitive machines, independently, a small skunkworks group in Palo Alto, outside of Central Engineering, focused on workstations and UNIX/Ultrix, entertained the idea of using an off-the-shelf RISC processor to build a new family of workstations. While the group looked at x86, 68K, MIPS, there was no SW solution for them to make a decision. Any choice was possible. Totally independently, the ULTRIX Advanced Development Group in Nahsua, NH obtained 2 MIPS M1000 workstations and did a port of little endian VAX Ultrix to the big endian MIPS M1000. Sometime late in March the plan started to come together with a now working ULTRIX on MIPS HW, it was easy for the marketing and sales side to come up with a plan to build a single board MIPS little endian workstation and get the ULTRIX SW team's backing to port and support the software and the needed apps to make a product. The group carried out due diligence covering HW and SW plans and support and applications. It all made sense, finally. From the initial meetings with MIPS to a prototype machine took only 90 days. Full production of a DEC version could begin as early as January 1989, whereas it would be at least another year before a PRISM based machine would be ready.[7]
- Neither x86 nor 68k are RISC, so if they considered x86 and 68k, as well as MIPS, it should be "entertained the idea of using an off-the-shelf processor...". Guy Harris (talk) 22:25, 1 May 2021 (UTC)
(From Mashey): Indeed the explanation is wrong. Here's more. We of course knew DEC had been experimenting with RISCs for a while. (Among other things, Forest Baskett had worked on Stanford MIPS, then went to DEC Western Research Lab in Palo Alto, then over to SGI, which was MIPS' closest customer. We knew the RISC fans inside DEC were pretty frustrated as strategies kept changing and people kept getting grabbed to work on faster VAXen, given the huge cash cow and goodness of having consolidated around single architecture, given end of DECSYSTEM-20s in 1983.
I attended a computer show in Washington DC in January 1988, and we had a MIPS M/1000 system (15MHz R2000) there. I'd known Bob Rodriguez at Bell Labs, he'd moved to DEC in Nashua, NH (home of ex-BTLer Armando Stettner's UNIX-Live Free or Die license plates) and was working on ULTRIX. Bob came by our booth, asked about MIPS and I gave him a MIPS Performance Brief, Issue 3.0, October 1987, which had many comparisons with other machines, including VAXen. Later, he came to me at the evening beer bust, saying "These are fast!" (They were, a deskside typically about 1.5-2X faster than a high-end ECL VAX 8700). "Do you think Ultrix could be ported to this?"
Me: Sure, the original development support was on a BSD VAX, we run BSD UNIX (akin to ULTRIX), the M/1000 is Big-Endian, but on reset can be set Little-Endian, we provide bootprom code & tools, I think software is fairly Endian-clean.
Bob: I want to do it!! and show (DEC CEO) Ken Olsen it doesn't take 3 years and hundreds of programmers. (we'd had a few beers).
Me: OK, give me your card, and let's see what we can do.
I then talked to MIPS Boston area sales guy: hey, the chances of DEC using outside RISCs are pretty low, but if we could get them it would a huge win.
It took a month or two to work through agreements with DEC lawyers.
Then we sent 2 M/1000s to Nashua, 1 as bringup target and the other as development support. Bob and 2 (I think) other guys then did a full Ultrix port in 2-3 weeks.
DEC West (including Workstation Sales run by Joe DiNucci), got wind of this ... and went nonlinear. They were getting killed by Sun SPARC workstations. THere was a big meeting back East, where Olsen introduced DiNucci, saying: sometimes at DEC we don't look at gems from outside and we need to. DiNucci presented, saying that if somebody was a VAX/MS customer, they were fine, but if someone was open to UNIX, MicroVAXen just weren't competitive with Sun SPARCstations, but the DEC folks in Palo Alto felt they could design very competitive products with MIPS chips. Shortly thereafter, Olsen issued orders, with a few days' notice, to senior DEC Fellows (i.e., very senior engineers), to fly to Silicon Valley and do due diligence on MIPS, and come back and report.
We had an all-day meeting. DEC had representatives of 4 groups: chip designers, systems designers, compilers, operating systems people. The chip crew (led by Bob Supnik) had many questions about clocking, reliability, processes, etc, which our chip designers had no problem answering. I think I recall them asking about 64-bit, we were already thinking about that for the R4000. The systems designers mostly asked about interfaces, i.e., were we flexible enough to fit systems them wanted to build. Of course, since MIPS chips had been designed for flexible use by other systems companies by engineers who'd often come from chip companies, that was no problem. Our chips weren't specific to the systems we built. The (few) compiler folks had few questions: MIPS compiler tech was well-known and highly-respected.
Normally, the challenge with a new architecture would be long time to port and stabilize operating system ... but Ultrix had already been running solidly for a few months, been heavily benchmarked, so that was easy.
We took them around our computer room, pointing out the servers we used for chip simulations. As we left, I heard one of the DEC engineers whisper to Supnik "We're DEC, and this little startup has more compute power for simulation than we do." Supnik (in voice to freeze air): yeah, that's why we've got a problem and will have to do something about it. (I knew then...)
The DEC Fellows returned and wrote an honest report, which had to be painful for the chip folks. Supnik and co said they could design as good or better RISC chips, but it would take years to do the software. Everybody else was fine and the systems folks were raring to go, to build compeitive UNIX products as soon as they could.
It didn't take too long to sign a deal in which they'd license our software,. They also got the rights to the design information, and to make their own variants (for example, for easier VAX transitions),
but they wanted at least one larger chip supplier than those we currently had as partners.
Motorola salespeople had hopes DEC would adopt its recently-announced MC88K RISC, not realizing how far along DEC was with MIPS. After some sometimes-surreal meetings, Motorola and MIPS signed a Memo of Understanding on a Friday, in which Motorola would drop its 88K in favor of MIPS R3000s ... which was OK with many customers they'd talked to, who said they'd be happy with R3000's supplied by Motorola, but they hadn't talked to Data General, which had invested heavily in building systems around MC88K. Over the weekend they did so, and by Monday changed their minds. Anyway, DEC helped us get NEC and Toshiba as semiconductor partners, which worked well.
DEC invested serious money in getting VAX/Ultrix third-party software moved to MIPS Ultrix, not very hard.
The MIPS-based business got to $2B/year (according to DiNucci), but it was growing too big to be run from Palo Alto, DEC wanted the management of that line of business to move back East to Maynard, but he didn't want to move. Instead, in a friendly departure, he moved to MIPS sales&marketing. We supported DEC sales sometimes. He once asked if I could give a talk for a big DEC sales meeting in Europe. I was a bit reluctant, but he said: it's in Nice (France) and they'll pay for it. I went. Nice was nice. DEC sales people knew me and of course they knew DiNucci. In early 1992, DEC chip folks announced Alpha chips at an ISSCC'92 conference, February 1992, well in advance of sellable systems. DEC sales people told us dispritedly: what? We've actually winning deals with and making good money with a solid MIPS-based product line. What does this mean? Now our customers are confused.
Alphas of course were very fast, well-designed (although with the surprising initial error of lacking byte/halfword load/stores, fixed after a few years). The UNIX was 64-bit only (unlike SGI, Sun, etc, who did 64/32 setups to run existing 32-bit binaries). I'd have done the same thing with a brand new architecture, but it did mean that DEC spent a lot of effort with third-party vendors getting their software 64-bit clean, which likely helped us all! See The Long Road to 64-bits. JohnMashey (talk) 23:45, 23 June 2021 (UTC)
Start a discussion about improving the DEC PRISM page
Talk pages are where people discuss how to make content on Wikipedia the best that it can be. You can use this page to start a discussion with others about how to improve the "DEC PRISM" page.