Article 1512 of alt.sys.pdp10: Path: nntp1.ba.best.com!news2.best.com!news.maxwell.syr.edu!newsfeed.icl.net!news.algonet.se!algonet!uninett.no!news.ost.eltele.no!newsfeed1.enitel.no!masternews.telia.net!news-sto.telia.net!news.defero.net!junk.nocrew.org!not-for-mail From: lars brinkhoff Newsgroups: alt.sys.pdp10 Subject: Re: Pipelining Organization: NoCrew Laboratories Lines: 44 Sender: lars@junk.nocrew.org Message-ID: <85lmvusg4i.fsf_-_@junk.nocrew.org> References: <8s3mis$i0t$1@news3.bu.edu> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii User-Agent: Gnus/5.0803 (Gnus v5.8.3) Emacs/20.7 Date: Thu, 12 Oct 2000 06:55:34 GMT NNTP-Posting-Host: 212.73.17.42 X-Complaints-To: abuse@defero.net X-Trace: news.defero.net 971333734 212.73.17.42 (Thu, 12 Oct 2000 08:55:34 MET DST) NNTP-Posting-Date: Thu, 12 Oct 2000 08:55:34 MET DST Xref: nntp1.ba.best.com alt.sys.pdp10:1512 Perhaps this alt.folklore.computers message is of interest here? budd@csa.bu.edu (Phil Budne) writes: In article , Sergej Roytman wrote: >The other day in lecture, the prof said something that I thought I'd >run by this newsgroup. The discussion was about interesting >addressing modes and how easy (or not) it would be for a compiler to >take advantage of same. Naturally, the PDP-11 came up, and the prof >said that the demise of this particular qarchitecture was the >impracticallity of pipelining the thing. I can't help wondering if he was thinking of the PDP-10. It has indirect addressing bit (not exactly a mode) which can be nested. The last DEC designed PDP-10 CPU (codename Jupiter) was killed after (among other things) the designers noticed that indirect addressing was hard to pipeline, and that it's use was important when writing extended addressing (for access to more than 256K of 36-bit words). I was in the meeting where someone from the Jupiter team gave a talk to the FORTRAN group on how to generate fast code for Jupiter. We were making FORTRAN generate extended addressing code (which was used to create CAD tools for the Venus (VAX 8600) project which was going on at the same time, in the same building -- Venus and Jupiter were both behind schedule when Jupiter got the axe). FORTRAN generated instructions which used "extended format indirect words" (EFIWs) to specify addresses longer than the 18-bit offset in the right half on an instruction. Traditional indirect words looked like instructions (and were called Instruction Format Indirect Words -- IFIWs). But I don't remember if the killer point was raised in that meeting or not. In retrospect, I don't see why you couldn't detect and handle a single level of indirection without breaking the pipe. I also don't see why we couldn't have made FORTRAN use indexing alone for extended access. It might have required keeping base registers, and doing address arithmatic explicitly.... [...] Article 2886 of alt.sys.pdp10: Sender: prep@k9.prep.synonet.com Newsgroups: alt.sys.pdp10 Subject: Jupiter From: Paul Repacholi Date: 28 Dec 2000 14:11:19 +0800 Message-ID: <87snn9m5x4.fsf_-_@k9.prep.synonet.com> Lines: 14 User-Agent: Gnus/5.0807 (Gnus v5.8.7) Emacs/20.7 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii NNTP-Posting-Host: 146.d01.pe.iqnet.net.au X-Trace: 28 Dec 2000 22:45:25 +0800, 146.d01.pe.iqnet.net.au Path: nntp1.ba.best.com!news1.best.com!newsfeed.mathworks.com!news.tele.dk!193.190.198.17!newsfeeds.belnet.be!news.belnet.be!skynet.be!newsfeed.iinet.net.au!news.waia.asn.au!usenet.per.paradox.net.au!127.0.0.1!nobody Xref: nntp1.ba.best.com alt.sys.pdp10:2886 StrangeBrew writes: > I used the same approach for the I/O box in the Jupiter CPU - it was > 100K ECL bus in the CPU to the TTL I/0 bus for the unibus and > massbus and CI adapters. How about more info on Jupiter? There is so much UL around it that it would be nice to have some real info. -- Paul Repacholi 1 Crescent Rd., +61 (08) 9257-1001 Kalamunda. West Australia 6076 Raw, Cooked or Well-done, it's all half baked. Article 2891 of alt.sys.pdp10: Path: nntp1.ba.best.com!news2.best.com!news.maxwell.syr.edu!newsfeed.skycache.com!Cidera!cyclone2.ba-dsg.net!cyclone1.ba-dsg.net!typhoon2.ba-dsg.net.POSTED!not-for-mail Message-ID: <3A4BA7AD.62A782AC@bellatlantic.net> From: StrangeBrew X-Mailer: Mozilla 4.75 [en]C-CCK-MCD BA45DSL (WinNT; U) X-Accept-Language: en MIME-Version: 1.0 Newsgroups: alt.sys.pdp10 Subject: Re: Jupiter References: <87snn9m5x4.fsf_-_@k9.prep.synonet.com> Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Lines: 114 Date: Thu, 28 Dec 2000 20:50:50 GMT NNTP-Posting-Host: 138.88.77.74 X-Complaints-To: newsadmin@bellatlantic.net X-Trace: typhoon2.ba-dsg.net 978036650 138.88.77.74 (Thu, 28 Dec 2000 15:50:50 EST) NNTP-Posting-Date: Thu, 28 Dec 2000 15:50:50 EST Xref: nntp1.ba.best.com alt.sys.pdp10:2891 Paul Repacholi wrote: > > StrangeBrew writes: > > > I used the same approach for the I/O box in the Jupiter CPU - it was > > 100K ECL bus in the CPU to the TTL I/0 bus for the unibus and > > massbus and CI adapters. > > How about more info on Jupiter? There is so much UL around it > that it would be nice to have some real info. From memory, I can tell you what I know for sure, I left before they cancelled it. The design team was small, especially compared to the major vax efforts at the time (Venus and Comet). My count is: Program manager, tech director, another manager, memory guy, 2 microcoders/simulator guys, 2 cpu design guys, 3 I/O guys, and one tech. The manager was Ron Bingham. A "contract" was established with corporate to develop a machine that was 5X faster than the KL. The KL was considered a 1 MIP processor by our math back then. The goal was to provide a new cost curver per mips by supporting 5 times or more the number of users. The design team folks who had done 10 work before ncluded the microcode/simulator folks, the memory guru, and two of the I/O guys. One I/O designer had worked on the KS10. The two CPU designers had not done PDP10 CPU design work before. They had done IBM Mainframe compatibilbe work. The way we divided up the I/O was one guy did the massbus interface, one did the I/O box, and I did the console. The I/O box guy left so I got that as well as the console. Fairchild had come out iwth some 100K ECL parts that seemed to be able to be turned into a pipelined PDP10 cpu based on a set of microengines (horizontally coded micros). The concept of a pipelined machine, running 10 code was novel to dec. The engineering budget for the effort was relatively small, with a target manufacturing cost of the CPU at about 50K in 1980 US dollars. A simulator was constructed based on the SUDS designs. Microcode based on the design was developed. Tweaking the simulator and the microcode, a couple of design changes were worked out that added a microengine and identified a kicker that we called a mid life kicker, an accelerator. The cache design was four way, set associative, and theoretically projected a hit rate up to 97 percent. The memory box did a quad word fetch to increase the apparent speed of the system. An 8 word fetch was discussed but with memory prices at the time.. it was shelved but would have allowed a bit more performance, theoretically. Testing with the simulator showed a conservative 10X perf of the KL. We were thrilled and jumped on the other boxes to design. I based the console on the Fonz-11, 11/23 CPU, and worked with George Lord in small systems to create a minimal chip implementation of the glue logic around the F11. We came up with a mealy state generator based design that allowed george to use the same chips in a follow on PDT, and me to construct the console. COnsole was modeled on the lines of the KL console BUT with more diagnostic paths, like the IBm 43xx family, a small winchester drive, remote field service support, microcode load capability for the engines, and automated prefailure tracking alerts. The idea was to look for changes in the system operating parameters so that we could project a failure and get FS to have the part there on time. The I/O box that I designed was a fairly simple 100K ECL to TTL and TTL to ECL converter. The memory box bus that I interfaced with was straight forward, limited length, synchronous bus, with the proper ECL drivers, layout rules, and all that. The I/O device bus was a simple synchronous variation on the omnibus, wider, shorter, and lots faster. Gordon Bell, Vaxland, and the committee pulled one of our cpu designers to help boost the performance of the venus effort - it was not in the same range as our simulated results, with a much larger investment. When we had our 12 meeting in 2 months to explain the concepts of our design, the writing was on the wall. When Bob E was pulled, that was the posters going up that we would be delayed and then killed off as sacrifices for the vaxen. That is why I picked the name Hephastus for the KC10. Testing with the simulator showed we could tweak the balance of the machine between low compute/high I/O and HIGH compute/Low I/O. The simulator gave us numbers in the range of 50X the KL. Simulating 1000 users, we could just barely keep the system busy. BUT THAT WAS THE SIMULATOR. Fairchild still had to deliver the parts to build the engines. Now, folks who stayed at DEC have told me stories that the machine did not properly implement all of the addressing requirements of the 10 running Tops. That is my story. bob > > -- > Paul Repacholi 1 Crescent Rd., > +61 (08) 9257-1001 Kalamunda. > West Australia 6076 > Raw, Cooked or Well-done, it's all half baked. Article 2907 of alt.sys.pdp10: Path: nntp1.ba.best.com!news1.best.com!newsfeed.mathworks.com!nntp-out.monmouth.com!newspeer.monmouth.com!news.monmouth.com!not-for-mail From: pechter@i4got.pechter.dyndns.org (Bill Pechter) Newsgroups: alt.sys.pdp10 Subject: Re: Jupiter Date: 28 Dec 2000 17:47:48 -0500 Organization: Unknown Lines: 31 Message-ID: <92gfuk$j5r$1@i4got.pechter.dyndns.org> References: <87snn9m5x4.fsf_-_@k9.prep.synonet.com> <3A4BA7AD.62A782AC@bellatlantic.net> NNTP-Posting-Host: bg-tc-ppp231.monmouth.com Xref: nntp1.ba.best.com alt.sys.pdp10:2907 In article <3A4BA7AD.62A782AC@bellatlantic.net>, StrangeBrew wrote: > >From memory, I can tell you what I know for sure, I left before >they cancelled it. > Great. By the time I left Field Service there were so many stories about the cancellation due to it not running at the correct speed... Among the rumors was mental breakdown of the only engineer who understood the internal design of the E-Box. (Which was pretty difficult to believe, knowing DEC's methods of doing things...) By the time rumors got to Field Service they were pretty wild. Among them (the dual microcoded Jupiter/Venus box) because of the 36 bit datapaths in part of Venus. (Was there plans to use the I/O subsystem on both machines?) --Bill -- -- bpechter@monmouth.com | FreeBSD since 1.0.2, Linux since 0.99.10 | Unix Sys Admin since Sys V/BSD 4.2 | Windows System Administration: "Magical Misery Tour" Article 2912 of alt.sys.pdp10: Path: nntp1.ba.best.com!news1.best.com!newsfeed.mathworks.com!europa.netcrusader.net!199.45.45.8!cyclone1.ba-dsg.net!typhoon2.ba-dsg.net.POSTED!not-for-mail Message-ID: <3A4BD248.742CD7AA@bellatlantic.net> From: StrangeBrew X-Mailer: Mozilla 4.75 [en]C-CCK-MCD BA45DSL (WinNT; U) X-Accept-Language: en MIME-Version: 1.0 Newsgroups: alt.sys.pdp10 Subject: Re: Jupiter References: <87snn9m5x4.fsf_-_@k9.prep.synonet.com> <3A4BA7AD.62A782AC@bellatlantic.net> <92gfuk$j5r$1@i4got.pechter.dyndns.org> Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Lines: 53 Date: Thu, 28 Dec 2000 23:52:35 GMT NNTP-Posting-Host: 138.88.77.74 X-Complaints-To: newsadmin@bellatlantic.net X-Trace: typhoon2.ba-dsg.net 978047555 138.88.77.74 (Thu, 28 Dec 2000 18:52:35 EST) NNTP-Posting-Date: Thu, 28 Dec 2000 18:52:35 EST Xref: nntp1.ba.best.com alt.sys.pdp10:2912 Bill Pechter wrote: > > In article <3A4BA7AD.62A782AC@bellatlantic.net>, > StrangeBrew wrote: > > > >From memory, I can tell you what I know for sure, I left before > >they cancelled it. > > > > Great. By the time I left Field Service there were so many > stories about the cancellation due to it not running at the > correct speed... There were rumors that the microcode - which controlled the balance between io and compute - was hosed and could not hit the balance. Another more credible rumor was the addressing was not handled properly for all cases. My most cherished memory was when we discovered that a jrst0, jrst0 sequence caused the second jrst0 to be skipped, so the timing loops that used that trick were totally out of whack!! > > Among the rumors was mental breakdown of the only engineer who > understood the internal design of the E-Box. > > (Which was pretty difficult to believe, knowing DEC's methods of > doing things...) I heard that story too, I was in MD by then. Note that the KC10 team was exceptionally small. many times in the past the folks who understood a piece of gear to the max was very very small. For example, KMC/DMC 11, there were 3 of us - Remi, Paul and me. DP8e, there were two of us, Remi Lisee and me. > > By the time rumors got to Field Service they were pretty wild. > > Among them (the dual microcoded Jupiter/Venus box) because of the > 36 bit datapaths in part of Venus. I know we looked at how to do venus wit the ebox but the problem was the instruction set. PDP10 instruction set is fixed width, the vax is variable. that hoses the pipeline very badly. > > (Was there plans to use the I/O subsystem on both machines?) No, the vaxbus, the 13.3MB/s bus was already designed and set. The console was going to be looked at but...they wanted to use an 8080/8085 or 8086 as the console cpu. > > --Bill > > -- > -- > bpechter@monmouth.com | FreeBSD since 1.0.2, Linux since 0.99.10 > | Unix Sys Admin since Sys V/BSD 4.2 > | Windows System Administration: "Magical Misery Tour" Article 2915 of alt.sys.pdp10: Path: nntp1.ba.best.com!news1.best.com!newsfeed.mathworks.com!newsxfer.eecs.umich.edu!newsxfer3.itd.umich.edu!newsxfer.itd.umich.edu!news.itd.umich.edu!sarr Newsgroups: alt.sys.pdp10 Subject: Re: Jupiter References: <87snn9m5x4.fsf_-_@k9.prep.synonet.com> <3A4BA7AD.62A782AC@bellatlantic.net> <92gfuk$j5r$1@i4got.pechter.dyndns.org> Reply-To: sarr@umich.edu Organization: University of Michigan From: sarr@engin.umich.edu (Sarr J. Blumson) Lines: 16 Message-ID: Date: Fri, 29 Dec 2000 00:56:20 GMT NNTP-Posting-Host: 207.75.146.69 X-Trace: news.itd.umich.edu 978051380 207.75.146.69 (Thu, 28 Dec 2000 19:56:20 EST) NNTP-Posting-Date: Thu, 28 Dec 2000 19:56:20 EST Xref: nntp1.ba.best.com alt.sys.pdp10:2915 In article <92gfuk$j5r$1@i4got.pechter.dyndns.org>, Bill Pechter wrote: > >By the time rumors got to Field Service they were pretty wild. > >Among them (the dual microcoded Jupiter/Venus box) because of the >36 bit datapaths in part of Venus. Customers who were waiting for Jupiter were told this, too. -- -------- Sarr Blumson sarr@umich.edu voice: +1 734 764 0253 home: +1 734 665 9591 JSTOR, University of Michigan http://www-personal.umich.edu/~sarr/ 301 E Liberty, Ann Arbor, MI 48109-2262 Article 2976 of alt.sys.pdp10: Message-ID: <3A4E2383.D0F23D52@ev1.net> Date: Sat, 30 Dec 2000 10:03:46 -0800 From: Charles Richmond Reply-To: richmond@ev1.net Organization: Cannine Computer Center X-Mailer: Mozilla 4.7C-CCK-MCD {C-UDP; EBM-APPLE} (Macintosh; I; PPC) X-Accept-Language: en MIME-Version: 1.0 Newsgroups: alt.sys.pdp10 Subject: Re: Jupiter References: <87snn9m5x4.fsf_-_@k9.prep.synonet.com> Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit NNTP-Posting-Host: taydal-207-55-153-210.ev1.net X-Trace: newsa.ev1.net 978192453 taydal-207-55-153-210.ev1.net (30 Dec 2000 10:07:33 -0600) Lines: 24 Path: nntp1.ba.best.com!news1.best.com!feed.textport.net!newsfeed.mesh.ad.jp!sjc1.nntp.concentric.net!newsfeed.concentric.net!newsfeed.nwlink.com!nntp2.savvis.net!newsa.ev1.net Xref: nntp1.ba.best.com alt.sys.pdp10:2976 Paul Repacholi wrote: > > StrangeBrew writes: > > > I used the same approach for the I/O box in the Jupiter CPU - it was > > 100K ECL bus in the CPU to the TTL I/0 bus for the unibus and > > massbus and CI adapters. > > How about more info on Jupiter? There is so much UL around it > that it would be nice to have some real info. > IMHO asking about the Jupiter project is equivalent to rubbing salt in the wounds of the old Dec people...and waving a red flag. It is certain to get some of them to explode!!! You should have posted a warning first: "I am about to ask about the Jupiter project..." Then the old Dec folks can prepare their blood pressure medicine and refill their nitroglycerine prescriptions... Heck, I did *not* even work at Dec, and hearing about how the Jupiter was screwed could make me bust a gasket!!! -- +-------------------------------------------------------------+ | Charles and Francis Richmond | +-------------------------------------------------------------+ Article 3114 of alt.sys.pdp10: Sender: eric@ruckus.brouhaha.com From: Eric Smith Newsgroups: alt.sys.pdp10 Subject: Re: Bad mouthing the '10 guys? (historic) References: <939oh1$2hb$1@teabag.cbhnet> <3A588798.545D2DD6@bellatlantic.net> <87lmsnmfsk.fsf@k9.prep.synonet.com> <6ur92f6lp1.fsf@chonsp.franklin.ch> X-Disclaimer: Everything I write is false. Organization: Eric Conspiracy Secret Labs X-Eric-Conspiracy: There is no conspiracy. Date: 08 Jan 2001 16:28:13 -0800 Message-ID: Lines: 19 User-Agent: Gnus/5.0807 (Gnus v5.8.7) Emacs/20.7 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii NNTP-Posting-Host: ruckus.brouhaha.com X-Trace: 8 Jan 2001 16:32:46 -0800, ruckus.brouhaha.com Path: nntp1.ba.best.com!news1.best.com!newsfeed.mathworks.com!europa.netcrusader.net!64.152.100.70!cyclone-sjo1.usenetserver.com!news-out.usenetserver.com!HSNX.atgi.net!news.kjsl.com!news.spies.com!ruckus.brouhaha.com Xref: nntp1.ba.best.com alt.sys.pdp10:3114 Neil Franklin writes: > Neither of the above problems was an technical K.O. for an fast 10, so > long you are prepared throw transistors and design time at it. The most difficult problem with making a fast -10 is the aliasing of the general registers into memory (locations 00 through 17 octal). This makes it very difficult to do any reasonable data dependency analysis, scoreboarding, register renaming, etc., because you can't predict what registers will be live until the EAs of all preceding instructions are calculated. You'd pretty much have to do both register renaming and speculative execution to get this right. This is now relatively common practice in state-of-the-art microprocessors, but not something likely to happen in a home-grown FPGA implementation. Rumor has it that the Jupiter (KC10) hardware team planned to eliminate the register aliasing, until the software people heard about it and screamed bloody murder. Article 3209 of alt.sys.pdp10: Path: nntp1.ba.best.com!news1.best.com!newsfeed.mathworks.com!panix!news.panix.com!not-for-mail From: Rich Alderson Newsgroups: alt.sys.pdp10 Subject: Re: KC10 aka 2080 Date: 16 Jan 2001 14:45:16 -0500 Organization: Systems Administration, XKL LLC, Redmond WA 98052 Lines: 20 Sender: alderson+news@panix2.panix.com Message-ID: References: NNTP-Posting-Host: panix2.panix.com X-Trace: news.panix.com 979674316 24773 166.84.0.227 (16 Jan 2001 19:45:16 GMT) X-Complaints-To: abuse@panix.com NNTP-Posting-Date: 16 Jan 2001 19:45:16 GMT X-Newsreader: Gnus v5.7/Emacs 20.7 Xref: nntp1.ba.best.com alt.sys.pdp10:3209 Mark Garrett writes: > There seems to be a mountain of info on the real nature of the Jupiter aka > KC10 aka 2080 though out the tops20 sources introduced around 6.1 That was an earlier designation. I have in my possession a number of binders (in a nice chocolate color) with the words DECsystem-4050 Sales Training Seminar printed on them. (I got my first from Mark Lottor, at the 1988 DEC-20 Day luncheon; the others came from Joe Dempster.) So Jupiter was going to be a DEC-40. Tops-40, perhaps? After all, wasn't it the sales/marketing folks who renamed the PDP-10 to be the DECsystem-10? -- Rich Alderson alderson+news@panix.com "You get what anybody gets. You get a lifetime." --Death, of the Endless Article: 19901 of alt.sys.pdp10 Path: iad-read.news.verio.net!dfw-artgen!dfw-peer!news.verio.net!newsfeed!wn13feed!worldnet.att.net!199.45.49.37!cyclone1.gnilink.net!spamkiller.gnilink.net!nwrddc04.gnilink.net.POSTED!53ab2750!not-for-mail Message-ID: <3ECB4958.5020308@bellatlantic.net> From: bob smith User-Agent: Mozilla/5.0 (Windows; U; WinNT4.0; en-US; rv:1.0.2) Gecko/20030208 Netscape/7.02 X-Accept-Language: en-us, en MIME-Version: 1.0 Newsgroups: alt.sys.pdp10,comp.arch Subject: Re: A Dark Day... References: <3EC77B0D.2050008@bellatlantic.net> <3EC784B9.1050909@bellatlantic.net> <4e56972b.0305202132.3783bf79@posting.google.com> Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Lines: 44 Date: Wed, 21 May 2003 09:38:37 GMT NNTP-Posting-Host: 138.88.67.163 X-Complaints-To: abuse@verizon.net X-Trace: nwrddc04.gnilink.net 1053509917 138.88.67.163 (Wed, 21 May 2003 05:38:37 EDT) NNTP-Posting-Date: Wed, 21 May 2003 05:38:37 EDT Xref: dfw-artgen alt.sys.pdp10:19901 comp.arch:72548 David Kanter wrote: >>>I just stated why cancelling Jupiter was a good thing. >> >>Cancelling Jupiter was, by itself, an act of suicide by Digital. > > > What were Jupiter and Venus? I tried looking them up on Google, but > nothing came up that was meaningful... Jupiter (nee Hephastus) or KC10 (c for 100K ECL) was an instantiation of the PDP10 instruction set using 100K ecl components and a series of microcoded machines. It was approved to support 30 bit addressing (CE or if you prefer CGB would not allow 32 bit addressing). Venus was the next gen high end vax. I don't recall which number it turned into 8xxx or soemthing. I know Al helenius and Trygve Fossum wore working on it. At the time, the differences between a variable lenght instruction and a fixed length instruction allowed us to project a much faster machine on the 10 side than the vax side. //bob > > > >>You seem not to have grasped how much that infuriated many many powerful >>entities. There were people who banned Digital from their corporate >>procurement list as a result. Even among people who did not go that far, >>you did not see them go to VMS -- they went to UNIX. > > > This sounds like an interesting story...where can I find the full > version? > > >>Yes, Jupiter was in a great deal of trouble (as was Venus); and perhaps >>Venus was more important to save. Nevertheless, it was inexcusable that >>Digital failed to put in the resources to save Jupiter as well. >> >>Don't confuse short-term results with long-term results. >> > > > David Kanter Article: 19916 of alt.sys.pdp10 Path: iad-read.news.verio.net!dfw-artgen!dfw-peer!news.verio.net!newsfeed!newstransit.mitre.org!news.tufts.edu!logbridge.uoregon.edu!news.u.washington.edu!140.142.17.34.MISMATCH!news.u.washington.edu!Ikkoku-Kan.Panda.COM!mrc From: Mark Crispin Newsgroups: alt.sys.pdp10,comp.arch Subject: Re: A Dark Day... Date: Wed, 21 May 2003 07:17:18 -0700 Organization: University of Washington Lines: 23 Sender: mrc@Ikkoku-Kan.Panda.COM Message-ID: References: <3EC77B0D.2050008@bellatlantic.net> <3EC784B9.1050909@bellatlantic.net> <4e56972b.0305202132.3783bf79@posting.google.com> <3ECB4958.5020308@bellatlantic.net> NNTP-Posting-Host: nntp2.u.washington.edu Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Trace: nntp1.u.washington.edu 1053526647 29648 (None) 140.142.17.39 X-Complaints-To: help@cac.washington.edu In-Reply-To: <3ECB4958.5020308@bellatlantic.net> Xref: dfw-artgen alt.sys.pdp10:19916 comp.arch:72557 On Wed, 21 May 2003, bob smith wrote: > Jupiter (nee Hephastus) or KC10 (c for 100K ECL) was an instantiation of > the PDP10 instruction set using 100K ecl components and a series of > microcoded machines. It was approved to support 30 bit addressing (CE > or if you prefer CGB would not allow 32 bit addressing). 30-bit addressing is the maximum in the PDP-10 instruction set. You can't have 32-bit addressing without changing the format of pointers, since you need to fit in the global/local bit, the indirect bit, and the index bits. Note that 30-bit addressing in the PDP-10 is equivalent to "32-bits plus 1" in a byte-addressed machine. This is because the PDP-10 addresses 36-bit words and not 8-bit bytes. If, as I propose, you use nonets to represent Unicode in UTF-9, you have exactly as many nonets of address in the PDP-10 address space as you do octets on a 32-bit machine. However, UTF-9 is a much more space-efficient means of storage than UTF-8. -- Mark -- http://staff.washington.edu/mrc Science does not emerge from voting, party politics, or public debate. Article: 19945 of alt.sys.pdp10 Path: iad-read.news.verio.net!dfw-artgen!dfw-peer!news.verio.net!newsfeed!wn13feed!worldnet.att.net!199.45.49.37!cyclone1.gnilink.net!spamkiller.gnilink.net!nwrddc02.gnilink.net.POSTED!53ab2750!not-for-mail Message-ID: <3ECC0082.2000905@bellatlantic.net> From: bob smith User-Agent: Mozilla/5.0 (Windows; U; WinNT4.0; en-US; rv:1.0.2) Gecko/20030208 Netscape/7.02 X-Accept-Language: en-us, en MIME-Version: 1.0 Newsgroups: alt.sys.pdp10,comp.arch Subject: Re: A Dark Day... References: <3EC77B0D.2050008@bellatlantic.net> <3EC784B9.1050909@bellatlantic.net> <4e56972b.0305202132.3783bf79@posting.google.com> <3ECB4958.5020308@bellatlantic.net> <3ECBFE12.1030909@iee.org> Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Lines: 42 Date: Wed, 21 May 2003 22:40:03 GMT NNTP-Posting-Host: 138.88.67.163 X-Complaints-To: abuse@verizon.net X-Trace: nwrddc02.gnilink.net 1053556803 138.88.67.163 (Wed, 21 May 2003 18:40:03 EDT) NNTP-Posting-Date: Wed, 21 May 2003 18:40:03 EDT Xref: dfw-artgen alt.sys.pdp10:19945 comp.arch:72581 Antonio Carlini wrote: > bob smith wrote: > >> Jupiter (nee Hephastus) or KC10 (c for 100K ECL) was an instantiation >> of the PDP10 instruction set using 100K ecl components and a series of >> microcoded machines. It was approved to support 30 bit addressing (CE >> or if you prefer CGB would not allow 32 bit addressing). > > > I gues that CGB is Gordon Bell, but who was CE? CGB, is indeed Chester Gordon Bell. CE is Central Engineering, and the board under bell. > >> Venus was the next gen high end vax. I don't recall which number it >> turned into 8xxx or soemthing. I know Al helenius and Trygve Fossum >> wore working on it. > > > Venus was the VAX 8600. I believe they had to downgrade the > speed to get it out at all (and it was pretty late > by then). When they eventually managed to up the speed, > the result came out as the VAX 8650 (aka Morningstar). > There was a rumour of a VAX 8670, but I have no idea > if it was ever more than a rumour. > > Antonio Thanks, I was not sure of the numbering, I just recall it was late, had a HUGE staff compared to 2080, and had a target manufacturing price for the CPU of 50K. bob > > > -- > > --------------- > Antonio Carlini arcarlini@iee.org > Article: 19954 of alt.sys.pdp10 Path: iad-read.news.verio.net!dfw-artgen!dfw-peer!news.verio.net!newsfeed!arclight.uoregon.edu!logbridge.uoregon.edu!news.u.washington.edu!140.142.17.34.MISMATCH!news.u.washington.edu!Tomobiki-Cho.CAC.Washington.EDU!MRC From: Mark Crispin Newsgroups: alt.sys.pdp10,comp.arch Subject: Re: A Dark Day... Date: Wed, 21 May 2003 17:23:07 -0700 Organization: Networks & Distributed Computing Lines: 33 Sender: mrc@ndcms.cac.washington.edu Message-ID: References: <3EC77B0D.2050008@bellatlantic.net> <3EC784B9.1050909@bellatlantic.net> <4e56972b.0305202132.3783bf79@posting.google.com> <3ECB4958.5020308@bellatlantic.net> <3ECBFE12.1030909@iee.org> <3ECC0082.2000905@bellatlantic.net> NNTP-Posting-Host: nntp6.u.washington.edu Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Trace: nntp1.u.washington.edu 1053563114 29038 (None) 140.142.17.40 X-Complaints-To: help@cac.washington.edu In-Reply-To: <3ECC0082.2000905@bellatlantic.net> Xref: dfw-artgen alt.sys.pdp10:19954 comp.arch:72584 On Wed, 21 May 2003, bob smith wrote: > >> Jupiter (nee Hephastus) or KC10 (c for 100K ECL) was an instantiation > >> of the PDP10 instruction set using 100K ecl components and a series of > >> microcoded machines. It was approved to support 30 bit addressing (CE > >> or if you prefer CGB would not allow 32 bit addressing). > > I gues that CGB is Gordon Bell, but who was CE? > CGB, is indeed Chester Gordon Bell. CE is Central Engineering, and the > board under bell. Neither of them had anything to do with 30-bit vs. 32-bit addressing. The PDP-10 pointer word format limits it to 30 bits (but that's still more memory than a 32-bit byte-oriented machine). On the other hand, I would believe a limit of 27-bit addressing. It is relatively straightforward to extend from 23-bit addressing (the limit in the KL) to 27 bits. Instead of a 5-bit section table, you have a full page (9-bit) section table, and page numbers become 18 bits instead of 14 bits. There's a fair amount of surgery needed to the virtual memory stuff, but it's relatively straightforward. Getting those last three bits, though, are a nightmare. To start with, only 18 bits are available for the page number in a storage address; bits 12-17 hold the storage medium. So that breaks immediate page pointers and indirect page pointers. It also breaks the page fail word which only has 27 bits. I understand that XKL did this; they would have had to redefine how the pager works and then change lots of stuff in the TOPS-20 kernel. -- Mark -- http://staff.washington.edu/mrc Science does not emerge from voting, party politics, or public debate. Article: 19956 of alt.sys.pdp10 Path: iad-read.news.verio.net!dfw-artgen!dfw-peer!news.verio.net!newsfeed!wn13feed!worldnet.att.net!199.45.49.37!cyclone1.gnilink.net!spamkiller.gnilink.net!nwrddc02.gnilink.net.POSTED!53ab2750!not-for-mail Message-ID: <3ECC2336.6090701@bellatlantic.net> From: bob smith User-Agent: Mozilla/5.0 (Windows; U; WinNT4.0; en-US; rv:1.0.2) Gecko/20030208 Netscape/7.02 X-Accept-Language: en-us, en MIME-Version: 1.0 Newsgroups: alt.sys.pdp10,comp.arch Subject: Re: A Dark Day... References: <3EC77B0D.2050008@bellatlantic.net> <3EC784B9.1050909@bellatlantic.net> <4e56972b.0305202132.3783bf79@posting.google.com> <3ECB4958.5020308@bellatlantic.net> <3ECBFE12.1030909@iee.org> <3ECC0082.2000905@bellatlantic.net> Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Lines: 57 Date: Thu, 22 May 2003 01:08:09 GMT NNTP-Posting-Host: 138.88.67.163 X-Complaints-To: abuse@verizon.net X-Trace: nwrddc02.gnilink.net 1053565689 138.88.67.163 (Wed, 21 May 2003 21:08:09 EDT) NNTP-Posting-Date: Wed, 21 May 2003 21:08:09 EDT Xref: dfw-artgen alt.sys.pdp10:19956 comp.arch:72586 Mark Crispin wrote: > On Wed, 21 May 2003, bob smith wrote: > >>>>Jupiter (nee Hephastus) or KC10 (c for 100K ECL) was an instantiation >>>>of the PDP10 instruction set using 100K ecl components and a series of >>>>microcoded machines. It was approved to support 30 bit addressing (CE >>>>or if you prefer CGB would not allow 32 bit addressing). >>> >>>I gues that CGB is Gordon Bell, but who was CE? >> >>CGB, is indeed Chester Gordon Bell. CE is Central Engineering, and the >>board under bell. > > > Neither of them had anything to do with 30-bit vs. 32-bit addressing. The > PDP-10 pointer word format limits it to 30 bits (but that's still more > memory than a 32-bit byte-oriented machine). > > On the other hand, I would believe a limit of 27-bit addressing. It is > relatively straightforward to extend from 23-bit addressing (the limit in > the KL) to 27 bits. Instead of a 5-bit section table, you have a full > page (9-bit) section table, and page numbers become 18 bits instead of 14 > bits. There's a fair amount of surgery needed to the virtual memory > stuff, but it's relatively straightforward. > > Getting those last three bits, though, are a nightmare. To start with, > only 18 bits are available for the page number in a storage address; bits > 12-17 hold the storage medium. So that breaks immediate page pointers and > indirect page pointers. It also breaks the page fail word which only has > 27 bits. > > I understand that XKL did this; they would have had to redefine how the > pager works and then change lots of stuff in the TOPS-20 kernel. Bingo, but when that same sort of arguement was presented to the committee of consulting engineers, Bell's staff, the answer was no. Yes, would have broken Tops 20. Yes, would have been a departure from PDP10 architecture in a subtle but majore way. Yes, it would have been a discriminator. Yes, we did use that ploy to get 30 bits approved. We did spend a lot of time looking at how we could support the extended addressing in fortran. I hope some of this discussion spurs some ofthe lurkers to pipe in here. It was a less than pleasent experience having to stop work and prepare a briefing or presenation for a new set of CE crew while trying to keep a deadline. It was not pleasent rejustifying over and over, the decisions made. It was fun when Ted Hess wore a suit and I wore jeans and a tee shirt. That kind of threw them for a minute. > -- Mark -- > > http://staff.washington.edu/mrc > Science does not emerge from voting, party politics, or public debate. Article: 19955 of alt.sys.pdp10 Path: iad-read.news.verio.net!dfw-artgen!dfw-peer!sjc-peer.news.verio.net!news.verio.net!newsfeed.cwix.com!newsfeed!panix!not-for-mail From: Rich Alderson Newsgroups: alt.sys.pdp10,comp.arch Subject: Re: A Dark Day... Date: 21 May 2003 20:51:21 -0400 Organization: Systems Administration, XKL LLC, Redmond WA 98052 Lines: 25 Sender: alderson+news@panix5.panix.com Message-ID: References: <3EC77B0D.2050008@bellatlantic.net> <3EC784B9.1050909@bellatlantic.net> <4e56972b.0305202132.3783bf79@posting.google.com> <3ECB4958.5020308@bellatlantic.net> <3ECBFE12.1030909@iee.org> <3ECC0082.2000905@bellatlantic.net> NNTP-Posting-Host: panix5.panix.com X-Trace: reader1.panix.com 1053564684 22225 166.84.1.5 (22 May 2003 00:51:24 GMT) X-Complaints-To: abuse@panix.com NNTP-Posting-Date: Thu, 22 May 2003 00:51:24 +0000 (UTC) X-Newsreader: Gnus v5.7/Emacs 20.7 Xref: dfw-artgen alt.sys.pdp10:19955 comp.arch:72585 bob smith writes: > Antonio Carlini wrote: >> Venus was the VAX 8600. I believe they had to downgrade the >> speed to get it out at all (and it was pretty late >> by then). When they eventually managed to up the speed, >> the result came out as the VAX 8650 (aka Morningstar). >> There was a rumour of a VAX 8670, but I have no idea >> if it was ever more than a rumour. > Thanks, I was not sure of the numbering, I just recall it was late, had > a HUGE staff compared to 2080, and had a target manufacturing price for > the CPU of 50K. Shortly before the cancellation, the 2080 had been renumbered as the 4050; I have a marketing materials binder with the DECSYSTEM-4050 name right there on the front for all to see. Grump. I never thought to scan it till just now; have to wait till I can get it out of the storage locker. -- Rich Alderson news@alderson.users.panix.com "You get what anybody gets. You get a lifetime." --Death, of the Endless