thomsen@trwspf.TRW.COM (Mark Thomsen) (10/06/87)
Help me on a search if you can.
In our laboratory we have considerable interest in running things quickly, and
have developed and bought a number of fast non-general processors for
implementing certain functions or languages quickly. We are preparing to work
with Prolog for some embedded expert and deductive processing functions, and
I am not familiar with what can be purchased to run Prolog programs fast. If
left in our current state of ignorance, we will probably buy Modula-Prolog,
put it on MC68020's (where we have an excellent Modula-2 compiler), and do our
own speed enhancements.
What is out there that runs Prolog fast? I have heard a little about the ICOT
effort on PSI, but have not technical article or description. I have also seen
articles in the past on extracting concurrency, Concurrent Prolog, and the
like -- are there concurrent Prolog machines?
I would like any references or descriptions on Prolog machines that you guys
know of. I will submit a summary of received data in a few weeks. Thanks!
Mark R. Thomsenram%shukra@Sun.COM (Renu Raman, Sun Microsystems) (10/06/87)
In article <454@trwspf.TRW.COM>, thomsen@trwspf.TRW.COM (Mark Thomsen) writes: > > Help me on a search if you can. > > In our laboratory we have considerable interest in running things quickly, and > have developed and bought a number of fast non-general processors for > implementing certain functions or languages quickly. We are preparing to work > with Prolog for some embedded expert and deductive processing functions, and > I am not familiar with what can be purchased to run Prolog programs fast. If > left in our current state of ignorance, we will probably buy Modula-Prolog, > put it on MC68020's (where we have an excellent Modula-2 compiler), and do our > own speed enhancements. > > What is out there that runs Prolog fast? I have heard a little about the ICOT > effort on PSI, but have not technical article or description. I have also seen > articles in the past on extracting concurrency, Concurrent Prolog, and the > like -- are there concurrent Prolog machines? > > I would like any references or descriptions on Prolog machines that you guys > know of. I will submit a summary of received data in a few weeks. Thanks! > > > > Mark R. Thomsen I am sure somebody from UCB can comment on this better than I can, but here goes my pitch anyway. If you talk of prolog hardware - the fastest available today is from UCB called PLM. (Yeah! they are a few orders of magnitude faster than Japanese PSI - How about a university project out-performing a national project) Al. Despain has been instrumental in its development and so are Dobry, Srini, P. Van Roy & co. The First Hardware was available thru Xenologic [X-1] which hooks onto a Sun VME backplane was performing at 300K lips and the new CMOS version is > 500K (maybe 800K. I think the PSI barely reaches 3 figures). Apparently the new CMOS version (I don't know how true this is) is available as a MOSIS tape and you can probably license it and fabricate it. There is a pretty efficient compiler that goes along with it (Written by Van Roy). I guess the new PLM hardware hooks onto an NCR (as it was developed thru an NCR grant). You may want to look at ECRC's ICM (There is a paper being presented by J.C.Syre in one of the conferences in Japan - forget the name now). I don't have figures about its performance - you could probably contact ECRC at Munich. Hope this helps. Info given here are from sketchy info that I have gathered. If there is any wrong info here - please correct me [I am sure Despain & Co. read this newsgroup]. --------------------- Renu Raman ARPA:ram@sun.com Sun Microsystems UUCP:{ucbvax,seismo,hplabs}!sun!ram M/S 5-40, 2500 Garcia Avenue, Mt. View, CA 94043
suzuki@russell.STANFORD.EDU (Hiroyuki Suzuki) (10/08/87)
In article <454@trwspf.TRW.COM> thomsen@trwspf.UUCP (Mark Thomsen) writes: > >What is out there that runs Prolog fast? I have heard a little about the ICOT >effort on PSI, but have not technical article or description. I have also seen >articles in the past on extracting concurrency, Concurrent Prolog, and the >like -- are there concurrent Prolog machines? > I was a user of PSI and Quintus on VAX 8700 in Japan a month ago. We had made several bench marks and the conclusion is Quintus on VAX 8700 is faster than PSI if you write PROLOG PSI is specially designed for ESP(Extended Self-contained Prolog). So many re-writings are needed, though there is a PROLOG program for translating PROLOG program into ESP program. Furthermore ESP is still changing now. So, FORGET about PSI/ESP For detailed, technical information, contact ICOT(Institute for new generation COmputing Technology) Mita Kokusai Building 21F 1-4-28 Mita, Minato-ku, Tokyo 108, JAPAN
tedc@dartvax.UUCP (Ted Cooley) (10/09/87)
There may also be the Xenologic Processor. This machine is to be an add-on to the SUN workstations. Claims of up to 300Klips are claimed. Xenologic is in Berkeley, CA. ...tedc@whips.dartmouth.edu
jouvelot@mit-vax.LCS.MIT.EDU (Pierre Jouvelot) (10/09/87)
This followup is from a friend (Akihiko Konagaya) who doesn't read USENET News. Pierre -- In article <379@russell.STANFORD.EDU> suzuki@russell.UUCP (Hiroyuki Suzuki) writes: >I was a user of PSI and Quintus on VAX 8700 in Japan a month ago. >We had made several bench marks and the conclusion is > > Quintus on VAX 8700 is faster than PSI if you write PROLOG > >PSI is specially designed for ESP(Extended Self-contained Prolog). >So many re-writings are needed, though there is a PROLOG program for >translating PROLOG program into ESP program. Furthermore ESP is >still changing now. So, > FORGET about PSI/ESP It is true that PSI is rather slower. But it's not unfair to compare a toy hardware and a workstation with a full programming environment. Processor performance is considerablly diminished by various traps, which are required for the environment. As for performance, new PSI achieves 200-250 KLIPS. This is fairly good because the performance of most useful application programs are determined by the performance of built-in predicates but "append program". Futhermore, we know that 2 Mega LIPS is not so difficult, if we develop a custom Prolog chip with 20 MHz. The fastest Prolog machine currently available would be CHI-II developed at NEC. It achieved 500KLIPS (See Habata etal, "Co-operative High Performance Sequential Inference Machine" in ICCD '87 New York for the details). The full programming environment that includes multi-window interface, mutiple process environment as well as ordinary programming tools such as an optimizing compiler, an interpreter with incremental compiler, would be available at the end of 1987. CHI-II provides an extended prolog, named SUPLOG, which is powerful enough to implement whole system programs on CHI-II. It supports not only whole prolog language features but also supports multiple name space, interprocess communication facilities and various data types (characters, arrays, streams etc). The subset of SUPLOG will be available on a UNIX workstation soon. The prototype CHI, oridinally named HPM, achieved 280 KLIPS. Its architecture and an programming environment are reported in: Nakazaki etal, "Design of a high-speed prolog machine HPM", in Proc. of Computer Architecture, 1985. Konagaya etal, "A Co-Operative Programming Environment for a Back -End Type Sequential Inference Machine CHI", in Proc. of Parallel Algorithms and Architecture, 1987, Akademie-Verlag Berlin. Copies are available until Jan. 1988 in the following address: Akihiko Konagaya MIT Laboratory for Computer Science, 545 Technology Square, room 252, Cambridge, MA 02139 In case of after Jan. 1988, try to access: Akihiko Konagaya Computer System Laboratory, C&C Systems Laboratories, NEC Coporation, 1-1 Miyazaki, 4-chome, Miyamae, Kawasaki, Kanagawa, 213 Japan
carlton@ji.Berkeley.EDU (Mike Carlton) (11/03/87)
In article <30115@sun.uucp> ram%shukra@Sun.COM (Renu Raman, Sun Microsystems) writes: >In article <454@trwspf.TRW.COM>, thomsen@trwspf.TRW.COM (Mark Thomsen) writes: ... >> I would like any references or descriptions on Prolog machines that you guys >> know of. I will submit a summary of received data in a few weeks. Thanks! ... > If you talk of prolog hardware - the fastest available today is from > UCB called PLM. (Yeah! they are a few orders of magnitude faster than > Japanese PSI - How about a university project out-performing a > national project) Al. Despain has been instrumental in its development > and so are Dobry, Srini, P. Van Roy & co. The First Hardware was > available thru Xenologic [X-1] which hooks onto a Sun VME backplane was > performing at 300K lips and the new CMOS version is > 500K (maybe 800K. > I think the PSI barely reaches 3 figures). Apparently the new CMOS version > (I don't know how true this is) is available as a MOSIS tape and > you can probably license it and fabricate it. There is a pretty > efficient compiler that goes along with it (Written by Van Roy). > I guess the new PLM hardware hooks onto an NCR (as it was developed > thru an NCR grant). ... > Hope this helps. Info given here are from sketchy info that I > have gathered. If there is any wrong info here - please correct me > [I am sure Despain & Co. read this newsgroup]. > >--------------------- > Renu Raman ARPA:ram@sun.com > Sun Microsystems UUCP:{ucbvax,seismo,hplabs}!sun!ram > M/S 5-40, 2500 Garcia Avenue, > Mt. View, CA 94043 Thanks for the kind words. Let me give some background on the Aquarius project here at Berkeley. The group has built several Prolog machines to date, beginning with an NCR 9300 based system. Prolog programs were compiled to WAM (Warren Abstract Machine) and then to microcode. With a cycle time of 150 ns., this system achieved 53 KLIPS for determinate concat in 1985. The next machine, the PLM, was a TTL implementation of the WAM with some local improvements, see [Dobry87]. The PLM was simulated to run at 400 KLIPS for determinate concat, based on a 100 ns. cycle. Another project implemented the WAM instruction set in microcode on a VAX 8600. This machine, with an 80 ns. cycle, reached 108 KLIPS on determinate concat. The most recent machine (chip actually) is the VLSI PLM, being fabricated right now. This is a 200K equivalent transistor, 1.4 micron CMOS implement- ation of an extended PLM. With a cycle time of 100 ns. this chip has been simulated at 416 KLIPS for determinate concat and 98 KLIPS for Warren's Chat program. The Prolog compiler used for most of this research was written by Peter Van Roy [VanRoy85], and is available directly from him. He can be reached via email at vanroy@ji.berkeley.edu. Unfortunately, the VLSI PLM is not available as a MOSIS tape as Renu's message indicated. The chip was funded in part by NCR and they are fabricating it for us and for their own use. Our focus has shifted away from single processor systems; current research centers around parallel execution Prolog machines. For some results from recent work in this area, see [Fagin87]. Altogether, more than 50 papers and reports have been produced by the Aquarius project. If anyone would like a list of the publications, please send electronic mail to Kathleen Nimr (nimr@ji.berkeley.edu). Most of these publications are still available from her. As for commercial products, Xenologic licensed the PLM design from the university and has designed an improved version, called the X-1. This is a set of 2 VME boards which plug into a Sun workstation and function as a coprocessor. I don't have exact performance figures for their machine, but have been told that they are better than the PLM. If you would like more information you can contact Xenologic at 39899 Balentine Dr., Suite 145, Newark, CA 94560. Their phone number is (415) 656-0708. Hope this helps, if you would like more information or have any questions please send us mail, Mike (and the rest of the Aquarius group) carlton@ji.berkeley.edu or ...!ucbvax!ji!carlton References: Dobry, T., "A High Performance Architecture for Prolog", Ph.D Dissertation, UCB, May 1987 Fagin, B., Despain, A., "Performance Studies of a Parallel Prolog Architecture", 14th ICSA, June 1987 Van Roy, P., "A Prolog Compiler for the PLM", UCB/CSD Report No. 84/203, November 1984 Disclaimer: I have nothing to do with Xenologic (other than wanting my own X-1 for the Sun in my office!). However, some members of our research group do have connections with Xenologic. This should not be construed as an advertisement for Xenologic.