Why so little parallelism?
|
From: Eugene Miya on 1 Nov 2006 16:24 In article <1162404604.130489.33410(a)e3g2000cwe.googlegroups.com>, jacko <jackokring(a)gmail.com> wrote: >back to the OPs question. why so little parallel? Look not everything is parallel. As a start, you should read Amdahl's (law) paper. It's barely 3 pages long. Amdahl himself has given permission and it's the comp.parallel FAQ panel on the 20th at the very end. It goes out monthly. I think for theoretic purposes, Backus' Turing lecture on getting out of the von Neumann paradigm has potential (you have to read these words carefully, FP didn't go far, and FL also stagnated). >it's to do with the language used to program gets turned into an RTL >(register transfer language) which specifies all variables as >registers, and then this has to be mapped onto a machine which only has >so many regs. Mapping is not so easy. You get into the 80s data flow (static and dynamic) which didn't go far (a number of machines were built, a number of languages and compilers were tried).... >this is why we use paper for calculations, so that the variable space >can fit on it while the short term memory (register space) can only >hold so much. Paper is somewhat irrelevant. Thinking is what's needed. >making better RTL -> machine code translators which efficiently >schedule transfers of variables between cores, to provide modular >computation units. > >it is not a hardware problem per se, but a communications problem. This is true, you are on the right track, but you are where the field stood oh maybe 20 years ago. >maybe the RTL is not expressive enough. What would you have in an >OpenRTL? OpenAPL?? Have you used APL? That's where part of the field was 30 years ago. --
From: Richard on 1 Nov 2006 17:36 [Please do not mail me a copy of your followup] eugene(a)cse.ucsc.edu (Eugene Miya) spake the secret code <45491e94$1(a)darkstar> thusly: >As a start, you should read Amdahl's (law) paper. It's barely 3 pages >long. Amdahl himself has given permission and it's the comp.parallel >FAQ panel on the 20th at the very end. It goes out monthly. Also: <http://en.wikipedia.org/wiki/Amdahl's_law> -- "The Direct3D Graphics Pipeline" -- DirectX 9 draft available for download <http://www.xmission.com/~legalize/book/download/index.html>
From: BDH on 1 Nov 2006 17:47 > Look not everything is parallel. > > As a start, you should read Amdahl's (law) paper. The core of that is pretty much obvious. But the slow things can be made more parallel. > I think for theoretic purposes, Backus' Turing lecture on getting out of > the von Neumann paradigm has potential (you have to read these words > carefully, FP didn't go far, and FL also stagnated). I don't know what FP and OO and half a dozen other acronyms are supposed to be. They both succeed when they do by making certain kinds of abstraction easier in a half-assed way, and then say they're really about something totally ridiculous. Let's get rid of state on a machine that's mostly state! Let's put our code in objects because...I don't know, some guy thinks without justification that it will make X easier! > >it's to do with the language used to program gets turned into an RTL > >(register transfer language) which specifies all variables as > >registers, and then this has to be mapped onto a machine which only has > >so many regs. Thanks a lot, Mr Von Neumann. That is not a very good system. > >this is why we use paper for calculations, so that the variable space > >can fit on it while the short term memory (register space) can only > >hold so much. You're going to argue that we should base computer I/O on human I/O? > Have you used APL? > > That's where part of the field was 30 years ago. What happened to APL? There's K and J but it just hasn't had the influence on modern popular languages of Lisp or Smalltalk or C. Or Pascal or Fortran or Algol probably. I've been half-assedly working on sort of an APL/Lisp hybrid.
From: Richard on 1 Nov 2006 17:57 [Please do not mail me a copy of your followup] "BDH" <bhauth(a)gmail.com> spake the secret code <1162421248.333748.269580(a)m7g2000cwm.googlegroups.com> thusly: >[...] Let's put our code in objects because...I don't >know, some guy thinks without justification that it will make X easier! Shirley, you're not serious. -- "The Direct3D Graphics Pipeline" -- DirectX 9 draft available for download <http://www.xmission.com/~legalize/book/download/index.html>
From: John Dallman on 1 Nov 2006 18:44
In article <1162421248.333748.269580(a)m7g2000cwm.googlegroups.com>, bhauth(a)gmail.com (BDH) wrote: > Thanks a lot, Mr Von Neumann. That is not a very good system. No. But it is both easy to design hardware for - which was the initial reason for it - and easy to teach to people. It has proved to be pretty versatile, and there doesn't seem to be a better-for-all-purposes replacement around. Because of the sunk investment in Von Neumann, any radically different paradigm will have to be a lot better, for almost every purpose. The way that Arabic numerals and the system for using them, as compared with Latin numerals, provide an example of the kind of difference you want. --- John Dallman jgd(a)cix.co.uk "Any sufficiently advanced technology is indistinguishable from a well-rigged demo" |