Why so little parallelism?
|
From: Eugene Miya on 8 Nov 2006 13:51 In article <eit3lg$9ta$1(a)gemini.csx.cam.ac.uk>, Nick Maclaren <nmm1(a)cus.cam.ac.uk> wrote: >In article <45520aa1$1(a)darkstar>, eugene(a)cse.ucsc.edu (Eugene Miya) writes: >|> Nick has been cited by various people for driving knowledgeable people >|> out of various groups for not knowing what he is talking about. That >|> might be their perception, but I could also perceive fatigue on their part. > >I have also been cited as causing the demise of mainframes in UK >academia. I doubt that I am that influential. I doubt that you are that influential as well. The mainframes guys like Lynn and others sort of dug their own trenches. The UK has its own set of problems going back to the handling of Turing thread yet again in a.f.c., and Alvey, and Lighthill, and god knows what other mistakes you guys shoot yourselves in the foot with. Your baggage. One on hand they had a good argument for compatibility. So much of science is codified in Fortran and other older languages, but is not the requirement to rewrite every single code into what ever new language comes forward. The problem is that computers are new, people are in various rates of hurry. Business is a different issue. Moore's law caught up with hardware. The mainframes types mistook the old side of discrete component transistors as being indicative of what computers were (the implementation of architecture). I have been in many new buildings made for computers as mainframes only to find vast open spaces (air conditioned roller rinks) with PCs sitting on small tables. The reality is it's a field of neglect and missing details left to widely varying implementations which come back to bite us. People are not willing to invest in infrastructure and the maintenance which goes with it. Not in the long term. I wish I could locate or find my copy of Rollwagon's "kicking and screaming" letter. >|> To my >|> eyes he's a sort of taller version of the Dr. Who Jon Pertwee (he >|> doesn't like that comparison). > >I see what you mean, though I don't agree, and don't particularly object >to the comparison. Why should I? Hey, merely neurons firing. --
From: BDH on 8 Nov 2006 15:18 > >What, in Dallas? > > Dallas? No. oh, i was thinking of sigarch for some reason
From: lynn on 8 Nov 2006 15:43 Eugene Miya wrote: > The mainframes guys like Lynn and others sort of dug their own trenches. > The UK has its own set of problems going back to the handling of Turing > thread yet again in a.f.c., and Alvey, and Lighthill, and god knows what > other mistakes you guys shoot yourselves in the foot with. > Your baggage. i'm not sure what you are referring to ... possibly the invention of the compare&swap instruction and the associated semantics in the late 60s and early 70s ... which was shipped as part of mainframe 370 machines in the early 70s. misc. past posts mentioning smp and/or compare&swap instruction use for various kinds of parallel processing operations http://www.garlic.com/~lynn/subtopic.html#smp minor note ... except finishing off rfc 1044 support in mainframe product (18 years ago), i haven't done any mainframe work in 20 yrs. the following reference was a long ways from any mainframe world and this was started nearly 20 years ago and this particular reference is to nearly 15 yrs ago related to scaleup work for ha/cmp project http://www.garlic.com/~lynn/95.html#13 enormous amounts of high-speed data transport project http://www.garlic.com/~lynn/subnetwork.html#hsdt done in the early and mid-80s ... referenced here had little to do with mainframes http://www.garlic.com/~lynn/2006s.html#50 Ranking of non-IBM mainframe builders http://www.garlic.com/~lynn/2006t.html#6 Ranking of non-IBM mainframe builders a lot of it was heavily oriented towards 801/risc http://www.garlic.com/~lynn/subtopic.html#801 part of the early issue in the ha/cmp cluster scaleup http://www.garlic.com/~lynn/subtopic.html#hacmp was that rios had no provisions for cache coherency used in smp parallel environment ... as a result we had to make design trade-offs for message passing parallelism in cluster scaleup environments ... while not precluding being able to also support cache consistent parallelism. since none of this was remotely mainframe oriented ... i guess you must be referring to the mainframe trench created with the invention of the compare&swap instruction and its deployment on 370 machines in the early 70s ... which would then place all subsequent hardware designs that implemented instruction and cache coherency semantics similar to that of compare&swap semantics ... in the same trench???
From: Eugene Miya on 8 Nov 2006 15:02 >> >What, in Dallas? >> Dallas? No. In article <1163017095.281305.91570(a)h48g2000cwc.googlegroups.com>, BDH <bhauth(a)gmail.com> wrote: >oh, i was thinking of sigarch for some reason I have not attended a ISCA or an ASPLOS in a long time. But I do occasionally email Patterson or Hennessy for various reasons. Dave was on ACM business. --
From: Thomas Womack on 8 Nov 2006 16:28
In article <4rcgj4Fqma70U1(a)mid.individual.net>, Del Cecchi <cecchinospam(a)us.ibm.com> wrote: > I think Nick might be one of the folks who like to make snarky little > remarks about the folks who are trying to protect us from other folks > who would like to kill as many of us as possible. It is hard to avoid making snarky remarks about organisations, however laudable their goals, who have so little visible oversight and so little visible budgetary constraint ... it is fairly clear how to waste money by bucketloads given the circumstances of an intelligence agency, and more than plausible that intelligence agencies are not populated exclusively by angels, so the safe assumption is that money is invisibly wasted. Being able to claim on one's own authority that any embarrassing admission would be a Threat to National Security provides a level of immunity to scrutiny with which I am wary to trust humans; it may work better in circumstances like the US where there are a larger number of in some sense more competitive agencies, but you get the feeling that they've divided up the field of espionage in such a way that there's no actual competition anywhere. Enron comes somehow to mind. Money is invisibly wasted many places, but usually you end up with reasonably detailed questions in the House of Parliament (consider NHS IT projects, the cost over-run of which exceeds most estimates of the budget of Thames House, Vauxhall Cross and the Doughnut taken together); huge budgets spent in ways expensively proofed against investigative journalism are intrinsically snarkworthy. Tom |