Notes on two presentations by Gordon Bell ca. 1998
|
From: Robert Myers on 22 Mar 2010 19:52 On Mar 22, 7:57 pm, eug...(a)cse.ucsc.edu (Eugene Miya) wrote: > In article <356cf4ca-9fac-4871-9cdb-e360d4139...(a)33g2000yqj.googlegroups.com>, > Robert Myers <rbmyers...(a)gmail.com> wrote: > > >I appreciate your willingness to engage me at all, but just once, I > >wish you'd lay a *fact* on me other than > > >(a) I (Robert) have a limited view of the world. (true, but I claim to > >do more of my own personal thinking than most, rather than borrowing > >my wisdom from others) > > Oh, sure, start by enumerating whom you think build (or used to be but > no longer build) supercomputers. Your answer, and most of the people in > the news group's answer would be highly media based. The past players > are quite a few: Cray Research, Cray Computer, much older, CDC, arguably > firms now would be like IBM (LLNL), some would argue the remains of SGI, > more arguably SUN, HP, firms who make what we call clusters. Oh yeah, a > few people might recognize Fujitsu, Hitachi (Nick's former machine bit > which he dangles before people), NEC. A few dozen older minisuper firms > existed, but those arguably were minisupers not supers. Let's just > stick to supers. > Yeah, but this is part of what's being argued. Once there were FPS boxes, and Convex and Alliant and whomever, the question became: why bother with high-end boxes? Hell, my attitude was: why bother with anyone. Greg Pfister wrote the book on Beowulf, I didn't, but I was already there. The fact is (Blue Waters perhaps being an exception--we'll see), we don't have high end boxes. We have press releases and power point presentations. Your "supercomputers" are to me more like FPS, Alliant, and Convex than what I grew up referring to as a supercomputer. These machines are nothing but lots of processors and a very unaggressive interconnect. IBM did some interesting engineering on Blue Gene with regard to power consumption, cooling, and computational density, and I have already praised those efforts. Other than that, what are we getting out of all these dollars and press releases? > So here's one example of an obscure fact: Bamford goes and writes a book > and writes about a box in AT&T in SF where all intercontinential traffic > comes in. Big deal box. So the name of this firm is Narus, and they > claim to make a supercomputer. So who has ever heard of this company? > Bamford clearly. Well they just happen to have offices where I live and > used to be in a center where Mozilla was done and other friends firms > like TellMe. We have the web now. The two things any naive person can > do is firmname.com or google(firmname). Does this firm make a supercomputer? > You tell me. > Anyone who follows me on posts should be able to anticipate my response. The "super" is in the connectivity these days. They were doing state-of-the-art switching? That's all there is that's left that matters. > One can go down a thread, realize that some people would say, well it's > not general purpose or stand alone, or any number of features. The > firms above lacked significant competition: remember Thinking Machines? > (They had better publicists.), Maspar, ever realize that Goodyear was > once in the supercomputer industry? And they still have computers > flying around? > People wanted glamor. They still do. The secret is out: there isn't much glamor to be had. You can do mediocre science on a desktop. Hell, for fraud, all you need is colleagues and a shared interest in being funded. Color plots really do help, though. > >(b) You (Eugene) know more important people than I (Robert) do > >(possibly true, depends on how you define important and what you mean > >by knowing someone). > > I don't know any more important people than you think you do. > If I do, not that much more important. > What can I say? If you're smart enough, you get to be around really interesting people. I'm sure there are lots of people like that hanging around. > >(c) You (Eugene) understand bureaucracies much better than I do and > >are much more adept at manipulating and negotiating them. > >(emphatically true). > > I hate to say it, I do cynically know a bit more about bureaucracies. > I would not use the words manipulating or negotiation (far too much > power sounding). Work navigates in the bureaucracies. > But it is about power, and you know it. > >I have a bad attitude toward the national security establishment. > >That's clear. I have also laid out my own technical concerns in some > >detail. Your preferred mode of response is to bat me away as if I > >were an annoying insect buzzing around your bureaucratically-blessed > >head. I may not have manners appropriate for the floor of the US > >Senate (and I don't know why anyone would be proud of such a skill,. > >anyway), but it's equally bad manners to respond to a technical beef > >with bureaucratic generalities. > > Maintain that attitude. It's a good attitude to have about them. > This is usenet Robert, all we can talk about here are generalities. > It's the behind the scenes email where things get done. > Politicians are compromisers, which is why people tend not to like them. > > There's bigger fish to fry. Go after MS Robert. The problem isn't at > the high end. It's commodity. > At one time, you took me to task for bad-mouthing Bill Gates. Which is it to be? Robert.
From: Eugene Miya on 26 Mar 2010 12:49 In article <ho8ter$k2o$1(a)smaug.linux.pwf.cam.ac.uk>, <nmm1(a)cam.ac.uk> wrote: >In article <4ba7f5c6$1(a)darkstar>, Eugene Miya <eugene(a)cse.ucsc.edu> wrote: >>Oh, sure, start by enumerating whom you think build (or used to be but >>no longer build) supercomputers. Your answer, and most of the people in >>the news group's answer would be highly media based. The past players >>are quite a few: Cray Research, Cray Computer, much older, CDC, arguably >>firms now would be like IBM (LLNL), some would argue the remains of SGI, >>more arguably SUN, HP, firms who make what we call clusters. Oh yeah, a >>few people might recognize Fujitsu, Hitachi (Nick's former machine bit >>which he dangles before people), NEC. A few dozen older minisuper firms >>existed, but those arguably were minisupers not supers. Let's just >>stick to supers. > >I sincerely hope that more than a few people would recognise NEC >among those! And there were others before and simultaneous with >CDC, but let's skip them. Among English speaking people (mostly USAian), I would tend to doubt it. At best they read. Few of us have hands on. 1se 2es CPU have been on US soil. Again, it's the same problem which sent Andrew Tanenbaum off usenet. I debated going and visit Fujitsu and the NEC Earth Simulator while on vacation in Tokyo, and in the end kept it a vacation. Skip mainframes as a supers as well. The issue of instruction set compatibility complicates this a bit. Acts of desparation (such as when one wishes for the presence of a superhero) might add to justification of IS incompatibilities as the ERA/CDC and Cray lines were. >>>(c) You (Eugene) understand bureaucracies much better than I do and >>>are much more adept at manipulating and negotiating them. >>>(emphatically true). >> >>I hate to say it, I do cynically know a bit more about bureaucracies. >>I would not use the words manipulating or negotiation (far too much >>power sounding). Work navigates in the bureaucracies. > >The word "navigates" implies more planning in the direction of the >decisions than fits the facts. Calling it Brownian motion is a >little unfair, though. It's more a sort of biassed random walk >with absorbing boundaries. You have more or less control than one thinks. In govts, there are political realities. Contracting Officers do the negotiation. But these are old hat historic (media) firms. I take it that you have never heard of Narus. -- Looking for an H-912 (container).
From: Eugene Miya on 26 Mar 2010 13:11 In article <048003e2-e495-4902-bf63-3958f011b9e9(a)v20g2000yqv.googlegroups.com>, Robert Myers <rbmyersusa(a)gmail.com> wrote: >On Mar 22, 7:57=A0pm, eug...(a)cse.ucsc.edu (Eugene Miya) wrote: >> In article <356cf4ca-9fac-4871-9cdb-e360d4139...(a)33g2000yqj.googlegroups.= >com>, >> Robert Myers =A0<rbmyers...(a)gmail.com> wrote: >> >I appreciate your willingness to engage me at all, but just once, I >> >wish you'd lay a *fact* on me other than >> >(a) I (Robert) have a limited view of the world. (true, but I claim to >> >do more of my own personal thinking than most, rather than borrowing >> >my wisdom from others) >> Oh, sure, start by enumerating whom you think build supercomputers. >> the news group's answer would be highly media based. >> >Yeah, but this is part of what's being argued. Once there were FPS >boxes, and Convex and Alliant and whomever, the question became: why >bother with high-end boxes? Excepting the C-3 and C-4, and maybe the SPP (I never used), these were minisupers. Not supers. I suggested leave them out. >Hell, my attitude was: why bother with anyone. >Greg Pfister wrote the book on Beowulf, I didn't, but I was already there. Greg, a good man, did more than that. He also did the RP3 which was an interesting research machine even if IBM couldn't sell it, copies of it, or pieces of it. But that's a longer story. Greg didn't just write the book. A lot of credit goes to Don Becker, Tom Sterling, and going back guys like Bart Locanti (who I have to bug about bisection bandwidth next Wed over dinner). >The fact is (Blue Waters perhaps being an exception--we'll see), we >don't have high end boxes. We have press releases and power point >presentations. Your "supercomputers" are to me more like FPS, >Alliant, and Convex than what I grew up referring to as a >supercomputer. These machines are nothing but lots of processors and >a very unaggressive interconnect. > >IBM did some interesting engineering on Blue Gene with regard to power >consumption, cooling, and computational density, and I have already >praised those efforts. Other than that, what are we getting out of >all these dollars and press releases? Your buzz. Meanwhile other firms are exploring/attempting different architectural ideas. I am mostly out of the direct line these days. >> So here's one example of an obscure fact: Bamford goes and writes a book >> and writes about a box in AT&T in SF where all intercontinential traffic >> comes in. =A0Big deal box. =A0So the name of this firm is Narus, and they >> claim to make a supercomputer. =A0So who has ever heard of this company? >> Bamford clearly. >> Does this firm make a supercom puter? >> >Anyone who follows me on posts should be able to anticipate my >response. The "super" is in the connectivity these days. They were >doing state-of-the-art switching? That's all there is that's left >that matters. Software. >> firms above lacked significant competition: remember Thinking Machines? >> (They had better publicists.), Maspar, ever realize that Goodyear was >> once in the supercomputer industry? =A0And they still have computers >> flying around? >> >People wanted glamor. They still do. The secret is out: there isn't >much glamor to be had. You can do mediocre science on a desktop. >Hell, for fraud, all you need is colleagues and a shared interest in >being funded. Color plots really do help, though. The Kindle is merely gray scale. >> >(b) You (Eugene) know more important people than I (Robert) do >> >(possibly true, depends on how you define important and what you mean >> >by knowing someone). >> >> I don't know any more important people than you think you do. >> If I do, not that much more important. >> >What can I say? If you're smart enough, you get to be around really >interesting people. I'm sure there are lots of people like that >hanging around. This assumes 2 people have mutual interests. Sure, interesting people at work. Interesting colleagues at other institutions (planning a trip right now). Mutual interests. Interest != important. Interest .NE. important. >> >(c) You (Eugene) understand bureaucracies much better than I do and >> >are much more adept at manipulating and negotiating them. >> >(emphatically true). >> I hate to say it, I do cynically know a bit more about bureaucracies. >> I would not use the words manipulating or negotiation (far too much >> power sounding). =A0Work navigates in the bureaucracies. >> >But it is about power, and you know it. I wouldn't hold with such a scalar concept. It's like a vector magnitude in math. Once you get to 2 and higher the concept of order goes by the way side. >> >I have a bad attitude toward the national security establishment. >> >That's clear. =A0I have also laid out my own technical concerns in some >> >detail. =A0Your preferred mode of response is to bat me away as if I >> >were an annoying insect buzzing around your bureaucratically-blessed >> >head. =A0I may not have manners appropriate for the floor of the US >> >Senate (and I don't know why anyone would be proud of such a skill,. >> >anyway), but it's equally bad manners to respond to a technical beef >> >with bureaucratic generalities. >> >> Maintain that attitude. =A0It's a good attitude to have about them. >> This is usenet Robert, all we can talk about here are generalities. >> It's the behind the scenes email where things get done. >> Politicians are compromisers, which is why people tend not to like them. >> There's bigger fish to fry. =A0Go after MS Robert. =A0The problem isn't at >> the high end. =A0It's commodity. >> >At one time, you took me to task for bad-mouthing Bill Gates. Which >is it to be? It's not either-or. It's knowing about change and what's present. 1) MS is now the Ballmer MS, not the Gates MS. 2) While Gates was the public face, people tend to forget that Paul Allen has his hands in a lot of things. Some times its good to be #2. 3) the DOD, the various uniformed services as well as the ununiformed get away with a lot, in particular the USAF is likely particularly wasteful. Why even some of those nonuniformed might be clients. Maybe an early 60s analogy would be who were knowledgeable about ERA in the computer industry or who in the general public heard of CDC before 1964? You only think of the buzz. Not the clients. I doubt that most people would even remember Cray's company. -- Looking for an H-912 (container). ------------ And now a word from our sponsor ------------------ Do your users want the best web-email gateway? Don't let your customers drift off to free webmail services install your own web gateway! -- See http://netwinsite.com/sponsor/sponsor_webmail.htm ----
From: Robert Myers on 26 Mar 2010 14:22 On Mar 26, 1:11 pm, eug...(a)cse.ucsc.edu (Eugene Miya) wrote: > In article <048003e2-e495-4902-bf63-3958f011b...(a)v20g2000yqv.googlegroups..com>, > Robert Myers <rbmyers...(a)gmail.com> wrote: > > >On Mar 22, 7:57=A0pm, eug...(a)cse.ucsc.edu (Eugene Miya) wrote: > >> In article <356cf4ca-9fac-4871-9cdb-e360d4139...(a)33g2000yqj.googlegroups.= > >com>, > >> Robert Myers =A0<rbmyers...(a)gmail.com> wrote: > >> >I appreciate your willingness to engage me at all, but just once, I > >> >wish you'd lay a *fact* on me other than > >> >(a) I (Robert) have a limited view of the world. (true, but I claim to > >> >do more of my own personal thinking than most, rather than borrowing > >> >my wisdom from others) > >> Oh, sure, start by enumerating whom you think build supercomputers. > >> the news group's answer would be highly media based. > > >Yeah, but this is part of what's being argued. Once there were FPS > >boxes, and Convex and Alliant and whomever, the question became: why > >bother with high-end boxes? > > Excepting the C-3 and C-4, and maybe the SPP (I never used), these were > minisupers. Not supers. I suggested leave them out. > Yes, but you have side-stepped my question. What is it, exactly, about supers that is so important? If you're designing throughput machines (aggregating lots of different, possibly non-local, activities) then measuring supers by mere size might be appropriate. If you're interested in doing things that otherwise wouldn't be possible (lots of jobs that could be done just as well on smaller machines don't count), what are you looking for in a super? My (aggressively oversimplifited) answer is that, if all the machine can do well are problems that are naturally localized or nearly embarrassingly parallel, it isn't a supercomputer, because you don't need anything special to do such problems. > >Hell, my attitude was: why bother with anyone. > >Greg Pfister wrote the book on Beowulf, I didn't, but I was already there. > > Greg, a good man, did more than that. He also did the RP3 which was an > interesting research machine even if IBM couldn't sell it, copies of it, or > pieces of it. But that's a longer story. Greg didn't just write the book. > > A lot of credit goes to Don Becker, Tom Sterling, and going back guys > like Bart Locanti (who I have to bug about bisection bandwidth next Wed > over dinner). > I've never had much (credited) influence over anything, and I certainly don't claim to have played any role in developing the Beowulf idea. My only claim is that, by the mid-eighties, the future was obvious to me and that I acted accordingly with respect to my own work. > >The fact is (Blue Waters perhaps being an exception--we'll see), we > >don't have high end boxes. We have press releases and power point > >presentations. Your "supercomputers" are to me more like FPS, > >Alliant, and Convex than what I grew up referring to as a > >supercomputer. These machines are nothing but lots of processors and > >a very unaggressive interconnect. > > >IBM did some interesting engineering on Blue Gene with regard to power > >consumption, cooling, and computational density, and I have already > >praised those efforts. Other than that, what are we getting out of > >all these dollars and press releases? > > Your buzz. > Is this the new inside-the-beltway putdown of choice? > Meanwhile other firms are exploring/attempting different architectural > ideas. I am mostly out of the direct line these days. > My only purpose is to emphasize qualities that seem important to me but not to the current "deciders." > >> So here's one example of an obscure fact: Bamford goes and writes a book > >> and writes about a box in AT&T in SF where all intercontinential traffic > >> comes in. =A0Big deal box. =A0So the name of this firm is Narus, and they > >> claim to make a supercomputer. =A0So who has ever heard of this company? > >> Bamford clearly. > >> Does this firm make a supercom puter? > > >Anyone who follows me on posts should be able to anticipate my > >response. The "super" is in the connectivity these days. They were > >doing state-of-the-art switching? That's all there is that's left > >that matters. > > Software. > I've gone on endlessly about two things: software and interconnect. They both, in the end, come down to the same thing. How easily can you control what's happening and when the control process has been defined, how transparent is it? My claim is that, all other considerations aside, current high-end hardware and software seem almost perversely designed to minimize transparency. With our "high-end" boxes it's partly because of the languages and user interfaces that are available and partly because you have to worry so much about where things actually are and how they got there. I posted here a guardian article about the realities of scientific software. If all that matters is position on a Top 500 list and an impressive facility to show visiting dignitaries, then it may not matter that most of what is actually being produced is riddled with errors. > >> firms above lacked significant competition: remember Thinking Machines? > >> (They had better publicists.), Maspar, ever realize that Goodyear was > >> once in the supercomputer industry? =A0And they still have computers > >> flying around? > > >People wanted glamor. They still do. The secret is out: there isn't > >much glamor to be had. You can do mediocre science on a desktop. > >Hell, for fraud, all you need is colleagues and a shared interest in > >being funded. Color plots really do help, though. > > The Kindle is merely gray scale. > But the iPad is color. Where is all the action going to go? > >> >(b) You (Eugene) know more important people than I (Robert) do > >> >(possibly true, depends on how you define important and what you mean > >> >by knowing someone). > > >> I don't know any more important people than you think you do. > >> If I do, not that much more important. > > >What can I say? If you're smart enough, you get to be around really > >interesting people. I'm sure there are lots of people like that > >hanging around. > > This assumes 2 people have mutual interests. Sure, interesting people > at work. Interesting colleagues at other institutions (planning a trip > right now). Mutual interests. Interest != important. Interest .NE.. important. > No. Important is people like Al Gore and the Club of Rome, both of whom made a career out of apocalypse without having much new in the way of insight. > >> >(c) You (Eugene) understand bureaucracies much better than I do and > >> >are much more adept at manipulating and negotiating them. > >> >(emphatically true). > >> I hate to say it, I do cynically know a bit more about bureaucracies. > >> I would not use the words manipulating or negotiation (far too much > >> power sounding). =A0Work navigates in the bureaucracies. > > >But it is about power, and you know it. > > I wouldn't hold with such a scalar concept. > It's like a vector magnitude in math. Once you get to 2 and higher the > concept of order goes by the way side. > You can order any vector space by choosing any vector whatsoever and doing a dot product with it. The vector you choose contains information about what's important to you. If you need non-linear measures, you can use sequences of dot products. I think I just described a neural net. In the end, you get a scalar which I strongly suspect is associated with maximizing action on a dopamine pathway in a decision-maker's brain. The question is: whose dopamine pathway matters in the end? If you don't like the word power as a descriptor of the quality that decisionmaker has, choose some other. You don't need me to explain any of this. You understand many more of the ways that decisions get skewed and steered than I do, although you did mention having a star in the flag (aka two senators) as particularly important. I'm not especially smart about these things, but neither am I naive. <snip> Robert.
From: Anne & Lynn Wheeler on 26 Mar 2010 14:42
Robert Myers <rbmyersusa(a)gmail.com> writes: > I've never had much (credited) influence over anything, and I > certainly don't claim to have played any role in developing the > Beowulf idea. My only claim is that, by the mid-eighties, the future > was obvious to me and that I acted accordingly with respect to my own > work. from an old thread in this newsgroup: http://www.garlic.com/~lynn/2000c.html#21 Cache coherence [was Re: TF-1] other pieces of above thread http://www.garlic.com/~lynn/2000c.html#9 Cache coherence [was Re: TF-1] http://www.garlic.com/~lynn/2000c.html#12 Cache coherence [was Re: TF-1] http://www.garlic.com/~lynn/2000c.html#22 Cache coherence [was Re: TF-1] http://www.garlic.com/~lynn/2001n.html#42 Cache coherence [was Re: IBM POWER4 ...] now I did have a proposal that predated RP3 ... that would intermix blue Iliad (first 32bit 801 chip design) and 370 boards ... big problem getting to something like 96 boards (aka processors) per rack ... was all heat issues. old post/reference http://www.garlic.com/~lynn/2004m.html#17 mainframe and microprocessor Now, one of the things RP3 is accused of is nice professional fullsize paste board mockups for display. However, at one point my wife did get tasked with auditing RP3 to see if funding should continue ... and it was thumbs down (well before we started ha/cmp effort & cluster scaleup with rios chips). possibly the "thumbs down" contributed to the comment in the cache coherence thread. past posts mentioning my wife getting asked to audit RP3 to see if funding should continue: http://www.garlic.com/~lynn/2000c.html#6 TF-1 http://www.garlic.com/~lynn/2006w.html#26 Why so little parallelism? http://www.garlic.com/~lynn/2006w.html#39 Why so little parallelism? related recent thread: http://www.garlic.com/~lynn/2010f.html#50 Handling multicore CPUs; what the competition is thinking http://www.garlic.com/~lynn/2010f.html#52 Handling multicore CPUs; what the competition is thinking http://www.garlic.com/~lynn/2010f.html#55 Handling multicore CPUs; what the competition is thinking http://www.garlic.com/~lynn/2010f.html#56 Handling multicore CPUs; what the competition is thinking http://www.garlic.com/~lynn/2010f.html#57 Handling multicore CPUs; what the competition is thinking http://www.garlic.com/~lynn/2010f.html#58 Handling multicore CPUs; what the competition is thinking http://www.garlic.com/~lynn/2010f.html#60 Handling multicore CPUs; what the competition is thinking http://www.garlic.com/~lynn/2010f.html#61 Handling multicore CPUs; what the competition is thinking http://www.garlic.com/~lynn/2010f.html#63 Handling multicore CPUs; what the competition is thinking http://www.garlic.com/~lynn/2010f.html#64 Handling multicore CPUs; what the competition is thinking http://www.garlic.com/~lynn/2010f.html#70 Handling multicore CPUs; what the competition is thinking http://www.garlic.com/~lynn/2010f.html#73 Handling multicore CPUs; what the competition is thinking -- 42yrs virtualization experience (since Jan68), online at home since Mar1970 |