Author: Robert Hyatt
Date: 20:11:52 10/11/01
Go up one level in this thread
On October 11, 2001 at 17:54:32, Michael Nieves wrote: >You guys reminded of an "accident" I had in the early nineties working on a >Cray. > >Our firm had contracted some CPU time on a Cray on an DOD facility. I migrated >a version of scientific code to the cray to do some testing. I fired off the >code and went to grab some coffee. > >I returned 5 minutes later and observed that the program had hung. I killed the >process and went back to our slow alphas to do some debugging. Turned out to be >a simple bug that caused an infinite loop. > >I didn't think much of it until a month later when I received a $10,000 bill >from the DOD. > >Needless to say, I treaded carefully thereafter. I had something equally funny happen. We were trying to track down a bug in the Cray multitasking library. I was working closely with the O/S guys and Dennis Moen in particular up at Cray. They had 6 crays on their network, and they said "sure, you can used one all you want to help find this." (I should note that Cray Blitz was the only program that would crash the library, so it wasn't exactly 'critical' although they were definitely concerned since it was clearly a lib problem.) We (Harry and I) started running tests. And while I had one running on one machine, I decided to use others. We worked for several hours until the director of operations up there called me (I got to know everyone up there pretty well). He asked "What in the hell are you guys doing? You have the cpu perfmeters pegged on the wall on every machine up here. We thought something was broken." Before I could answer, the director of software development (a female I won't name to protect the guilty) called on the other phone line and asked "Do you mind if I borrow about half of the operating system group back? Every time I ask one of them to look at something they say "sure, but right now I am working on a problem with Hyatt..."" Needless to say, I became quite famous. Or "infamous" up there. They sent me a funny-money bill as a practical joke. We burned 6 Cray YMPs (8 cpus each) for a total of over 480 hours during that 48 hour testing binge. At 20K per hour, the bill was nearly one million. I think I still have the "bill" in my old Cray file. It was sent along with a note from John Rollwagen (president of CRI) that asked "Bob, I can't beat this damned chess challenger 7. Can you tell me how to run Cray Blitz so I can stomp the damned challenger with it?" Turns out he was a chessplayer as well. :) They also added a footnote "please ignore the bill, just in case you thought it was serious" or something to that effect. I never told them I had never given it a moment's "serious" thought. :) I will say I had a hell of a time working with the software guys up there. Only problem was none of 'em had reasonable names. Had I needed to spell any of their last names, I would have been dead. :) > >On October 11, 2001 at 11:23:38, Robert Hyatt wrote: > >>On October 11, 2001 at 00:43:57, Slater Wold wrote: >> >>>On October 11, 2001 at 00:27:17, Robert Hyatt wrote: >>> >>>>On October 10, 2001 at 22:55:27, Slater Wold wrote: >>>> >>>>>On October 10, 2001 at 22:36:01, Robert Hyatt wrote: >>>>> >>>>>>On October 10, 2001 at 16:55:16, Olaf Jenkner wrote: >>>>>> >>>>>>>>Some took hours on a Cray (Stiller). Figure months on a PC for those. >>>>>>>>No idea how long the really bad ones will take (kppkpp for example.) >>>>>>>>The promotion cases will certainly take a _long_ while... >>>>>>> >>>>>>>Didn't use Stiller the connection machine? >>>>>>>I remember that I read 1991 about this. >>>>>>> >>>>>>>OJe >>>>>> >>>>>> >>>>>>He used both. Burton Wendroff/Tony Warnock got him some time on a C90 >>>>>>sometime in the 1993 time frame... The advantage of the C90 is that it >>>>>>comes with 32 gigabytes of RAM, which was helpful for making the generator >>>>>>run very fast... Of course, he didn't do any compression, and he couldn't >>>>>>write the files out to anything for saving them... but he did compute some >>>>>>statistics about them before chucking them. >>>>> >>>>>Any idea where to get those statistics Bob? >>>> >>>> >>>>I believe Lewis published some of them quite some while back. But the >>>>thing he published was mainly "deepest mate" and the like. I didn't follow >>>>this very closely, and only remember Wendroff asking for "my contact" up at >>>>Cray in trying to set the time up... >>> >>>Ok. Thanks. >>> >>>> >>>> >>>> >>>> >>>>> >>>>>Who is compiling them now? >>>> >>>>Only Eugene so far as I know... >>> >>>On what? And how is he doing the 6 pieces? I thought you needed a 64-bit OS >>>and a LOT of memory and HD space. >> >> >>Last I heard he was using some "unused" alphas at microsoft. Since MS >>no longer supports the alpha with NT, the machines were free and he >>grabbed them. :) >> >> >> >>> >>>> >>>> >>>> >>>>> >>>>>How hard is it to get time on a Cray? >>>>> >>>> >>>>Have you noticed a "lack" of "cray blitz"? Draw your own conclusion. :) >>> >>>Well, it was running at one time! What's the difference between then and now!?! >> >> >>we could get time to play in one tournament a year. Maybe a few nights of >>testing between 1am and 5am if we were lucky. >> >> >> >> >>> >>>> >>>>The machines sell for $60,000,000.00 or so. So yes, it takes a lot of >>>>haggling to get time... >>> >>>Jeez.... >>> >>>> >>>>At one time CPU time was going for $20,000 per hour... >>> >>>Man oh man. I went looking up Cray's, (other than www.cray.com) and all I could >>>find was crap about Cray, Inc. suing the United States of America. (The CIA was >>>one of the organizations involved.) >>> >> >> >>Not sure what that is about. The US Government (the NSA and other "black" >>agencies as well) are their best customers. NSA has a "cluster" of crays >>if you can call that a "cluster". :) >> >> >> >> >>>So renting one for a day is pretty much outta the question for me.....as I don't >>>have $480,000 to blow. >>> >>>What a shame. >>> >>>> >>>> >>>> >>>> >>>> >>>> >>>>> >>>>> >>>>>Slate
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.