Re: AW: Mersenne: Prime Net Server
On 10 Sep 2001, at 7:34, [EMAIL PROTECTED] wrote: I beg your pardon: you didn´t really expect to get informed about a planned outage BEFORE it happens or a crash AFTER it happened, did you? Well, as a service operator myself (though without any responsibility to PrimeNet), I do try to warn users when a service is at unusual risk. As for crashes, or unplanned outages for other reasons (like the power failing), I would try to post a notice explaining what happened as soon as I know myself. In the case of a system which fails on a Friday evening I do not think it is unreasonable that there is no information until Monday. Presumably the system support staff are not paid on a 24x7 basis. BTW some people claim that the PrimeNet server is working, just very, very slowly. In that case it's more than likely that the server is sufferring a DoS attack :( Such an outage didn´t occur for the first time in my (nearly) three years supporting GIMPS and others will follow. Is that a statement, or a threat? BTW I do remember a few service breaks, though this one is the worst; in fact the service has been pretty reliable recently, at least until last Friday night. May be that´s one reason why GIMPS lost about 8.000 to 9.000 machines during the last six months. Other reasons may apply: (a) new projects which may be attractive to many people; (b) the increasing size of our work units; processor speeds really haven't been keeping up! (c) problems with utility (mains) power supply in some areas e.g. California discouraging people from running systems 24x7; (d) adverse publicity involving distributed computing projects e.g. the sysadmin who installed RC5 clients on his systems without consent of his employer. Regards Brian Beesley _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Mersenne: RE: L2 Cache
George Woltman wrote: b) More importantly, the CPU caches work less effectively as the FFT gets larger. A 64K FFT can fit within some CPU's L2 caches. Jumping to a 128K FFT means half the data is in the L2 cache and half is in main memory (when switching from pass 1 to pass 2 of the FFT). So, will the Pentium 4 Northwood core (with 512KB L2 cache as compared to the current Willamette's 256KB L2 cache) perform even better on Prime95? Will there be a setting to tell Prime95 that a Willamette or Northwood core is specifically being used (as opposed to just Pentium 4)? Stephan T. Lavavej _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: RE: L2 Cache
Hi, At 12:44 AM 9/10/2001 -0700, Stephan T. Lavavej wrote: So, will the Pentium 4 Northwood core (with 512KB L2 cache as compared to the current Willamette's 256KB L2 cache) perform even better on Prime95? It won't be worse! There should be some improvement, I wouldn't expect much. Of course, running at 2.2 GHz won't hurt. Will there be a setting to tell Prime95 that a Willamette or Northwood core is specifically being used (as opposed to just Pentium 4)? No. Regards, George _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: AW: Mersenne: Prime Net Server
-Original Message- From: [EMAIL PROTECTED] [EMAIL PROTECTED] Date: Monday, September 10, 2001 1:58 AM BTW some people claim that the PrimeNet server is working, just very, very slowly. In that case it's more than likely that the server is sufferring a DoS attack :( That was my first thought. Isn't mersenne.org physically located on Entropia's servers? I still have been unable to get to mersenne.org at all, but was able to get to Entropia's home page (although it took several minutes to partially download before I gave up waiting). Regards, Steve Harris _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Mersenne: Entropy of mersenne calculations.
What is the entropic efficiency of our calculations? Is it possible to say how much work must be performed in order to verify whether a number is prime? If it is, then how efficient are our methods in comparison? For instance, can it be shown that there is theoretically a better method, but one that no-one has discovered? -- === Gareth Randall === _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Prime Net Server
The thing is that even a 2-3 day outage is no big deal, because if we are all responsible GIMPSers, then we have our "days of work" configuration set to more than a couple days worth, right? So the worst that should have happened is that you have a result to check back in and have to wait for that while the next number is already crunching. :) - Original Message - From: [EMAIL PROTECTED] To: [EMAIL PROTECTED] Cc: [EMAIL PROTECTED] Sent: Sunday, September 09, 2001 10:34 PM Subject: AW: Mersenne: Prime Net Server I beg your pardon: you didn´t really expect to get informed about a planned outage BEFORE it happens or a crash AFTER it happened, did you? If you did expect this, then you should have joined another distributed-computing project like SETI. They do inform their participants about such things. But people who are cool enough to find million-digit-primes should be able to find out that the server is down the whole weekend. Well, you have to pay for it (in Germany), but who cares? Such an outage didn´t occur for the first time in my (nearly) three years supporting GIMPS and others will follow. May be that´s one reason why GIMPS lost about 8.000 to 9.000 machines during the last six months. Regards Achim Ursprüngliche Nachricht-Von: Matt Goodrich [mailto:[EMAIL PROTECTED]]Gesendet: Samstag, 8. September 2001 21:36An: Mersenne List (E-mail)Betreff: Mersenne: Prime Net Server Is there scheduled maint. going on with the server today, or is this a unscheduled outage? Matt
Re: Mersenne: Prime Net Server
Hi, it isn't that easy at all. The default parameter for networkretries is, afaik, 10 minutes. So most clients try to connect every 10 minutes until thy got a timeout. And the timeout seems to be very long :-/. I think we loose much power on these gimpsserver fails. Regards Jörg --Original Message Text--- From: Aaron Blosser Date: Mon, 10 Sep 2001 08:21:57 -0700 The thing is that even a 2-3 day outage is no big deal, because if we are all responsible GIMPSers, then we have our "days of work" configuration set to more than a couple days worth, right?So the worst that should have happened is that you have a result to check back in and have to wait for that while the next number is already crunching. :) - Original Message - From: [EMAIL PROTECTED] To: [EMAIL PROTECTED] Cc: [EMAIL PROTECTED] Sent: Sunday, September 09, 2001 10:34 PM Subject: AW: Mersenne: Prime Net Server I beg your pardon: you didn´t really expect to get informed about a planned outage BEFORE it happens or a crash AFTER it happened, did you? If you did expect this, then you should have joined another distributed-computing project like SETI. They do inform their participants about such things. But people who are cool enough to find million-digit-primes should be able to find out that the server is down the whole weekend. Well, you have to pay for it (in Germany), but who cares? Such an outage didn´t occur for the first time in my (nearly) three years supporting GIMPS and others will follow. May be that´s one reason why GIMPS lost about 8.000 to 9.000 machines during the last six months. Regards Achim Ursprüngliche Nachricht- Von: Matt Goodrich [mailto:[EMAIL PROTECTED]] Gesendet: Samstag, 8. September 2001 21:36 An: Mersenne List (E-mail) Betreff: Mersenne: Prime Net Server Is there scheduled maint. going on with the server today, or is this a unscheduled outage? Matt PGP signature
Re: Mersenne: Entropy of mersenne calculations.
On 10 Sep 2001, at 14:47, Gareth Randall wrote: What is the entropic efficiency of our calculations? Um. We're not even gaining information in the strict (information theory) sense, since all Mersenne primes existed long before anyone formulated the concept - yet we are certainly raising the entropy of the universe as a whole by our activities. In that sense, the entropic efficiency must be zero, though it's hard to see how any other computing activity could be any different. Is it possible to say how much work must be performed in order to verify whether a number is prime? If it is, then how efficient are our methods in comparison? For instance, can it be shown that there is theoretically a better method, but one that no-one has discovered? So far as Mersenne numbers are concerned (and some other types of numbers, e.g. Fermat numbers, numbers for which Proth's Theorem applies) there is AFAIK no theoretical work which would show that the tests we're using are anything other than optimal. Theoretically there should be ways to improve primality tests for general numbers so that they are about as efficient as the LL test is for a number of the same approximate size. At present the best algorithms for general numbers (e.g. ECPP) are a great deal less efficient than this; the largest number proved prime using ECPP is about the hundredth root of 2^6972593-1 in magnitude. Practical implementation of tests is a different matter. At present the best multiplication algorithms are O(n log n) - which would in itself be a huge surprise to anyone working in this field 50 years ago - yet it appears that there may be scope to improve on this. With suitable hardware (VVLW architectures) O(n) is possible. Also, the possible advent of quantum computing may have a substantial effect on the choice of algorithm. Regards Brian Beesley _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: RE: L2 Cache
On 10 Sep 2001, at 0:44, Stephan T. Lavavej wrote: So, will the Pentium 4 Northwood core (with 512KB L2 cache as compared to the current Willamette's 256KB L2 cache) perform even better on Prime95? Will there be a setting to tell Prime95 that a Willamette or Northwood core is specifically being used (as opposed to just Pentium 4)? Well - there was no need for any change to the code when the PIII changed from 512KB L2 cache running at half speed (Klamath) to 256KB L2 cache running at full speed (Coppermine), or when a similar change occurred in the Athlon line with the introduction of Thunderbird. Neither is there any need for the code to differentiate between Celeron 2 and PIII chips - the essential difference here is that the C2 has only 128KB of L2 cache - or between Thunderbird and Duron in the Athlon line. So I don't see any need for there to be anything to differentiate between different cache sizes on Pentium 4. A larger L2 cache will do no harm, though with data set sizes much bigger than the L2 cache, it probably won't help much either. (What's the difference in Prime95 benchmarks between a Celeron 800 and a PIII 800 (100 MHz FSB) in the same system?) Larger caches will benefit multiprocessor systems more than uniprocessor systems, _provided_ the work is allocated to processors using an efficient algorithm so that cache contents are not wasted by processing the data on another processor. If there is a change in the instruction set that can be exploited, or a significant change in relative timings for particular instructions between one variant of a processor and another, then maybe it might be worth forking the code (again), but IMHO it's better to keep forks to the minimum for two basic reasons: (a) reducing development maintainance load; (b) avoiding confusing users, as far as possible. BTW the new Tualatin PIIIs built using 0.13 micron technology have 512KB L2 cache. I'm not sure whether they can be used in existing FP-CGA mobos - you're likely going to need at least a BIOS upgrade; there may be other electrical considerations e.g. provision of the lower voltage required by the Tualatin core may not be possible without changes to the voltage regulator hardware. For the sake of anyone wishing to upgrade I hope the compatibility issues are resolved soon - Intel is already saying that the desktop PIII will be withdrawn from sale in early December! Regards Brian Beesley _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Prime Net Server
John R Pierce wrote: I dunno, I've let about 1/2 my machines drop out, they were mostly p133 and below, they've been replaced with fewer much faster machines. My P133 is still doing useful GIMPS work; just not very fast. Cheers... Russ _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Prime Net Server
On Mon, 10 Sep 2001 19:56:34 +0200, Jörg Thomsen [EMAIL PROTECTED] wrote: Hi, it isn't that easy at all. The default parameter for networkretries is, afaik, 10 minutes. So most clients try to connect every 10 minutes until thy got a timeout. And the timeout seems to be very long :-/. I think we loose much power on these gimpsserver fails. Regards Jörg However, AFAIK, the client continues working while it is trying to connect; it's multithreaded. As Aaron said, by default more than enough work is cached to get you through these sorts of outages. Nathan _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Prime Net Server
Aaron, you´re right. I had to check in a result and if you have to use the manual way to do this and if you have to pay per minute for every internet connection than it´s wiry that the server is down for days and nobody told you that this was planned or not planned and when the server will be available again. Regards Achim -Ursprüngliche Nachricht- Von: Aaron Blosser An: Mersenne@Base. Com Gesendet: Montag, 10. September 2001 17:21 Betreff: Re: Mersenne: Prime Net Server The thing is that even a 2-3 day outage is no big deal, because if we are all responsible GIMPSers, then we have our "days of work" configuration set to more than a couple days worth, right? So the worst that should have happened is that you have a result to check back in and have to wait for that while the next number is already crunching. :) - Original Message - From: [EMAIL PROTECTED] To: [EMAIL PROTECTED] Cc: [EMAIL PROTECTED] Sent: Sunday, September 09, 2001 10:34 PM Subject: AW: Mersenne: Prime Net Server I beg your pardon: you didn´t really expect to get informed about a planned outage BEFORE it happens or a crash AFTER it happened, did you? If you did expect this, then you should have joined another distributed-computing project like SETI. They do inform their participants about such things. But people who are cool enough to find million-digit-primes should be able to find out that the server is down the whole weekend. Well, you have to pay for it (in Germany), but who cares? Such an outage didn´t occur for the first time in my (nearly) three years supporting GIMPS and others will follow. May be that´s one reason why GIMPS lost about 8.000 to 9.000 machines during the last six months. Regards Achim Ursprüngliche Nachricht-Von: Matt Goodrich [mailto:[EMAIL PROTECTED]]Gesendet: Samstag, 8. September 2001 21:36An: Mersenne List (E-mail)Betreff: Mersenne: Prime Net Server Is there scheduled maint. going on with the server today, or is this a unscheduled outage? Matt
Mersenne: Multithreaded?
Date: Mon, 10 Sep 2001 19:56:51 -0400 From: Nathan Russell [EMAIL PROTECTED] ?Subject: Re: Mersenne: Prime Net Server On Mon, 10 Sep 2001 19:56:34 +0200, Jörg Thomsen [EMAIL PROTECTED] wrote: Hi, it isn't that easy at all. The default parameter for networkretries is, afaik, 10 minutes. So most clients try to connect every 10 minutes until thy got a timeout. And the timeout seems to be very long :-/. I think we loose much power on these gimpsserver fails. Regards Jörg However, AFAIK, the client continues working while it is trying to connect; it's multithreaded. That wasn't my experience. During the outage, I'd get the message that it was trying to contact the server. Then it would sit there for 6 minutes, doing nothing, then state that it would try again in an hour, then pick up where it left off. That 6 minutes per hour was a loss of 10% for 2 days. So don't know what Nathan means by multithreading. And, BTW, why does it take 6 minutes to give up? Wouldn't 1 minute be sufficient? _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Multithreaded?
On Mon, 10 Sep 2001 19:44:31 -0700, Bob Margulies [EMAIL PROTECTED] wrote: That wasn't my experience. During the outage, I'd get the message that it was trying to contact the server. Then it would sit there for 6 minutes, doing nothing, then state that it would try again in an hour, then pick up where it left off. That 6 minutes per hour was a loss of 10% for 2 days. I can't speak to some of your other questions, but I can offer a suggestion for that one. If you select 'do not communicate automatically' in the advanced-manual communication dialogue box, the client will wait for you to ask it to communicate. I do the same here at college, since the firewall automatically blocks off my access to off-campus sites 8 hours after I last log into it. I also must initiate all connections regardless, but this doesn't impact GIMPS (or, in my opinion, any sane distributed computing client). Nathan _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers