Enlisting the Help of Universities for GIMPS
 

This fall I will be attending a university having graduated from a community college in May. I have already met the President and he has quite a reputation for being forward thinking and is well liked by both faculty and students. I would like to present a case to him for enlisting some of the CPU resources on campus for aiding GIMPS. Being a university, these resources are plentiful as there are many "idle" computers sitting around on campus. I would not even suggest this except for the fact that I think he would be open to it. However, I would like to present a structured case including the current status of the project and some background information. The background information is rather easy to obtain... as is the majority of the status information. However, I would like to include the overall status of the project and not just the main range of testing. The PrimeNet page only gives information for exponents up to 100M where as I want to make a short and concise table that also includes 100M digit prefactor project and Operation Billion Digits. This information is not as easy to obtain so I have a few questions to ask. Any help in answering them will be greatly appreciated as well as any input or advice on my idea.

1) If Trial Factoring or P-1 finds a factor, is that the end of it and no double check is assigned where as with an LL a double check is always done?

2) Where can I find a page that displays information on exponents 100M to the cutoff for one billion digits around 332M? I could not see a homepage for this project.

3) I know that Operation Billion Digits has a homepage...however I believe that some indicated more exponents had been trial factored than were indicated on the table on said homepage. Where can I get that information?

Thanks! If I think of anything else I will post it.

[quote=Primeinator;176964]1) If Trial Factoring or P-1 finds a factor, is that the end of it and no double check is assigned where as with an LL a double check is always done?[/quote]
Correct. Double checks are just a second LL run to verify that the first LL was correct. They aren't necessary if the number can be proven to be composite by the presence of a factor. A factor can be verified very quickly (the server probably verifies all incoming factors), while an LL can only be verified by being run again and checking that the final residue is the same (Prime95 technically only checks the 64 least significant bits of the final residue).

Good luck with your suggestion. Sorry, but I don't know the answers to your other questions.

[QUOTE]Correct. Double checks are just a second LL run to verify that the first LL was correct. They aren't necessary if the number can be proven to be composite by the presence of a factor. A factor can be verified very quickly (the server probably verifies all incoming factors), while an LL can only be verified by being run again and checking that the final residue is the same (Prime95 technically only checks the 64 least significant bits of the final residue).

Good luck with your suggestion. Sorry, but I don't know the answers to your other questions. [/QUOTE]

Thanks! That answers one question and is definitely a start!

[quote=Primeinator;176964]<snip>[/quote]

You've got your exponents and digits mixed up. The PrimeNet page shows exponents up to 1G (IIRC), which is about 300 million digits. The billion digit line is at 3.32G.

I just realized this. How embarrassing. Okay... so I am in the process of getting information for exponents 0 to 1 billion... but I imagine that past 100M or so this PrimeNet table may not have all the information for people that are doing individual work for the 100M digit prefactor project?? (100 M digits starting at 332M roughly)

Be sure to let them know that it costs money to run this project. A rough estimate is 50 watts per machine.

[quote=Prime95;176980]Be sure to let them know that it costs money to run this project. A rough estimate is 50 watts per machine.[/quote]

If they haven't enabled the power saving options the more power consumption might be less.

I'll check. I thought of another question...

Is there a quick way I can find out how many primes are between 332,192,831 and 333,000,000? Thanks!

[QUOTE=Primeinator;176994]Is there a quick way I can find out how many primes are between 332,192,831 and 333,000,000? Thanks![/QUOTE][URL="http://v5www.mersenne.org/primenet/"]PrimeNet reports[/URL] that from [B]332[/B] to [B]333[/B] there are 51080, of which 27982 have been factored already (a 1.825 ratio). Of the remaining 23098: 2644 are out for factoring, 1041 are assigned for LL, and 19411 are available for assignment.

Come join us over here: [url]http://www.mersenneforum.org/showthread.php?t=10693[/url] Using the new LMH mini-how-to, one could get numerous assignments in the range.

I knew how many from 332 to 333M but I wasn't sure if I could find out from an intermittent exponents.

[QUOTE=Primeinator;177002]I knew how many from 332 to 333M but I wasn't sure if I could find out from an intermittent exponents.[/QUOTE]
41287 total expos. You could have looked on-line for a prime lister program. Or written a bute force one for this range. I happened to have one on my HDD. That with a quick line count (via excel). Viola!

[url]http://primes.utm.edu/links/[/url]

[QUOTE]41287 total expos. You could have looked on-line for a prime lister program. Or written a bute force one for this range. I happened to have one on my HDD. That with a quick line count (via excel). Viola!

[url]http://primes.utm.edu/links/[/url][/QUOTE]

Uncwilly, you are a god among men! I looked for such an online lister but could not find one. Evidently I am not that good at searching.

18505 are still unresolved in this range, per primenet again: [url]http://v5www.mersenne.org/report_factoring_effort/?exp_lo=332192831&exp_hi=332999999&bits_lo=0&bits_hi=99&txt=1&B1=Get+Data[/url]

And using excel to count.

And all to 64 bits already....

Where can I find such a program to download or is there a way to just script from number x to n and count how many primes? Any list I find has them in multiple columns. Thanks.

I recommend Pari/GP (Google it).

[quote=Primeinator;177019]Where can I find such a program to download or is there a way to just script from number x to n and count how many primes? Any list I find has them in multiple columns. Thanks.[/quote]

If you just want to count primes in range, [url=bbuhrow.googlepages.com]YAFU[/url] can do that (up to 2^32). It used to be able to list them out as well, but evidently I disabled that option :unsure:. There are win32 binaries you can download.

[QUOTE=Primeinator;177019]Where can I find such a program to download [/QUOTE]You should find some if you follow the links at [url]http://primes.utm.edu/links/[/url] (in posts 11 & 12 above :big grin:).

Chris Caldwell's site has "The nth prime page" which is excellent -- but I must admit I can;'t remember who actually wrote it!

[QUOTE=Primeinator;177019]Where can I find such a program to download or is there a way to just script from number x to n and count how many primes? Any list I find has them in multiple columns. Thanks.[/QUOTE]
I find that using excel, I can cut and paste a list into a blank spreadsheet and use the END down-arrow sequence to see the last element, the line number is the count. Or use the open function to open a text file and it does the same thing.

Attached is Luigi's primelister program (DOS mode), it is designed to make a text file to be a worktodo type file.

I think the old bsd games package for unix/bsd/linux systems has a primes command that will list any range of primes at least up to 32 bits.

[quote=CRGreathouse;177046]Chris Caldwell's site has "The nth prime page" which is excellent[/quote]
I second this recommendation.
[url]http://primes.utm.edu/nthprime/[/url]
[quote=CRGreathouse;177046]but I must admit I can;'t remember who actually wrote it![/quote]
[QUOTE]The text of this page, the programs, and all of the necessary data sets were provided by Andrew Booker.[/QUOTE]

Additional power/watts to run
 

[quote=Prime95;176980]Be sure to let them know that it costs money to run this project. A rough estimate is 50 watts per machine.[/quote]

Speaking of, I've got a question regarding this. If you have an older CPU w/o the newer power management / cpu power scaling features, running mprime under Linux should not cause the CPU to use more power, right?

In this case I'm thinking of some P3s I have at some colos.

The servers are on 24/7 anyway for dns, email, webserving and the like, and as far as I can tell don't support any cpu power scaling.

Second, if a newer CPU does support cpu power scaling, is there any way to get mprime to be intelligent about the cpu power scaling? For instance, when I'm just typing on IRC or reading a webpage, my CPU is mostly sitting at 800MHz. If I do something more intense, it may ratchet up to 1600 or I think peak at 2000 MHz.

I wouldn't mind running mprime, but I don't want it to cause the cpu power scaling to go up. In other works, if the load I'm generating has room for mprime to do some crunching within the 800MHz of idle, then I wouldn't mind it taking the idle cycles - so long as it doesn't bump the cpu power scaling up to 1000MHz or whatever. It appears my CPU supports "100mhzsteps", but I've never seen it run below 800MHz.

I think if mprime or prime95 or whatever did support this sort of intelligence and power saving with cpu power scaling and idle -> suspend/hibernate wasn't impacted, then places like a University would gladly accept running such a client.

[quote=jroysdon;177582]Speaking of, I've got a question regarding this. If you have an older CPU w/o the newer power management / cpu power scaling features, running mprime under Linux should not cause the CPU to use more power, right?
[/quote]
Any CPU capable of running Prime95 will use significantly more power when working than not. One of the first power optimizing techniques was to switch off the FPU when it was not doing anything. 486SL and newer CPUs also switch off other circuits to conserve power when not in use.

Linux (and I think newer versions of Windows) will also issue a HLT instruction whenever the scheduler don't find a process ready to run. The HLT instruction will put the processor in low power C1 sleep state until an interrupt wakes it up again. When mprime is running, the scheduler will always find work to do. Since mprime is using the FPU all the time (unless you are trial factoring), the FPU can't be switched off either.

[quote]The servers are on 24/7 anyway for dns, email, webserving and the like, and as far as I can tell don't support any cpu power scaling.[/quote]For this purpose CPU power scaling would actually make the servers use more power, not less. The CPU gets most work done per watt at high frequency. While the CPU is sleeping, the power usage is low and almost independent of CPU frequency. The amount of work to be done by a typical server is independent of CPU frequency, so it's better to get the work done quickly at a high frequency and get back to sleep ASAP, than to spend more time active doing work.

I did some measurements three years ago with a cheap wattmeter on a dual 3GHz Xeon, and measured the same wattage idling at all clock frequencies. Running mprime at full clock speed used about 100 watts more than when idling. The difference is probably higher on newer CPUs and chipsets.

Another interesting experiment to try is to start some other CPU intensive process when mprime is running. Your power consuption will decrease because the FPU will be idling and can be switched off when the other process is running.