Can Old PCs Help New PCs on Same Team?

[lfm]

Originally Posted by Unregistered

To all who've been helping me (and anybody else who'd like to add their two cents),

Thank you for the info you've provided so far. I'm a writer and not a mathematician, so these concepts don't come easily to me, but I *think* I'm slowly making progress.

Thats OK, we all have our crosses to bear. :-)

Pentium 75MHz, 128MB RAM tower, Windows for Workgroups 3.11 (yup!)
Pentium 233MHz, 80MB RAM notebook, Windows 98 Second Edition
Pentium II MMX 400MHz, 384MB RAM tower, Windows 98 Standard Edition
Pentium Dual-Core CPU T4200 @ 2.00GHz, 4GB RAM laptop, Vista 64-bit
Pentium Dual CPU E2200 @2.20GHz, 4GB RAM tower, Vista 64-bit

So, to judge from the Assignment Thresholds page, I'm guessing that I can have the two Pentiums do Trial Factoring, the Pentium II some P-1 Factoring, and the two Vista machines pretty much anything, except maybe a "preferred exponent." This leads to several questions:

1. Did I get that right?

Mostly, they seem to be on the lower end of the ranges, The 75 mhz will not be able to contribute a significant load but you are welcome to run it since you are paying the power bill for it (or you employers or whatever) You might have trouble with the MS-Windows 3.11 OS. Not sure if it still works for prime95 at all. Come to think of it, I'm not sure if ms-win 98 still works.You could always load Linux on those older machines, then they would work, they'd still be slow though of course.

2. Can (or may) the three lower-end PCs perform any "higher" operations than what I said above?

Theoretically you could force them to do other work but it could take years to complete even one result. The Pentium 75 mhz for instance will probably take over a month to return a TF result. anything else will be more time.

3. How can you tell (ahead of time) what the "reliability" and "confidence level" are for your CPU?

As they return their various results the reliability and confidence rating will be adjusted. You could try running the torture test for a few days to see if those machines will stand up to the strain.

4. What is a "preferred" exponent (what makes them preferable?), and how is that different from a non-preferred exponent?

The preferred exponents are marked manually by the administrators I believe. Not sure if there is any marked just now.

5. How would I assign (or ask for) work so that my slower machines will directly assist my faster machines in their pursuit of the cash award?

You can't really. Each machine works on it's own part of the problem. Only if you have machines with multiple "cores" can you group the cores together to work on a single item.

6. I seem to recall reading somewhere on this forum that the latest version of the Prime95 software is not the optimal one for older machines. Is that true, and if so, then which version(s) should I get for the older PCs, and where would I get it/them?

If you put Linux on them then you can run the current mprime. Not sure what you should do if you are forced to leave the obsolete OSes in place.

If you can't try Linux then MS-Windows XP is the oldest Microsoft OS you could use the current prime95 on I think.

Good luck.

lfm,

Thanks for the information, I learned a lot and I appreciate it!

I found this forum page (scroll down to post #3): http://mersenneforum.org/showthread....ight=prime.zip

Check out what it says about Windows 3.1 and the file prime.zip. Is that obsolete information? (The link didn't work, but maybe there's a different way to download the file.)

Also, see Uncwilly's post #5 in our thread here. Am I being way too optimistic thinking that it sounds like it's possible to coordinate different machines to help each other out?

I've been reading up on "Lone Mersenne Hunters," which I guess is what I'd be doing. Could it be that things work differently in that corner of the project (i.e., that you can set up old PCs to do work assisting newer ones in the same room)?

The more I read about it, the more the GIMPS project grows on me, so I may just go ahead and offer some systems to the general pool of work, but I'm still intrigued by the idea of setting up the old machines in my office to directly help the newer ones in finding the next Mersenne prime by doing prep/auxiliary work. Like you said, I'm paying the power bill. ;-)

Thanks for the good wishes.

Rodrigo

[chalsall]

Originally Posted by Unregistered

Also, see Uncwilly's post #5 in our thread here. Am I being way too optimistic thinking that it sounds like it's possible to coordinate different machines to help each other out?

To be perfectly honest, yes -- you are being way too optimistic. And, to be explicit, you should give up on your hope of having your machines "help" each other.

The reason is while it is theoretically possible, the reality is that all the low available exponents which qualify for the prize money have already been factored well past what an older machine can do efficiently.

This means that either your faster machines will wait literally years for your slower machine to do the appropriate factoring work, or else you'll have to focus on extremely high exponents which will take you so long to complete that someone else will almost certainly have found a qualifying Mersenne Prime (MP) well before you do. (Understand that the prize money is a "one off" for the first discoverer of any prime number which qualifies.)

Originally Posted by Unregistered

The more I read about it, the more the GIMPS project grows on me, so I may just go ahead and offer some systems to the general pool of work, but I'm still intrigued by the idea of setting up the old machines in my office to directly help the newer ones in finding the next Mersenne prime by doing prep/auxiliary work. Like you said, I'm paying the power bill. ;-)

Thanks for the good wishes.

We all welcome anyone and everyone who contribute to the project in any way (beyond "poaching", right Cheesehead ).

But it is important for you (as a "newbie" (absolutely no disrespect intended with that)) to understand the realities of the situation....

[lfm]

Originally Posted by Unregistered

I found this forum page (scroll down to post #3): http://mersenneforum.org/showthread....ight=prime.zip

OK, that is a good page. The only problem is it is a bit old now. For instance, there is special support for several types of CPUs besides just Pentiums now. Various Intel and AMD CPUs all have special code to make the most of any features available right up to the currently new Intel i7s.

Check out what it says about Windows 3.1 and the file prime.zip. Is that obsolete information? (The link didn't work, but maybe there's a different way to download the file.)

I'm not totally sure about what level of support remains for the old ms-win 3.x and 9x OSs. I guess you should just try and see. I know that if you have problems there will probably be a lot less support there for you.

Also, see Uncwilly's post #5 in our thread here. Am I being way too optimistic thinking that it sounds like it's possible to coordinate different machines to help each other out?

Yes, I guess UncWilly is just a little more optimistic about your level of expertise than I am. I guess you could try what he outlines but it presumes you will gain a higher level of understanding about the processes involved than what I was assuming you would be willing to put up with. He's an optimist. I'm a pessimist. Neither is totally wrong. To partly justify my pessimism, I have been fiddling about with this stuff for several years now and I don't think I'd want to try what UncWilly was suggesting, but maybe you will want to.

I've been reading up on "Lone Mersenne Hunters," which I guess is what I'd be doing. Could it be that things work differently in that corner of the project (i.e., that you can set up old PCs to do work assisting newer ones in the same room)?

I'm not sure if "assisting newer ones in the same room" means what you think it means. I'm not sure it means what I think it means.

The more I read about it, the more the GIMPS project grows on me, so I may just go ahead and offer some systems to the general pool of work, but I'm still intrigued by the idea of setting up the old machines in my office to directly help the newer ones in finding the next Mersenne prime by doing prep/auxiliary work. Like you said, I'm paying the power bill. ;-)

I'd suggest starting with your up-to-date Vista machines and running them for a while just to get your feet wet, then if you're still keen you could try to reserect the machines from the closet.

[Uncwilly]

Originally Posted by lfm

Yes, I guess UncWilly is just a little more optimistic about your level of expertise than I am.

I am talking theory. If one had their heart set on it they could, for example:
Get a 100m digit number assigned to their duo core machine.
Start that machine on the P-1 (because it will have the practical speed and memory).
Once it finishes that start it on the L-L test.
Start a couple or 3 Pentium (I or II) machines on the factoring that remains to be done. (This will take months to do on all 3 machines).
If a factor found, stop all machines that are working on that number, transfer the needed files to the fast machine and report the results.
Repeat.

As noted by others, this is not the best use for those machines.

Based upon the machines listed, I would suggest the following:
P75, leave off, put to use in the range that NRtarheel suggested, or look at OBD (linux would be a good choice.)
P233, put to use in the range that NRtarheel suggested, or OBD
P II, put to use in the range that NRtarheel suggested, factoring the exponents shown here at 64 and 65, or OBD
The other 2, do 100m digit numbers assigned by primenet for LL, if that is your desire, (that is where the prize is).

[S485122]

As I replied immediately :

Originally Posted by S485122

The old PCs will only contribute to the general pool. Their contribution to your direct search of a 100 000 000 digit prime (to earn the EFF cash price) would be marginal and difficult to arrange...

Jacob

chalsall,

I appreciate the frank but gentle talk! It helps me to understand faster and better.

The way the project operates is starting to come into focus. Let me know if I'm getting this right: The work on any given exponent is sequential rather than concurrent -- meaning that, for any given exponent, you (or I) need to finish steps A, B, and C before steps D, E, and F can take place. And, having the slow machines do the simple prep work means that the faster machines twiddle their chips in the meantime.

That said, is it possible to, for example, have the Vista machines do L-L tests on lower available exponents while they wait for the older systems to do the basic factoring work on very high exponents -- anticipating that when they finish with that, then the modern machines can take over the heavy crunching work? That way "everybody" would be doing something useful all along.

Looks like the practical problem would be getting all of these PCs to finish up and be ready together in such a neat and tidy fashion. Plus, I guess this is where the strategy point that you raised comes in -- I'd be committing all my machines to do a long-range operation while somebody else was finding the next MP at a lower range.

Am I getting it? :-)

Rodrigo

lfm,

Thanks for following up. Your reply helps to crystallize my thinking.

Here’s what I have in mind when I takla about assisting newer PCs in the same room:

I could request two exponents. To get started, I could have one of the modern PCs do L-L testing on one exponent that’s ready for that stage, while for the second exponent the other modern PC does (for example) P-1 work and the older PCs do the simplest tasks (is that trial factoring? or else some suitable rearrangement of the preliminary work as the project veterans would recommend). Then, when all of that is done, the two modern PCs would start L-L tests on the second exponent, and the older PCs could begin preparatory work on a third exponent. That way, all the PCs in my office would be working together (though at different times) toward the same goal on the same exponents.

There could be tweaks to this. For instance, I understand that having two cores doesn’t really help to speed up an L-L test, so I could have one core do P-1 on one exponent while the other does L-L on another exponent. I could end up handling three or more exponents at the same time among all these PCs.

How does that sound? I will trust you and the other good folks here to tell me if this is crazy, or feasible.

There does seem to be the issue chalsall brought up, that slower machines aren't as efficient at TF on larger exponents. If I did have them do TF, then -- relative to leaving them offline -- would that be 1) a fatal obstacle to my notion, or 2) slow down the GIMPS project as a whole, or 3) constitute merely a suboptimal use of these machines?

Probably the most sensible thing is, as you suggest, to get my feet wet by running the Vista machines for a while. But I’m trying to develop a sense of the possibilities here. :-)

Rodrigo

Uncwilly,

Please see my replies to chalsall and lfm, and let me know if any of this makes sense.

The arrangement that you suggest to carry out my idea sounds attractive, but I don’t yet have a firm grasp on the order of complexity for the various tasks that need to be performed for a given exponent. That would help me to see how each of my machines would fit in. Here's where my understanding of it stands now:

From the way I read the page http://www.mersenne.org/thresholds/ it looks like, for a given exponent, the two Pentiums (75 and 233MHz) could do TF, the Pentium II (400) could do P-1, and the two dual-cores could do DC or L-L. (I’m not sure how the first table, dealing with the size of the L2 cache, affects this.) Let me know if I'm missing some important factor (so to speak).

Lastly, what is OBD? That’s a term I haven’t come across yet (or missed).

Thank you!

Rodrigo

[cheesehead]

Rodrigo,

Here are some additions and alternatives to others' earlier responses:

Originally Posted by Unregistered

Also, see Uncwilly's post #5 in our thread here. Am I being way too optimistic thinking that it sounds like it's possible to coordinate different machines to help each other out?

You appear not to yet understand the technicalities of what Uncwilly described. It involves hand-editing various control files after you receive a L-L assignment, and "fooling" the PrimeNet server. It's not something you can set up to work automatically -- you'd have to manually intervene in each of your assignments. It's rather easy once you understand enough of what's going on, but you're not there yet.

I've been reading up on "Lone Mersenne Hunters," which I guess is what I'd be doing. Could it be that things work differently in that corner of the project (i.e., that you can set up old PCs to do work assisting newer ones in the same room)?

No.

All the considerations about dividing-up work that we've described above are exactly the same for LMH. "Things work" exactly the same for them -- it's just that LMH has a different focus. LMH participants are _NOT_ trying to win prize money. Instead, they are trying to help clear the way for future prize-seekers.

The more I read about it, the more the GIMPS project grows on me, so I may just go ahead and offer some systems to the general pool of work, but I'm still intrigued by the idea of setting up the old machines in my office to directly help the newer ones in finding the next Mersenne prime by doing prep/auxiliary work.

You can do that by ordinary participation with no manual interventions necessary -- if you don't care that both the old and new machines are your own. Instead, your old machines would be helping someone else's (99.99% certainty) newer ones.

What you can't do is arrange it so that your old machines directly help your newer ones -- except for the manual intervention and editing Uncwilly and I mentioned above.

(OTOH, someone else's old machines could be helping your newer ones, unbeknownst to you.)

[Uncwilly]

Originally Posted by Unregistered

The work on any given exponent is sequential rather than concurrent -- meaning that, for any given exponent, you (or I) need to finish steps A, B, and C before steps D, E, and F can take place. And, having the slow machines do the simple prep work means that the faster machines twiddle their chips in the meantime.

Yes, but. I was suggesting that if you were really desperate, you could do some concurrent work, at the risk of wasted effort. The "prep" work is actually just doing simple and shorter tests to eliminate numbers and avoid the long L-L primality test (the only one that proves a number prime and wins money.)

That said, is it possible to,...

Yes, the lower numbers would take about a month each on the fast machines. All of the factoring would take several months on the slow ones (Trial Factoring M(332192831), the smallest potential prize winner, from 70 to 77 bits would take a P90 machine 18 years, while a 3GHz single core would take ~30 days.)

I'd be committing all my machines to do a long-range operation while somebody else was finding the next MP at a lower range.

But, if prize money is your goal, the lower numbers don't have a chance.

Originally Posted by Unregistered

To get started, I could have one of the modern PCs do L-L testing on one exponent that’s ready for that stage, while for the second exponent the other modern PC does (for example) P-1 work and the older PCs do the simplest tasks (is that trial factoring? or else some suitable rearrangement of the preliminary work as the project veterans would recommend).

It is not feasible to have the modern machines share a number. Once you start to involve a second motherboard, your loses in communication speed make it impractical. P-1 is a different kind of test than trial factoring. TF is an exhaustive step wise test, while P-1 is different and can find bigger factors.

There could be tweaks to this. For instance, I understand that having two cores doesn’t really help to speed up an L-L test, so I could have one core do P-1 on one exponent while the other does L-L on another exponent. I could end up handling three or more exponents at the same time among all these PCs.

2 cores will give you a speed up (but less than 2x) vs a single core, while working on a single number. The best total throughput for the project is to have each core working on a different number, when one is tied up with system business, the others don't wait.

There does seem to be the issue chalsall brought up, that slower machines aren't as efficient at TF on larger exponents.

Mainly he was speaking of the 'bit levels'.

Originally Posted by Unregistered

the Pentium II (400) could do P-1

Using the number I mentioned above, it would take the machine ~7 years to do P-1 on that number, using default settings.

two dual-cores could do DC or L-L.

A good idea would be to have them do DC's (double checks, which are the same as L-L, just verifications) on one core of each and on the other core of each start a 100 million digit (potential prize winner), it will do all of the remaining TF and P-1 on the number, then it will start the L-L. After the number start into the L-L phase, you can switch to 2 cores working on the number (that take a single 3 GHz core ~1900 days).

Lastly, what is OBD? That’s a term I haven’t come across yet (or missed).

Operation Billion Digit, a fanciful, quixotic project doing factoring for the prize for a Billion Digit prime.
http://www.mersenneforum.org/forumdisplay.php?f=50
http://home.earthlink.net/~elevensmooth/Billion.html

Rodrigo

Please register for the forum, it is painless.

Originally Posted by cheesehead

Rodrigo,

Here are some additions and alternatives to others' earlier responses:

All of his answers are sound. I was initially answering your question of "could", not if one should. Others have been saying that it is not practical to do what you ask, this is true. Cheesehead and Jacob can tell you more about total throughput to benefit GIMPS.