Missions

[Mersenne-ary]

I see in Completed Missions that Mersenne-aries in the past were performing Redo TF missions and few special DC missions. Is there a chance that we get again such diverse type of missions?

Thanks.

[dave_0273]

Well, this is something that I have always planned on implementing, but I guess things just kept on getting in the way. The main problem was that we didn't have much of a lead on the leading edge of doublechecking (with our p-1 checking) and some even got left behind due to the fact that we weren't moving fast enough. However, we are now just starting to get a decent lead, so now is a good time to try some other things.

Before I put my suggestions down as to what the future of mersenne-aries is, is there anyone who would like to give some suggestions. Is there some particular stats that you would like to see?? Would you like the blocks of work to be smaller/bigger/about the same, or even have a range of different sizes.

About the only thing that i don't want to do is do anything too far ahead of the leading edge of first time LL tests. The other thing is that the "redo-factoring" assignments were more just a once off. By accident, someone found some factors that really should have been picked up if they were indeed factored to the limit which was returned to the primenet server. Unless a range is once again discovered, I don't think that there will be any more redo-factoring missions, however I am still relatively new to this so who knows, there could be.

Please, feel free to offer any suggestions that you may have, no matter how small or "silly" they seem. I am eager to hear what people think about mersenne-aries since I took it over from GP2.

[ET_]

My impression is that without your coordination effort, no one would ever try to organize that P-1 search

Research is 5% genius and 95% sweat, and here we are, working to put each and every exponent to its place.

As for stats, I know that people running distributed projects love daily checkouts to improve their standing position, sometimes looking for more power to pass their "neighbours". But also know that maintaining a regular stat page may turn into a burden, so if you don't have much spare time to devote to it, just forget it

Luigi

[dave_0273]

Thanks for that ET. It is actually something that I have been thinking about for awhile now.

Ok, so lets say that "hypothetically" that I was to set up a stats page. What sort of stats would you want on there.

It would be easy enough to have..
number of exponents tested
number of factors found

What other stats would you like to have??

I am also in the process of getting some "interesting" LL tests to do. I have messaged the people who are doing similar work to make sure that I won't be stepping on anyones toes, and as soon as I get a reply, I would post the exponents here.

[ET_]

Originally Posted by dave_0273

It would be easy enough to have..
number of exponents tested
number of factors found

What other stats would you like to have??

You asked for it...

List of found factors in descending order
(with the discoverer?)

Most active factorers (number of tests done)
Factorers that discovrered most factors (number of factors found)

Global project time
Exponents (or P90 days) per day
Searcher's factoring time and/or speed

...and I stop here

Luigi

[Mersenne-ary]

First of all, thank you dave for your efforts in keeping Mersenne-aries alive.

Regarding stats - it's not a must for me. I enjoy being part of Mersenne-aries and doing some useful work for our project. I feel privileged to be part of this group.

I track for myself missions completed, and if you will start some stats page that's the info I would like to see for each Mersenne-ary.

Maybe smaller missions would be better (let's say 50 p-1 max), but that's just my opinion. They don't have to be all the same size. Maybe some smaller missions, 10 - 20 p-1s, for those who want to devote less time. It's OK with me if missions doesn't have round number of exponents (let's say 67 p-1s is OK mission for me).

I'm looking forward to new assignements, p-1s. TFs, DCs, LLs. What ever is needed I'll crunch it.

Marin's Mersenne-ary

[James Heinrich]

I'm not dying for online stats, but if they come I won't object. I offer my help if it's needed (as long as I can do it in PHP )

If this helps, this is a screenshot of the stats I keep for myself in Excel:
http://www.jamesheinrich.com/temp/p-1.png
edit: you can't put images in posts here?

The one thing I don't track, but would like to, is the total number of exponents processed, and the ratio of no-factor vs found-factor.

Another stat (that I don't track) could be the number of CPU-years spent factoring, and relate that to the number of CPU-years saved by found factors (hopefully time saved > time spent )

Histogram of found factors length:

46 |    1 | *
...
25 | 1234 | *********
24 | 2433 | ****************
...

100-blocks of P-1s is a good size for me; maybe you could put the jobs up in batches of 100s (as you've been doing) for people with a lot of machine time, but split one of the 100 blocks up into four 25-blocks so that people with less CPU time could each grab a small chunk.

[dave_0273]

Firstly ET_: Thanks for your suggestions.

Most active factorers (number of tests done)

easy

Factorers that discovrered most factors (number of factors found)

easy

Global project time
Exponents (or P90 days) per day
Searcher's factoring time and/or speed

Ok that one is a little more tricky. The problem is that I don't really get a daily update of amount of p-1 tests done. I use the "weekly" status files to track our progress.

So, how about some middle ground. How about I have a table that shows how many exponents are factored by mersenne-aries each time a new set of "weekly" status files comes out. Then, if I were to divide that by the number of days since the last set of status files, we could get a (rough) estimate as to the number of exponents cleared per day.

As for an individual speed, this is just as tricky. However, how about this. I could do it on a per month basis. I could rank people by the number of exponents tested over the last month. The reason that I don't suggest weekly is that it takes just over a week on a resonably fast computer to do a set of work at the moment. Therefore, if I were to do it on a weekly basis, it would look like they didn't do any crunching that week when really they were crunching all week. I know that some people submit work as they go, but the majority of people submit at the end of each block of work.

List of found factors in descending order
(with the discoverer?)

Ok, I don't think that I have the time to keep up with this one. I personally have found over 200 factors and I am sure that there are at least a few others that have done about the same amount of work. But, how about if I track the largest factor found by each person??

So, how about these compromises?? Your suggestions are very much appreciated. If you can think of anything better than what I have suggested, please post it.

Mersenne-ary: As for your suggestion about smaller sets, I will start to post some blocks of 50 or so. They used to be all blocks of 50, however a lot of our crunchers asked for bigger sets, so I upped it to 100. That way they could leave it for longer without having to manually update it. However, now that I know that people prefered the blocks of 50, I will now do a selection of both.

As for a "round" number in each group, that is something that I do to make it easier for myself. They are easier to post and easier to keep track of this way. However, there is the occassional "non-round" set that I put on the boards. This is when I get to the end of a 0.5M range and I just post what is left over.

Thank you for your suggestions. Please keep them coming!!

[James Heinrich]

I freely admit I know nothing about how you pull stats (or anything, including exponents to work on) out of PrimeNet, but I know it'd be possible to extract all the needed information out of results.txt dumps (including datestamps). The problem is, of course, that everyone would need to double-submit their results (one to PrimeNet, one to stats parser). I've been considering writing my own stats parser/grapher purely for my own purposes (to track things like factorsfound/totalexponents), although I have not yet done so.

[dave_0273]

Sorry James, I didn't leave you out, you just posted while I was writing.

The one thing I don't track, but would like to, is the total number of exponents processed, and the ratio of no-factor vs found-factor.

Easy, and done.

Histogram of found factors length:

Hmm, let me think about this one.... Interesting though. I will definately keep the longest factor found by each person, however... yeah, let me think about it.

As for the block size, I will now post some 50 and 100 exponent block sets. I would like to stay away from 25 block sets if I could. I just think that they would be too fidley. The other thing that I am worried about is that you are only "expected" to find a factor every 30 exponents or so. So I would prefer not to post sets that are smaller than the size set where you would expect to find at least one factor. (Does that make sense??)

Another stat (that I don't track) could be the number of CPU-years spent factoring, and relate that to the number of CPU-years saved by found factors (hopefully time saved > time spent )

This is something that I tried to do in my own stats (like you, I keep record of all my work too). However, it was too much work for what is was worth. I don't think that I will be doing anything to do with CPU years, unless a very easy way to work it out falls into my lap. Sorry.

Thanks for your suggestions.

[James Heinrich]

Sorry dave_0273, I edited my post above your last one while you were replying
I've moved it down here:

Originally Posted by dave_0273

I could do it on a per month basis. I could rank people by the number of exponents tested over the last month. The reason that I don't suggest weekly is that it takes just over a week on a resonably fast computer to do a set of work at the moment. Therefore, if I were to do it on a weekly basis, it would look like they didn't do any crunching that week when really they were crunching all week.

What about a past-4-week moving average? Taking any arbitary chunk of time (week, or month) you end up with problems when you complete 100 exponents just after the cutoff. A rolling average over 4 weeks or so should smooth out any spikes, and still give you a current number each week.

Originally Posted by dave_0273

Hmm, let me think about this one.... Interesting though. I will definately keep the longest factor found by each person, however... yeah, let me think about it.
....
I don't think that I will be doing anything to do with CPU years, unless a very easy way to work it out falls into my lap.

I can't really comment more in-depth as to what's doable, or not doable, since I don't know what format the input data you're working with is, and what kind of stats processing you envision (stored in a database of some sort, I imagine?)