assigning large amounts

[kracker]

Ok, maybe it's somewhere here.. but how do I reserve large numbers of assignments? (over ~700) I put it in Prime95, but it stops after a while, and it's sorta slow.

Any "tips" would be nice

[petrw1]

Originally Posted by kracker

Ok, maybe it's somewhere here.. but how do I reserve large numbers of assignments? (over ~700) I put it in Prime95, but it stops after a while, and it's sorta slow.

Any "tips" would be nice

Are you talking about the "send new end dates" process?
- It is not uncommon for it slow or freeze at times with very large batches but the assignments are still yours.

Or are you talking about the Manual Assignments page...
- I've hardly used it

Make sure in prime.txt that:
- MaxExponents (default 15 per worker) and UnreserveDays (Default 30?) are set high enough.

[Uncwilly]

Some of the people that work in the LMH zone (above the current LMH wave's bit level), just grab the data, post their intent to the proper thread and go.
Mike (Xyzzy) is experienced in doing large blocks.

[chalsall]

Originally Posted by Uncwilly

Some of the people that work in the LMH zone (above the current LMH wave's bit level), just grab the data, post their intent to the proper thread and go.

As Uncwilly says, for ranges which are not likely to be assigned by Primenet most people simply announce their intentions, and do the work. This has less impact on Primenet since it doesn't need to deal with thousands of AIDs.

For those ranges where others might automatically be assigned the work, it is best to get a real AID to ensure no "toes are stepped on" (read: duplication of effort). G72 uses a simple "slaved" mprime instance driven by Perl to import work to be assigned.

The following Perl code is similar to what I use -- it's been modified to not require a DB API -- simply works with flat files. Place the script in a directory where you have an instance of mprime configured to fetch only a single assignment using your username and a machine's name and run it from there, after building a list of desired candidates.

#!/usr/bin/perl

open (IN, "input.txt");  # List of desired candidates

while (<IN>) {
   print "$_";

   if (/.*=(\d*),(\d*),(\d*)/) {  # Look for "Factor=[Exponent],[From],[To]"

      open(OUT, ">worktodo.txt");
      print OUT "$_";
      close(OUT);

      `./mprime -c`;  # Run mprime to get an AID.

      # Adds the results to a file.
      `cat worktodo.txt >> full_fetched.txt`;

      sleep(1);  # Lessen load on Primenet.
   }
}

# Remove any lines which are not an assignment.
`grep -v N\/A full_fetched.txt | grep -v Worker | grep -v ^$ > fetched_with_aid.txt`;

# Clean up after ourselves so the next time the script runs we don't have duplication.
`rm full_fetched.txt`;

# Full assignments are in the file "fetched_with_aid.txt".

Lastly, another option is to simply use the manual assignments page, although you should limit the requests to a few hundred at a time.

[kracker]

So if it is a range that is not at all likely to be worked on, is it enough to just "announce" that I am working on it?

[chalsall]

Originally Posted by kracker

So if it is a range that is not at all likely to be worked on, is it enough to just "announce" that I am working on it?

Yup. Several of us were doing this a couple of years ago. So you know, 7xxM and 9xxM are still being worked, as is (obviously) 33xM. But everything else is pretty much open.

[Uncwilly]

Originally Posted by chalsall

still being worked, as is (obviously) 33xM.

But the 33x has a thread and those of use that are working in the range watch it. Beside I can tell you where in the 332.19-332.99 is safe. Right now 332.46 to 332.99 is idle and is largely at 70 bits. There are other smaller ranges that are also open for higher bit levels. Also anything that shows up as unassigned above 76 is fair game (when taken in chunks of a week's worth of work at a time or so.

[TheMawn]

I've grabbed another batch of 400M factoring work, taking stuff up to 2⁶⁸. If you look on mersenne.info, you'll be able to see my footprint (I'm kind of interested in seeing how long it'll stay there).

I've been reading up a bit on the LMH stuff to see if anything is current, and it sort of looks like no. I was going to officially announce my work in 400M but the people attacking from below are still in the mid 200M and the people attacking from above are just finishing 600M, so 400M is a bit of a graveyard for now, so I don't think announcing it is going to have much of an effect.

I did come across this, though, which made me smile.

Originally Posted by ckdo

I have a lone P3 happily crunching away in the 740M-741M range, at like 0.04M/week.

0.04M/week. Oh how times have changed...


EDIT: I just remembered I had another question about this. Even taking exponents up to 2⁶⁸ is fairly quick, and I was wondering if there was a way to submit the factors results and somehow say "...and no factors anywhere else". I take the exponents 50,000 at a time and I usually submit results through Primenet in small chunks as I go, but it would seem a bit easier to say no factors for M4XX,XXX,XXX through M4YY,YYY,YYY from 2⁶⁶ to 2⁶⁸ if possible.

[chalsall]

Originally Posted by TheMawn

Even taking exponents up to 2⁶⁸ is fairly quick, and I was wondering if there was a way to submit the factors results and somehow say "...and no factors anywhere else".

Not without involving George, Scott, or James' time at an SQL console.

So long as you don't submit too many results at a time, everything should be fine. And, interestingly, the "no factor found" results are fairly inexpensive -- simply a DB update.

On the other hand, "factor found" results appear to need the launch of an external program by Primenet to verify the factor. This has empirically been shown to result in a small memory leak which eventually results in resource starvation. This is the probable reason for the occasional need for a reboot of Primenet's IIS service.