Distributed sieving discussions

[geoff]

Originally Posted by michaf

The sieve would start from where the prp-ing left off, and eliminating the newly prp-d numbers every now and then. Same as when a new prime is found --> eliminate that one from the sieve

When a new prime is found all the previous factors and prp residues are no longer needed, as it doesn't matter if we missed a smaller prime.

But if we start the sieve from n=70,000 or wherever the prp effort gets to, or remove terms as they are prp tested, then we lose the opportunity to double check the earlier prp effort as part of the main sieve, we would have to start a seperate sieve later for that. It might be reasonable to do it that way, I guess it depends on how likely it is that we have already missed some primes.

[axn]

Originally Posted by geoff

But if we start the sieve from n=70,000 or wherever the prp effort gets to, or remove terms as they are prp tested, then we lose the opportunity to double check the earlier prp effort as part of the main sieve, we would have to start a seperate sieve later for that. It might be reasonable to do it that way, I guess it depends on how likely it is that we have already missed some primes.

The speedup obtained by removing all n < 100000 (from a sieve file containing n upto 2M) will be very negligible (about 2.5%).

Anyway, current sieve file is for 0 < n <= 2M.

[konrad127123]

I have 100MB of webspace at the moment that could be used. As long as people are reasonable with not downloading the same file repeatedly, it should be OK.

[ltd]

I have also webspace available. There should be no problem with download traffic as i have 1GB free per month.

Lars

[gribozavr]

ohmdpp.5gigs.com has a transfer limit of 5Gb/month. Currently I'm using about 20Mb/month, so I can give out pretty much.

[michaf]

Originally Posted by geoff

Will the distributed sieve start from n=0? Is 2 million the best upper bound?

Do we have the tools to record who found the factors, perhaps for use in a future scoring system?

Should we be collecting prp/llr residues now?

starting from 0 is more logical indeed when you think about it; 2M the best well... if we get that far, a new sieve from 2 to 5M or whatever is created ''quick'' enough with distributed sieving ...

For me, scoring isn't important, just for fun (and with only one P4, I'll be dangling on the end of such a list anyway :>)

collecting residues are maybe a good option; two primes are already found by double checking on sierpinski's, so it DOES happen too much. If we do collect them however, there should be an easy way to compare them, and someone to collect them of course.

[geoff]

Originally Posted by michaf

collecting residues are maybe a good option; two primes are already found by double checking on sierpinski's, so it DOES happen too much. If we do collect them however, there should be an easy way to compare them, and someone to collect them of course.

Everyone should keep a copy of their PRP/LLR results.txt files, and then we can work out how to deal with them later. It will be a much easier job once the number of remaining candidates has been reduced a bit.

[geoff]

My reservations have all been prp tested to n=80,000. I will continue with them until the distributed sieve starts if that is OK.

[axn]

With only one range below 70k pending, we are fast approaching the distributed sieving phase.

Right now, I am sieving 331 sequences from 1 <= n <= 2M. The sieve is currently at p=8G and the sieve rate is 8000 p/sec (on an Athlon32 2400+). Between 1.5 and 3 sec / n.

9.26M candidates left. Unzipped, the ABCD format, it comes to 36.5MB. Zipped, it is 7.3MB.

Q: Of the many offers for hosting in this thread, which one should be used? Or should I start a poll for that one?

[michaf]

Originally Posted by axn1

Q: Of the many offers for hosting in this thread, which one should be used? Or should I start a poll for that one?

I'd say either three offers with the larger available bandwidth should suffice;
my 50Mb is a bit smallish compared to the others...

no preferred place to put the files for me.

[masser]

We should be careful with terms like 'billions' and G; these mean different things in different countries. Perhaps we should use 5e9 = 5*10^9, or some such terminology...