mersenneforum.org - How/Where to get Jens Kruse Andersen's prime constellation sieve?

There are alternatives to it that are open source, but his benchmark numbers make it sound faster, but that's just based on a couple data points.

Jörg Waldvogel also has published results about constellations and believes his software is faster than mine based on time for a particular cluster search. Without source it's hard to tell. I know his method is similar in gross overview "a sieve to eliminate the small divisors in an entire block of numbers" followed by primality tests. In more specifics he generates candidates from the first 15 primes, then sieves from 53 to 541, then uses Pari's BPSW test. It's also done in a cluster framework.

My version chooses the candidate size dynamically so it will differ based on the size and clsuter, trying to meet a very approximate target goal of candidates per segment. Then sieves to 997, followed by BPSW in GMP. I do the dynamic limit partly because as part of a library I believe I should try to limit memory use, and work without someone changing the source code for each search. I suspect the real key is in reducing the time taken for sieving out the candidates, especially the first 5-20 next primes.

Anyway, email him as Paul said.