Thoughts on future NPLB searches

[gd_barnes]

Originally Posted by henryzz

sounds good to me
one thing though:
some people might prefer to take one k all the way to 350k in one chunk
will reservations for the next level be available

No. Sieving would continue for n>150K. But if a person preferred to search one k, they could reserve it ahead of time. It might be several weeks or months between when they do n=50K-150K and the next higher n-range up but they could have their reservation in ahead of time.

The idea is to bring all the k's up to a "nominal" level, then a low level, then a medium level, and then finally to the level where they should be based on where k<=1001 is at; i.e. a little below it at n=500K.

My proposal at this point based on the discussion so far would be:

By k-value for:

n=50K-150K
n=150K-250K
n=250K-300K
n=300K-350K

Another option would be to make the n-ranges slowly reducing in order to even out the testing time for each "phase" of the drive. Perhaps something like:

n=50K-200K
n=200K-300K
n=300K-350K


Henry, the latter might be a little more to your liking with larger pieces in each testing range.

Any preference on the above? Any other ideas? After typing out the latter idea above, I think I may like it a little better. It will be a little less administrative work and people can search each k a little further. The only drawback is that the files would be a little bigger. Well...we could compensate by allowing people to take part of a k if they wanted. Even so, I suspect most people here would be willing to take even a high-weight k from n=50K-200K. It shouldn't take too long.

Edit: Keep in mind that after sieving has been completed far enough, the n=350K-500K portion of this range in our 9th drive will be going on at the same time as these lower ranges. LOTS of work to do! :-)


Gary

[kar_bon]

i'm preferring the fixed-k search too.
it's easier for me to update the primes for one k instead of jumping around the page.

all i need are the LLRnet-resultfiles, so if they available by downlaod, there's no proeblem for me to handle them.

[Flatlander]

Originally Posted by gd_barnes

...
n=50K-200K
n=200K-300K
n=300K-350K
Gary

My zipped lresults file for 15k-19k combined, n from 100k-260k is 397KB, so a file for just one k should be much less than the 244KB forum limit for posting. i.e. No need for emails.

[gd_barnes]

After suggesting the fixed-k idea, I feel I need to play "devils advocate" here to make sure I don't reduce the popularity of the effort...

Do people feel that reserving a k or a few k's at a time for n=50K-200K (or n=50K-150K) would be less "interesting" to people than to search, say, all ~500 k's for n=50K-52K, followed by n=52K-54K, etc.

The fun part about the low n-ranges is the huge # of primes that can be found very quickly whereas the higher part of a n=50K-200K range will slow that down to a certain extent.

On the other hand, if we progress by n-range, the effort could peter out near n=200K as people won't find it as interesting without top-5000 primes and with not a large # of primes coming out every day or two. The fixed-k approach kind of "averages" it out so perhaps it is better...not really sure.

I want to make sure that everyone gets both sides of the story here. Regardless of how we break it up, I think the less interesting part of the effort for people will be n=200K-350K.


Gary

[henryzz]

what is the lowest n value that you think a llrnet server could cope with
starting running on llrnet at n=200k would help solve the problem

[gd_barnes]

Originally Posted by henryzz

what is the lowest n value that you think a llrnet server could cope with
starting running on llrnet at n=200k would help solve the problem

It depends on the # of cores that are on it. Heck, it could likely start at n=50K if there were only 15-20 cores max on it and none of those cores requested more than about 5 pairs at a time.

Realistically, I think n=200K is a good point to start a server at...maybe n=150K as long as we keep an eye on the # of cores that are running it.

So, yes perhaps putting a server on it at n=200K would take care of the less interesting aspect of a somewhat long effort with no top-5000 primes.


Gary