Misc. reserves/statuses/primes for k>2000 & k=1003

[PCZ]

Ok then i won't split the work up the way i indicated above.

Assume i will be using a quad for the sieving, so 4 cores.
2927 - 2953 so 14 k's
20K to 1M
I'll install a 64bit OS, windows not NIX.

[gd_barnes]

Originally Posted by PCZ

Ok then i won't split the work up the way i indicated above.

Assume i will be using a quad for the sieving, so 4 cores.
2927 - 2953 so 14 k's
350K to 1M
I'll install a 64bit OS, windows not NIX.

Before, getting into it, make sure you create a separate folder for each core of your machine and put a copy of srsieve, sr2sieve, and srfile in each one.

Here is what I would recommend:

1. LLR test one candidate at 70% of the n=350K-600K range, i.e. n=525K, and make note of the testing time.

2. Run srsieve to sieve the entire n-range to P=500M using the -a switch.

3. Run sr2sieve as follows:
(a) Run sr2sieve on all 4 cores to sieve to P=~100G splitting it up more-or-less evenly on each core.
(b) Use srfile using the -a switch to remove the factors found on all 4 cores from the sieve file in (a). Then copy that file that has all the factors removed to each of your folders. You'll use the -a switch because we want to continue sieving.
(c) Repeat (a) and (b) sieving this time to P=~1T. Before continuing, be sure and remove all factors found in all cores to get an even smaller sieve file for upcoming use.
(d) Use the file in (c) to determine optimum sieve depth. Let it run for 1-2 hours at P=1T to see what the removal rate is. If your testing time in #2 is 500 secs. and your current sieving removal rate is 100 secs, then you need to sieve to 500/100=5 times as high or P=~5T.
(e) Repeat (a) and (b) sieving this time to the optimum sieve depth as determined in (d).
(f) Run srfile one last time using the file in (e) and the -G switch to write to a file that LLR will understand.
(g) Manually split the file in (f) up in 2 parts: (1) A file that has only n=350K-600K. (2) A file that has only n=600K-1M. The manual splitting should be no problem because the file will be sorted by n-value.

4. Copy the file for n=350K-600K from #3(g)(1) to a folder with the LLR program in it and LLR the entire thing.

5. Run srfile using the -a switch on the file for n=600K-1M from #3(g)(2). This will create a new sieving file for only that n-range that needs to be sieved further.

6. Copy the file from #5 to the 4 sieving folders.

7. LLR test one candidate at 70% of the n=600K-1M range, i.e. n=880K, and make note of the testing time.

8. Repeat steps #3(d) thru #3(e) as many times as you feel is reasonable to get it sieved to the optimum depth for n=600K-1M.

9. Use srfile and the -G and -k switches to remove any final factors needed from #8 and write it to an LLR-readable file.

10. LLR the entire n=600K-1M range.


One thing that I'll mention here: This is a HUGE amount of work. People tend to forget how long it takes to test even a single average-weight k to n=1M. Before starting on the n=600K-1M range, I would suggest taking your test at n=880K as a basis for determining how much work the n=600K-1M range will be. I can guarantee it will be several CPU years. But on 2 quads or more, it should only be a few months so it's all a matter of how much processing power that you want to dedicate to it.

One final thing: Larger k-values LLR slower than smaller k-values at the same n-range. You'll still get the most bang for your testing buck running our well-sieved drives. Example: The 5th/6th/7th and individual-k drives are all sieved to P=26T. The individual k's (k=300-400) can be reserved right now. Files are waiting for anyone who wants them.


Gary

[PCZ]

Gary Thanks for all the info.

I have made a start.
Srseive bit is done and i have 2 instances of sr2sieve running at the moment, in different folders.
One doing 500M to 100G and the other doing 100G to 200G.
When there done i will remove the factors found from the sieve2937-2953.txt file and then split up further onto 4 cores as you suggested.

[Flatlander]

3023*2^400538-1 is prime.

[Flatlander]

3153*2^400790-1 is prime.

[Flatlander]

3171*2^404431-1 is prime.

[Flatlander]

3091*2^404441-1 is prime.

[PCZ]

Sieving is progressing.
Finished up to 4T last night, 5T some time tonight.
It's certainly a slow process.

At the weekend i'll review, and maybe shift some resources towards the sieve.

BTW
The AMD 4800+ sucks badly at sieving compared to the core2's.

[Flatlander]

3117*2^404826-1 is prime.

[gd_barnes]

Originally Posted by PCZ

Sieving is progressing.
Finished up to 4T last night, 5T some time tonight.
It's certainly a slow process.

At the weekend i'll review, and maybe shift some resources towards the sieve.

BTW
The AMD 4800+ sucks badly at sieving compared to the core2's.

Brian,

Do you suppose I could talk you into pulling a quad off of port 8000 and putting it on the team sieving drive for k=1003-2000/n=500K-1M. It would be nice if we were close to completion of the sieving for n=500K-600K, by the time port 8000 has reached n=500K.

Thanks for your huge contribution to the project!


Gary

[PCZ]

Yes i can do that.
Currently most of my computers are running 32bit though.
I'll need a day or 2 to get a 64bit OS on more of them.

At what depth should i start sieving ?
500G ranges should be OK to start, then see how it goes from there.
You just need the factors.txt files sent you yes ?