View Single Post
Old 2021-02-12, 03:49   #2
Romulan Interpreter
LaurV's Avatar
Jun 2011

100100111010112 Posts

Good job. Some random advice: that range is way too large. Split the range is 5-10 smaller ranges for sieving. You will be able to sieve higher and faster for each range, albeit the rate of eliminating the candidates per total range will be a bit slower, but you will compensate by having different sieve limits for each range, in such a way to equilibrate the LLR duration. Right now, your LLR may take (in a good computer) like 20 minutes per candidate for the start of the range, and 8 hours for the end of the range, so how do you calculate how much to sieve? Alternative, you may want to stop the LLR every few days/weeks and do some more sieving for a day or so, for the remaining candidates, to be sure that you always chose the task that eliminates the numbers faster. Write down how long it takes to eliminate candidates by sieving, and do a LLR test for a candidate which is down the list, about 5-10%. For example, the list is 3 thousand candidates, do a test for a candidate in position 200-300 in the list. Write down the time. Eliminate that candidate from the list once the LLR is done. Continue sieving on the list until you take the same time to find new factors. Then LLR the first 500-600 candidates. Repeat.

There is nothing like "i should have sieve it higher", stop the LLR, take a text editor, get rid of the tested (LLR-ed) candidates at the beginning of the list, then sieve the remaining list as high as you want. In fact, is recommended to do this periodically, as your LLR time gets higher per candidate, in a certain point you will eliminate them faster by continuing the sieving process. You don't need to start the sieve from scratch, you can continue any time from where it left, to higher primes. Make a backup in case you screw up the editing, if you didn't do that before.

The size of the sieving ranges needs to be not too large, and not too small. If too large, you will lose a lot of wall-clock time either by sieving too high (when you could possibly eliminate smaller candidates faster using LLR, therefore increasing the speed of the future sieving sessions) or sieving too low (when it would have been faster to eliminate larger candidates by more sieving). If too small, you will lose a lot of wall-clock time with the overhead of switching from sieving to LLR, sieve initialization, manual work, etc. Choosing the right range size is a matter of experience, system speed, number of cores, etc. That is why crus, for example, works in smaller ranges, 100k, 200k, etc.

Last fiddled with by LaurV on 2021-02-12 at 03:56
LaurV is offline   Reply With Quote