Testing new Ranges for Sierpinski/Riesel

[rogue]

Originally Posted by wombatman

My house had a short power outage. Is there a good way to try and pick back up with srbsieve, at least in terms of having it remove all the k's that have a prime found, or, even better, near the last n it was searching?

You can take the created files and remove the n from the original list. That might be a lot of work though. I should probably have it checkpoint after each phase is complete.

[wombatman]

Could there a line in the ini file--something like LastN=""? That way, someone could just pick up from there or enter in the appropriate n value. Alternatively, a checkpoint would work fine too.

Edit: Is there a good way to restart it with the already proven k's removed? With, say, the pl_prime or using the pl_remain files that are created? That would be better than starting over completely.

[pepi37]

Originally Posted by wombatman

With, say, the pl_prime or using the pl_remain files that are created? That would be better than starting over completely.

Fantastic idea!
And I also noticed one thing: if I start srbsieve for test and exit : then when I try to edit srbsieve.ini : got lock worning: and until I kill that cmd windows that is invisible ( kill it in the task manager) cannot edit anything.

Last: please could some one compile it for 64 bit Linux: I try but lot of warnings, and there is no binaries at end :(

[pepi37]

Rogue one thing is still big mystery for me. Speed on different basses.
On first post you say

Note that this range of 5M took a little over 6 hours.

So you process on one core at base 3 up to 25K with little over 6 hours.

In my case I started with only 89 candidates left and 5 hours later I still have it 20 and test is up 13.8K

I know the base 3 is very prime base, but so much difference is impossible ( or not)

I take you settings , and yesterday on S708 take my settings but there is no significant difference.
Running on I7 -2700K under Win 7 -x64 on 4Ghz.
Any advise?

P.S with this speed I will finish my base around 1.5 days from start. on four cores...

[VBCurtis]

Originally Posted by pepi37

Rogue one thing is still big mystery for me. Speed on different basses.
On first post you say
So you process on one core at base 3 up to 25K with little over 6 hours.

In my case I started with only 89 candidates left and 5 hours later I still have it 20 and test is up 13.8K

I know the base 3 is very prime base, but so much difference is impossible ( or not)

I take you settings , and yesterday on S708 take my settings but there is no significant difference.
Running on I7 -2700K under Win 7 -x64 on 4Ghz.
Any advise?

P.S with this speed I will finish my base around 1.5 days from start. on four cores...

Consider: How big a number is 708^10000? How big is 3^25000?

[pepi37]

Originally Posted by VBCurtis

Consider: How big a number is 708^10000? How big is 3^25000?

On base 3 and on exponent 25K length is 11933 digits
On base 708 and on exponent 25K length is 71258 digits

x7 is factor :) so it will be at least 7x more time....
Thanks VBCurtis

[wombatman]

Originally Posted by wombatman

Could there a line in the ini file--something like LastN=""? That way, someone could just pick up from there or enter in the appropriate n value. Alternatively, a checkpoint would work fine too.

Edit: Is there a good way to restart it with the already proven k's removed? With, say, the pl_prime or using the pl_remain files that are created? That would be better than starting over completely.

I ended up just making a quick and (very) dirty python script to generate a file with all the n's in a given range (like n=3000 to 4000) based on the k's that remain in pl_remain.txt. That should work alright for now, and I can resume the file with pfgw if the power goes out again.

[Batalov]

Originally Posted by pepi37

x7 is factor :) so it will be at least 7x more time....

If M is the size factor, it will be ~ M³ more time, times ratio of series densities.

[pepi37]

Originally Posted by Batalov

If M is the size factor, it will be ~ M³ more time, times ratio of series densities.

And M is what? ( or in my case what is X time more?)

I am tempted to stop all, make deep sieve and finished with LLR :)
It looks faster :)

[rogue]

I've added recovery capability to this version, but it is completely untested. You will need to start a range from scratch so that it can build the checkpoint file as it executes. If it does work out of the box (which would surprise me), then there are conditions under which it cannot recover, such as incomplete writes to files. I can only think of that happening if there is a hardware problem, power less, or a software bug in my code.

[wombatman]

I will test it with R708 as a new base up to n=25k.