Question about cycle counting in MSieve

[schickel]

Jason,

How much time does MSieve spend cycle counting?

Would you save much time running it with the '-c' switch so it doesn't bother to count cycles before enough relations are gathered?

For instance, at 97 digits, MSieve wants to collect 77,000+ releation to finish the factorization. Would it make sense to run in 'client only' mode until 35,000+ full relations are collected, then run MSieve without the '-c' switch to allow it to finish?

Thanks,
Frank (a regular user of MSieve....)

[xilman]

Originally Posted by schickel

Jason,

How much time does MSieve spend cycle counting?

Would you save much time running it with the '-c' switch so it doesn't bother to count cycles before enough relations are gathered?

For instance, at 97 digits, MSieve wants to collect 77,000+ releation to finish the factorization. Would it make sense to run in 'client only' mode until 35,000+ full relations are collected, then run MSieve without the '-c' switch to allow it to finish?

Thanks,
Frank (a regular user of MSieve....)

I think it extremely unlikely that it will be useful. Unless you are uncannily good at predicting how many cycles will be generated, you are more likely to spend more time oversieving, or undershooting and having to sieve some more, than you will ever save by not counting cycles.

I've been using MPQS in one implementation or another since the late '80s and the above is the distillation of my experience. YMMV.


Paul

[jasonp]

Originally Posted by schickel

How much time does MSieve spend cycle counting?

Would you save much time running it with the '-c' switch so it doesn't bother to count cycles before enough relations are gathered?

For instance, at 97 digits, MSieve wants to collect 77,000+ releation to finish the factorization. Would it make sense to run in 'client only' mode until 35,000+ full relations are collected, then run MSieve without the '-c' switch to allow it to finish?

The computational cost of counting cycles is essentially zero; the only reason you would ever not want to do it is if you have multiple computers sieving, and don't want each computer allocating cycle structures that grow over time and will never be used.

Paul is right, the number of cycles is a nonlinear function of the number of relations found, and by the end of the factorization cycles are accumulating very rapidly. It is very likely that letting the program figure out when to stop yields the shortest runtime.

jasonp

[schickel]

Jason,

The computational cost of counting cycles is essentially zero; the only reason you would ever not want to do it is if you have multiple computers sieving, and don't want each computer allocating cycle structures that grow over time and will never be used.

OK, sounds good. Since I don't really have a software infrastructure to run on multiple computers, I'll just let my spare systems run free.

But since I do some factoring on my primary system, I'll run here without the cycle counting so I don't bog down...

Later,
Frank