Sievulator

[hhh]

Dear friends, please have a look at the little spreadsheed I attached, together with a manual.
Quickstarting guide:

In the sheed:
Grey fields are for you to fill out (they are prefilled, but you can change these)
All the rest, don't touch it.
Yellow is giving you the results.

If you want to put in your timings, proceed as follows: Run an LLR test, note its n and the time it took. Put those two in.

Start a range with sr2sieve (no need to finish it), and note the expected remaining time in hours and how many factors are expected.

In the percentage field, you can paste your estimate of what percentage of the factors is going to be actually used.

If you would like to share your results, please post the stuff in the grey and yellow boxes only.

(Template)
PSP/SoB
[Your computer type]
N:xxxxxxxx
Time for that N:xxx h

Sieve range start: xxxxxxxx G
Time for that range: xxx h
Expected factors: xx,x

Result: strongly sieving!!! (Won't change, I promise)

Have fun, H.

[VJS]

hhh,

Why don't you use the following n's

firstpass 5M
secondpass 1.8M

80% n's found before 50M
use a core quad for the analysis I'm sure someone has some times.

Use that for the starting template...

I'm sure I did something wrong b/c when I worked the numbers it looked like we should do more llr

[hhh]

Did you use the Riesel Template or the PSP template? Riesel is the upper one.

Which n is used for timing shouldn't change much. Yet, honestly, I dont quite get your question, I have to admit.

H.

[VJS]

Humm,

Yup I did use the RS template, didn't think it would make much difference.

I guess my only problem with the template is the basis of the hours for LLR testing. I guess its a garbage in garbage out sort of thing but your template shows that a factor is worth 100 times it's weight in LLR testing depending on what you put in. I'd have to give it a little more thought.

[hhh]

Originally Posted by VJS

... that a factor is worth 100 times it's weight in LLR testing depending on what you put in.

I'm not quite sure about what you mean by "depending on what you put in", but I think that's kind of the point. A factor might be for n=5000000 and be worth a couple of hours on a fast pc, or for n=45000000 and be worth a couple of months. The chances of both events are the same, so a factor is roughly speaking worth at least a month, so until we wait a month for a factor, sieving is worth it.

OK, I have to admit that with riesel sieve, they approach the optimal sieve depth. If they sieve twice as deep or so, they are done; and sieving further will only save 1/50 of the tests or so; but yet, more effectively.

Timings? Anyone? I will upload a spreadsheet with the results for different machines as soon as I get the numbers.

H.

[Sloth]

Have numbers off a couple machines

PSP/SoB
P4 2.4GHz - socket 478 but not sure bus speed
N:5402590
Time for that N:45.42 h

Sieve range start: 10000000 G
Time for that range: 54 h
Expected factors: .53


PSP/SoB
C2D E4500 @ 2.2GHz
N:5053493
Time for that N:22.0 h

Sieve range start: 10000000 G
Time for that range: 29 h
Expected factors: .53

[hhh]

Thanks. I just made a new one, with even more conservative hypotheses: 80% of the factors go out of the window, and so on. Sieving is still an order of magnitude better than LLR.

Cheers, H.

[mdettweiler]

Originally Posted by hhh

Thanks. I just made a new one, with even more conservative hypotheses: 80% of the factors go out of the window, and so on. Sieving is still an order of magnitude better than LLR.

Cheers, H.

What exactly do you mean by "80% of the factors go out of the window"? Do you mean those are found for lower n's that have already been passed in terms of LLR? (If that's what you mean, then, why not simply remove everything from the dat file that's been already LLR tested--I'm sure you'd get a big speed boost! I know that removing individual numbers from the dat doesn't do much, but shrinking the range--now that's a different story. )

[hhh]

Originally Posted by Anonymous

What exactly do you mean by "80% of the factors go out of the window"? Do you mean those are found for lower n's that have already been passed in terms of LLR?

No. Those are not taken into account anyway. What I mean is that we are all the time finding factors for high n; Yet, those will probably never save an LLR test, because a prime for their k and a lower n is found, hence stopping all LLR-testing. 80% means that only 2 or 3 k resist to yield a prime until 50M.

H.