PSP greeting to a new user (PSP@BOINC)

[ltd]

It's in Test mode at the moment.
You can already opt in for PSP LLR on the "your account" page but at the moment only a small number of WU is handed out to make sure everything works correct. The second set of 100 WU is handed out at the moment while the data from the first set comes in. Acording to Rytis everything looks fine.

[hhh]

Too soon, all this... It's nice to find primes, but we should still sieve some months; I would even prefer to sieve until the threshold of computability.

If there is any way to slow down the release of these numbers (only a handful a day or so), you should do so. IMO.

H.

[ltd]

the BOINC server is now handing out PSP llr test. We are out of the beta test and in "production"

@hhh: I think it is not to early to do the llr testing. Our goal is to find the the first prime per k and not to test all "n".

And when you look at the results comming back from sieving you can see that we are not undersieved for the level of llr testing we are doing now.

[hhh]

Well, I have some reasons for my belief; I post about it from time to time, but never get an answer, probably because finding primes is more sexy then stupid sieving. Or perhaps because my posts are too long and nobody cares to read.

I will continue to post, but first I will recall what I wrote at different locations. For instance here:
http://www.rieselsieve.com/forum/viewtopic.php?t=1288

You are still and always considering the relative performances between different machines in different tasks.
What is largely neglected IMO, is the comparison in ABSOLUTE performance for ONE machine, comparing sieve and LLR. If you take the sluggishest sieve-machine, say a P4, but sieve is not very advanced and spits out tons of factors, you should sieve with that P4.
Sieve is advancing very quickly. So I'd like to propose some questions, which I cannot answer, because I'm not very much in Rieselsieve:

What is the absolute performance relation, say, the current average number of saved LLR-hours per sieve hour on a poor sieving machine?

Where is the current limit of the sieve software (it was extended some time ago, but to what extent)?

The transition of (absolutely) "sieving more efficient" to "LLR more efficient" is it already reached?

If not, shouldn't rieselsieve SIEVE?

Yours H.

The usual (WRONG) calculation is: It takes 1 day to run a test, and you have to wait 1 day for a factor. Let's start testing!

Why is it wrong? Because the factor is worth much more than 1 day of work. If it corresponds to an (average) test with n=25M, it saves WEEKS of work. And even if you have a 3/4 chance of finding a prime before n=25M, and to never need this factor, you still have a 1/4 chance to need it later, and to save say one week of work.

Additionally, the sooner you sieve, the more valuable factors you find, and the higher your efficiency is.
And the sooner you start LLRing, the more factors are passed by and the more tests are going to be run although you could have found a factor in no time.

If you sieve only one year full speed, you will have done all the sieving, you can sit back and do nothing but LLR. If you start LLR now, you get a prime in half a year and are happy, you can print it out and put at your wall, but it doesn't improve your efficiency at all (please don't speak about the 2% speed increase in sieving. It's ridiculous compared to fully saved LLR tests).

One cannot object my reasoning without (a) exterior factors like psychology that I didn't take into account, or (b) statistics that prove that sieving is indeed really too highly advanced. But in this case, it should be stopped completely, and nobody seems to be suggesting this.

I would like to know these exterior factors.

Yours H.

[VJS]

hhh,

I had a huge long post trying to answer your questions and it was lost due to login??? ARRGGGHHHH!!!!

ANyways long story short, NO we shouldn't stop sieveing but Yes we have to do more llr.

Long story short.

Assume we sieve to 9000T and stop. By doing this we will find 100% of the factors from 0T to 9000T.

What have we done thus far??

We have already sieved to roughly 1700T about 20% of the work.
With this first 20% we already found about 85% of the factors we would expect to find between 0-9000T.

If we keep extended to say 500000T which is HUGE!!! and would probably take our life time to complete.

We would have completed 0.3% of the work and already found about 55% of the factors we would ever find.

In order to complete this project by sieve we would have to sieve to roughly.
p=2^25000000, 5000000T is less than p=2^70.

[hhh]

All right, if you want to have the long story short, let me ask some questions.

Let's assume an average PC. Choose whatever you want. P4, or CoreDuo, say.

How long takes an LLR test at the levels 5M, 10M, 20M, 30M, 40M, 50M?

How long does it take to find a factor at p= 2P, 4P, 8P, 16P?

How many primes do you expect until 50M?


From the answers, you can calculate a sieve depth at which sieving does not save any cycles anymore. If you give me the numbers, I will do it.

And here comes my claim: For a maximum efficiency, one should sieve to that very sieve depth FIRST, and THEN do LLR. This sieve depth is not going to be 500P, but not 2P neither. Something between 5P and 20P, I'd estimate, but I am looking forward to the numbers, and then I'll be glad to be proven wrong (don't know if you can say that in English).

Yours H.

[VJS]

Originally Posted by hhh

All right, if you want to have the long story short, let me ask some questions.

Let's assume an average PC. Choose whatever you want. P4, or CoreDuo, say.

FINE THIS DOESN"T MATTER YOU NEED TO USE RATIOS ANYWAYS.

How long takes an LLR test at the levels 5M, 10M, 20M, 30M, 40M, 50M?

This is in the scoring debate for SoB it's something like

Time = n^2 log n or Time = n log n

How long does it take to find a factor at p= 2P, 4P, 8P, 16P?

As I stated below your chances of finding a factor halve with doubleing of p

How many primes do you expect until 50M?

Good question, I'd guess 4-7 remain after 50M and 1-2 remain after 100M.

From the answers, you can calculate a sieve depth at which sieving does not save any cycles anymore. If you give me the numbers, I will do it.

And here comes my claim: For a maximum efficiency, one should sieve to that very sieve depth FIRST, and THEN do LLR. This sieve depth is not going to be 500P, but not 2P neither. Something between 5P and 20P, I'd estimate, but I am looking forward to the numbers, and then I'll be glad to be proven wrong (don't know if you can say that in English).

Yours H.

ANswers within the text

[ltd]

Primegrid did reach another milestone today.

With sieving under BOINC we did complete an amount of work that
is equal to a range of 1500T.
This is the same amount as the complete manual sieving (inclusive SOB)
has done or only reserved by now.

[VJS]

The power of sieve is simply awesome on this level.

If you look at the stats over at sob sieve they are still dropping like flies as well.

[benjackson]

Is there a plan for manual sieving once it gets to 1500T? It's been slowing down a lot lately but still should run out of ranges in about two months or less.

[Joe O]

Originally Posted by benjackson

Is there a plan for manual sieving once it gets to 1500T? It's been slowing down a lot lately but still should run out of ranges in about two months or less.

I plan to contact all 12 users of the manual reservations system and ask them to transfer to BOINC.