Printable View
Just an arguement...
With the recent prime discovery and another not predicted until 10m+ probably not until late next year. Isn't it more important to sieve as deep as possible now, even to go as far as stopping prp???
If we could get all users to actually sieve...
About 5000 currently pending tests, 1000 users, about 500T until we reach 2^50 the limit of cmov...
That's really only 100G per active computer or 500G per user. This could really be done in a week!!!
It would be a little difficult to organize because the client cannot do it automatically. Were it built in, we could just assign it all out in 50 G blocks and it would be easy, but this is currently not the case. :(
O.K. first off congrats to all who have participated in finding the prime, prp'ers, sievers, p-1'ers....
We find each of our primes late in the year this is due to many reasons however our most recent prime...
28433•2^7830457+1 was found as predicted.
7th prime in the 6.5M-10.1M range our next prime
8th prime is predicted 10.1M-16.8M we will be at roughtly 11m by this time next year.
Odd
Well what this means is we have about 1 year of testing for primes before we are predicted to get another! PREDICTED, yes it is possible we find another next week and I feel as though we may find one in the low 8's. Someone somewhere said primes come in bunches???
In either case let's assume we have one year for another prime.
We currently have 1000 active users with 5000 active machines. If we were to divide the remaing sieve range... the limit of the sieve client up. That would turn into.
100G per machine or 500G per user. If everyone stopped now and started sieving this would take 1 week, 2 weeks tops.
In that time we would normally do ~3000 tests.
Sieving the final 500T will probably result in another 20,000 factors, 4000 of those factors will be in the 8m-10m range. These 4000 factors saved would in effect give us back >2 weeks of prp testing before the next prime.
We would also eliminate a significant number of k/n pairs yet untested, probably about 12,000 and 19,000 double checks for free!!!
So basically all I'm saying is those of you with heavy horsepower out there, even the small guy. Lets sieve as deep as possible now that we found a prime. We can afford the loss in prp production to help the project complete the ultimate goal.
Also if you're considering leaving until next year same time, don't, try sieving for two weeks first.
At this point in time, I'd almost suggest all p4's sieve as well, but I won't go that far...
How much time will it take at the current speed to complete sieving?
jaat
sieving will not be completed. Before we reach the limit of the proth sieve client a new version will be released that can go alot farther. And as far as when sieving become less productive than PRP, well that is a highly debated number but definatly not before 2000T which at our current rate is still years off. And speed-ups in sieving software would of course move that number up to a significantly higher one. And expansion of the sievered range (sieving 1.5M-50M instead of 1.5M-20M) would also move the highest efficient range up too because more factors would be found. However it must also be noted that as the time required to finish PRP tests goes up there is more and more need for sieving. So the sieve will continue for the life of the project. Currently the sieve is much more efficient than PRP for eliminating canidates thoguh so it needs more help now then it will later.
As for an estimate on time to the limit of proth 2^50... depends upon participation, if everyone stopped prping and started sieving based upon available computer power probably 1-2 weeks!! But what your really asking?? Probably another year...
Keroberts is right, however the method used by the new sieving client is different than proth. We are starting to get into another topic altogether, if it is really fast then yes sieving will continue until way into the future, also it may run concurrently with proth and P-1 for non 64-bit machines who knows it's still alpha!?! There are also other factors to consider like new prp programs etc...
But the point is, as you know, we will go beyond 20m, also probably beyond 50m even 100m for some values of k. But in the very short term if we don't expect another prime until 10 or 11m the above arguement holds true. But stopping all prp (which won't happen) and just sieving out everything we can with proth to 2^50. We would basically get all that lost prp time back through newly eliminated k/n's before the new prime is found, and have eliminated all those k/n pairs above the next prime already done. Heck even if we find one prime tomorrow the arguement still holds true we just don't gain as much.
Man if we could simply throw an auto install switch for the whole project that would be great!!!
After sometime sieving would become slower than Prping in the present range of 1m-20m. Clearly, one could start sieving in the larger range of 20m-xxm but for Prping to reach till 20m would take quite a bit of time. So I thought it might make sense to shift resources at that point to Prping or P-1 factoring. I wanted to know if somebody had some calculations/guesses for this time event when this would happen. Clearly, it would be a bit vague as you guys pointed out that different softwares gets updated differently. Now that you mentioned it, I sense that maybe it might never be optimal to stop sieving as PrPing gets slower as n increases.
Also, one more thing. How does sieving time depend on size of p.
jaat
sieving gets slightly faster with p but we can only go upto 1000T with proth (p=2^50),
Yes at some point it will be better to switch to higher n values.
Some may argue now is the time to sieve above 20m, 1.2m-50m for example, then once we get close to 20m resieve the portion we missed with a 20m-50m dat or 20m-70m dat.
Other say wait until 18m until and restart at p=0 from 20m-40m...
This is really a point for great debate, check out some other threads for more info on this topic it's been discussed before.
We have the luxury of time to think about alternatives and create some brainstorming. No one knows for sure how relative effectiveness of sieve and PRP clients will change going forward.
Roughly speaking, we have 562T left to react 2^50, and with current average speed of 1.4T per day, it would take 13 months to reach there. For the time being, there is no rush to change things dramatically. What we're currently doing is trying to do is to see the potential outcomes of what if strategies.
Of course, things will change when a much faster sieve (or PRP) clients comes to the scene, but what to do is to be discussed based on relative performances of the new clients.
To SoB project coordinators:
Will the new SoB client support sieving (automatically connecting to the server and making reservations)?
I'm not a co-ordinator but I can probably answer your question.
The new client will probably not support Sieve.
There has been thought of implementing P-1 factoring however, this would be easier to design. Simply create a client front end which first trial factors your particular k/n prior to prp testing. If a factor is found submit the factor and abort the test.
I suppose the same could be done with sieve but I think the co-ordinators have alot on their plate with other client issues; computer integrety, first-pass, second-pass, user specific requests, let alone all of the server side refinements.
Agreed, i would like to see automated sieving but intergrating everything is a lot of work. The project has the most to gain by integrating P-1 factoring into the client (compared to sieving).