Curiosity, some general questions

[JBerg]

I decided to get myself a new toy recently and ended up buying a 3.2ghz P4 with a 19" rackmount RAID box (8 drives, 5ms stepping, 10k RPM SCSI-3 ultra-wide) for a total of 200gb of fast drive space.

But then I realized I didn't have anything to really work a P4 and consequently started looking around for something to run on it. After all, with ECC2-109 finished, I couldn't very well run that on it. :cheers:

Finally I realized that 17orBust uses the Woltman PRP code and would make for a good test for the box. So today I loaded up the client and turned it loose. Wow, I didn't realize just how high you guys had the N sizes worked so high! Sheesh!

So naturally now I've got some questions about the project and can't seem to find any repository of information to easily find the answers. So...

1. How high are the candidates currently being sieved? With an N of 5.7m, I'm hoping that the seiving is well into the hundreds of trillions! If it was possible to find out what the sieving level was on an N when it is assigned, people might be interested in doing some sieving of their own on the single assigned N. With NewPGen, it would be possible to do a fixed N sieve with a single K and drive it upwards very fast. If you have a spare box, it might be worth running a non-stop sieve during the duration of the PRP test.

2. Have the general pool of candidates already had some degree of P-1 run? I see that some people are doing some P-1 testing, but it's not clear to me what candidates are being tested with P-1.

3. Has P+1 or ECM been explored for using on any of the candidates? With so many Athlons running around, it would seem that they would be better suited for such factor hunting rather than PRP testing.

4. I've noticed the client has some difficulties in getting connected to the server. Is this due to server load? If it would be of some help, I've got a fair amount of bandwidth and horsepower available now that ECC2-109 has finished. While I'm not about to volunteer as a staff member (been there, got the T-shirt, aint going back!), I wouldn't mind donating some bandwidth and/or horsepower to the cause.

Jay Berg
admin@eCompute.org

[Joe O]

Originally posted by JBerg
So naturally now I've got some questions about the project and can't seem to find any repository of information to easily find the answers. So...

1. How high are the candidates currently being sieved?
2. Have the general pool of candidates already had some degree of P-1 run?
3. Has P+1 or ECM been explored for using on any of the candidates?


1. At least to primes around 253 000 000 000 000. See this thread. (http://www.free-dc.org/forum/showthread.php?s=&threadid=3501)
2. Yes, but not enough people to keep up with PRPing. See this thread. (http://www.free-dc.org/forum/showthread.php?s=&threadid=3239)
3. Yes to both. I've done as much as I can with the machines available to me. Louie and others have done more. I've found P+1 disappointing, but perhaps a better implementation than PFGW or GMP would help. ECM gave me quite a few factors. I'm currently runniing Pollard's RHO on a few numbers, and am planning an SIQS implementation. I've also got an SNFS program that I plan to try.

Check out this website. (http://www.aooq73.dsl.pipex.com/)

Congratulations on the ECC2-109 solution. I worked on the 3 previous challenges, but gave this one a pass.

[Ken_g6[TA]]

Hey, Jay! I haven't seen you in a forum since we were optimizing the ECCP-109 client!:cheers:

As for #4, I think the alpha version of my Queue (see link in my sig) should be good enough to run a team Queue or something. It might freeze, but it shouldn't drop results.:)

[Death]

Originally posted by JBerg
After all, with ECC2-109 finished, I couldn't very well run that on it. :cheers:

Jay Berg
admin@eCompute.org

OMG!!! That's Jay!!! :notworthy

[JBerg]

Thanks for the information guys.

Now for one more question if I might?

The SoB client is cranking away on an N of 5752827. It is a 3.37ghz P4, just over 50% done with the N, and has almost exactly 5 days as the target completion time. The P4 box has nothing else running on it and is dedicated currently to the SoB client.

Now over on another box, I have standard PRP running on an N of 6000000. This box is also a P4, but running at 1.8ghz. Also it is an older P4 with smaller caches and slightly slower memory. PRP is just over 10% done on the N, with a target completion time of almost exactly 7.5 days.

Now think about this. I have a slightly smaller N on the high speed box. The slower box is running just over 50% the speed of the fast box. On the fast box I have the SoB client, with stock PRP running on the slow box.

So why does the SoB client take 5 days and PRP only needs 50% longer? Give that the SoB box is nearly twice as fast as the PRP box and the SoB client has a smaller N, I'd expect a bigger difference in time. I'm starting to think I need to run a PRP/SoB benchmark to see exactly what's going on!


PRP
- - -
Box speed 1.8ghz
N is larger (N=6m)
Total expected time: 7.5 days

SoB
- - -
Box speed 3.37ghz
N is smaller (N=5.75m)
Total expected time: 5.0 days

[prime95]

That's strange my 2.0GHz P4 shows just 2 days to PRP n=6000000.

Oh wait, my bad, that's the experimental version.....

[Alien88]

Jay,
The server did have some issues for a little bit. They should be fixed now - dave coded up a new server since we couldn't figure out the bugs with the old one.

See this thread (http://www.free-dc.org/forum/showthread.php?s=&threadid=6609) for more details about that.

--
Mike Garrison