24 hour rally

[mdettweiler]

About a year and a half ago in the Sierpinski/Riesel Base 5 forum, Carlos (em99010pepe) proposed a 24 hour rally on the Riesel LLRnet server there (and later a 48-hour one, also on the Riesel server). The goal: for everyone who wanted to participate, to throw as much computing power as possible on the project for a 24-hour period, with the object of finding a bunch of primes! They were both highly successful, and thus I was thinking, "Why not do something like that here at NPLB?"

So, here is what I propose: that, beginning at 12:00 AM on Saturday, February 9 EST (GMT-5), everyone is invite to make a 24 hour run on the NPLB LLRnet server. This is the time to pull out all those old Pentium II's and get them crunching!

Just before the 24 hour period starts, Carlos can pull the results file from the LLRnet server, and I'll generate stats from it. Then, at the end of the 24 hours, he can pull the results file again, and I'll again generate stats--then I'll simply subtract the beginning totals from the end totals to come up with how much work was done by each user during the 24-hour rally. (Carlos, if the 12:00 AM EST time won't be a good time for you to pull the stats, with time zones and all, please feel free to suggest a new timeframe. ).

In order to count for this rally, testing must be done with the NPLB LLRnet server. Manual testing is nice too, but we can't count it with the rally because there's no way to know whether it was done within the 24-hour period.

Let's get crunching!

[mdettweiler]

I can safely pledge my primary machine (2.2Ghz Core 2 Duo) on both cores for the rally, but I can't say anything for sure about any other machines. I can safely promise, though, that I'll be crunching with whatever I can scrounge up.

[em99010pepe]

I always have good ideas...
I am in. Gonna search for more machines...

BTW, which range to crunch?

[mdettweiler]

Originally Posted by em99010pepe

I always have good ideas...
I am in. Gonna search for more machines...

BTW, which range to crunch?

We'll be crunching the 320K-333.2K range from the 2nd Drive, since this is for the LLRnet server.

As for my machines: Does anyone think an AMD K6 350Mhz (yeah, it's old) would be at all worth it? I read on the GIMPS web site that K6's absolutely stink at LLR, LL, and other FPU-intensive apps because the K6 doesn't do FPU pipelining and thus is twice as slow at FPU-intensive apps as a comparably clocked Pentium. (Right now I've got it running SR5 sieving--since that and other similar distributed sieving tasks are the only things it can manage to do in a reasonable time frame. It takes weeks to complete a 5G range that, on my main machine, usually takes less than 12 hours.)

[TimSorbet]

I'm in. (2.5 Ghz Athlon X2, both cores) Since I'm most likely not going to be awake at what's 11 PM my time zone, I'll start it the night before and stop it the next night (or morning after that). I can't really throw my CPU at NPLB full time yet, as I'm finishing a few GIMPS numbers, but they can stand to be delayed 24 hours to help a NPLB drive.

Make sure the server has plenty of numbers ready to test, and that you make sure it doesn't run out by reserving more during those 24 hours if necessary. (Just stating the obvious...)

Originally Posted by Anonymous

As for my machines: Does anyone think an AMD K6 350Mhz (yeah, it's old) would be at all worth it? I read on the GIMPS web site that K6's absolutely stink at LLR, LL, and other FPU-intensive apps because the K6 doesn't do FPU pipelining and thus is twice as slow at FPU-intensive apps as a comparably clocked Pentium. (Right now I've got it running SR5 sieving--since that and other similar distributed sieving tasks are the only things it can manage to do in a reasonable time frame. It takes weeks to complete a 5G range that, on my main machine, usually takes less than 12 hours.)

I'd say to use it. Even if it only gets, say, ten to thirty numbers completed in the drive, that's still better than nothing.

[mdettweiler]

Originally Posted by Mini-Geek

I'm in. (2.5 Ghz Athlon X2, both cores) Since I'm most likely not going to be awake at what's 11 PM my time zone, I'll start it the night before and stop it the next night (or morning after that). I can't really throw my CPU at NPLB full time yet, as I'm finishing a few GIMPS numbers, but they can stand to be delayed 24 hours to help a NPLB drive.

Make sure the server has plenty of numbers ready to test, and that you make sure it doesn't run out by reserving more during those 24 hours if necessary. (Just stating the obvious...)

Okay, I'll leave that up to Carlos at his discretion, since he's now handling the LLRnet reservations for NPLB.

Originally Posted by Mini-Geek

I'd say to use it. Even if it only gets, say, ten to thirty numbers completed in the drive, that's still better than nothing.

Okay. I was sort of thinking the same thing, I just wanted to get a second opinion.

[em99010pepe]

I'll bring:

4 x P4 3.0GHz (3 are confirmed)
1 x Q6600@ 2.9GHz
1 x AMD 64 3000+ (2GHz)

[mdettweiler]

Originally Posted by em99010pepe

I'll bring:

4 x P4 3.0GHz (3 are confirmed)
1 x Q6600@ 2.9GHz
1 x AMD 64 3000+ (2GHz)

Awesome!

My (hopeful) list:
1 x Core 2 Duo @ 2.2Ghz (definitely)
1 x AMD K6 @ 350Mhz (likely)
1 x P3 @ 1Ghz (somewhat likely)
1 x P3 @ 550 Mhz (maybe)

Total: 4.1Ghz, though it's actually more like 3.95Ghz since the AMD K6 only counts for 150Mhz when doing LLR. But hey, 3.95Ghz is still pretty cool.

If I can manage to get all of the above list crunching for the rally, then that will be the most machines I've ever had crunching.

[gd_barnes]

This sounds like a blast! Good idea.

I'll see what I can do. The problem that I have is that most of my home cores are doing main sieving ranges on 300<k<400. I don't really want to delay that by even a day. I'll think of something.

Maybe an older 0.8 Ghz P4 laptop that a friend lent me. I also have a 333 Mhz laptop from 2000 but that's ridiculous. I don't think it has the correct internet software and I'm sure it doesn't even have a wireless setup.

Just to confirm...this would be 11 PM Friday to 11 PM Saturday CST US (GMT-6)? When I hear 12 AM Saturday, I automatically think late Saturday night and not early Saturday morning.

One thing I should mention...That would be 5 AM for Carlos and 6 AM for Karsten on Saturday morning local time. I'm assuming that they could just have their machines hooked up before they go to bed unless they are VERY early birds!

Oh, just had a thought...I have my kids this weekend and they always bring their dual-core Athlon laptop over. I always do some sieving in the background on it. Athlons are slow at LLRing but with 2 cores on it, it'd be like 1 core of a 3.2 Ghz P4.

The only thing is...I'll have to figure out the server stuff for 2 cores on one machine. Is there any issue having 1 machine connected on 2 cores? (Loading of software twice, etc.)


Gary

[mdettweiler]

Originally Posted by gd_barnes

This sounds like a blast! Good idea.

I'll see what I can do. The problem that I have is that most of my home cores are doing main sieving ranges on 300<k<400. I don't really want to delay that by even a day. I'll think of something.

Maybe an older 0.8 Ghz P4 laptop that a friend lent me. I also have a 333 Mhz laptop from 2000 but that's ridiculous. I don't think it has the correct internet software and I'm sure it doesn't even have a wireless setup.

No, it's not ridiculous. I'm going to be using an AMD K6 350Mhz, and since K6's don't have FPU pipelining, on LLR it only counts as 175Mhz.

Originally Posted by gd_barnes

Just to confirm...this would be 11 PM Friday to 11 PM Saturday CST US (GMT-6)? When I hear 12 AM Saturday, I automatically think late Saturday night and not early Saturday morning.

Correct.

Originally Posted by gd_barnes

One thing I should mention...That would be 5 AM for Carlos and 6 AM for Karsten on Saturday morning local time. I'm assuming that they could just have their machines hooked up before they go to bed unless they are VERY early birds!

Hmm...and Carlos has to be available to pull the results file immediately before the rally starts. Carlos, are you OK with that? If you'd rather reschedule to a different time of day, that's fine with me (as long as it stays roughly within the same time frame).

Originally Posted by gd_barnes

Oh, just had a thought...I have my kids this weekend and they always bring their dual-core Athlon laptop over. I always do some sieving in the background on it. Athlons are slow at LLRing but with 2 cores on it, it'd be like 1 core of a 3.2 Ghz P4.

I don't think the difference for Athlons is as pronounced as you're thinking--it's only the K6's that are twice as slow. You should have almost as much power (no more than 10% of a detriment, I would think) as you would with a comparible P4.

Originally Posted by gd_barnes

The only thing is...I'll have to figure out the server stuff for 2 cores on one machine. Is there any issue having 1 machine connected on 2 cores? (Loading of software twice, etc.)

No, not at all. Just make sure you keep the two instances separate: You'll need to extract the LLRnet zip file twice, into a separate folder each time. Then, when you configure the llr-clientconfig.txt, make sure you set the GUIPort value to something different for one of them. For the corresponding llr-guiconfig.txt, you'll have to change the GUIPort value in it, too. Other than that, simply configure the llr-clientconfig.txt files the same--same username, same queue, etc. Then all you have to do is just start both of the llrnet.exe's. (You can even have both LLRnet's set up as a Windows service, as long as you adjust the server name for at least one of them in llr-clientconfig.txt so that they're not the same.)

[TimSorbet]

Originally Posted by em99010pepe

I'll bring:

4 x P4 3.0GHz (3 are confirmed)
1 x Q6600@ 2.9GHz (x 4 cores)
1 x AMD 64 3000+ (2GHz)

25.6 GHz combined (counting Q6600 as 4 x 2.9, since quad-core)

Originally Posted by Anonymous

1 x Core 2 Duo @ 2.2Ghz (definitely) (x 2 cores)
1 x AMD K6 @ 350Mhz (likely)
1 x P3 @ 1Ghz (somewhat likely)
1 x P3 @ 550 Mhz (maybe)

6.1 GHz combined (C2D as two, as before)

and me:
1 x Athlon X2 2.5 GHz (x 2 cores)
5 GHz total

That's probably a rough approximation of the relative work we'll complete over the 24 hours...admittedly, probably very rough because of factors such as how much, if at all, the CPU's used by other things, the different speeds because of the different architecture's and how efficient the code is on it, as well as how it bottlenecks (lookin' at you, quad-core ), etc. etc.

36.7 GHz total

Originally Posted by gd_barnes

Just to confirm...this would be 11 PM Friday to 11 PM Saturday CST US (GMT-6)? When I hear 12 AM Saturday, I automatically think late Saturday night and not early Saturday morning.

I was thinking it was 11 PM Sat - Sun CST...gotta love our 12-hour system...clarification please? Is it 24:00 (that is, the moment ending the day of...) Fri or Sat - 24:00 Sat or Sun, EST?