SBprp - a better PRP tester

[Ken_g6]

First there was Proth.
Then there was PRP.
Then LLR/LLRNET.

And now you can use the Seventeen or Bust software to do PRP tests!

Introducing SBprp, part of the SBServer Package. SBprp is a Java program that takes over a PRP job by reading its ini and input files, serving work to Seventeen or Bust clients, and recording the results like PRP would. Currently, the SB client is faster than PRP on non-SSE2 chips, because it uses optimizations George Woltman hasn't yet added to PRP. You can also use several computers to work on the same PRP range in parallel! And because SBprp is partially open source (only the API is closed), you can modify it to suit the needs of your project!

Now for the bad news. Because the SB client is limited to Proth numbers (K*2^N+1), so is SBprp (which is why I posted in this forum.) And while the closed-source API code is pretty stable, the open-source SBprp code doesn't do a lot of checks on your input. It just ignores the first line of the input file, and assumes the .ini file is correct. Also of note, work units are currently set to time out after 24 hours (meaning they go to another client if you have one), and intermediate blocks are ignored. But, of course, the code in question is open-source, so you can fix these problems.

When used carefully, SBprp works very well! I found my most recent large prime, 383037*2^365350+1, on between one and three computers (none of them P4s) with SBprp. So why don't you give it a try?

[Citrix]

Thanks for the info. I think we will wait for George Woltman to put the improvements in the next PRP release, as most of the computers involved in the PRP section of this project are P4's. If you or someone else wants to modify SBPRP for PSP, we will for sure give it a try.


Thanks,
Citrix

[Mystwalker]

Originally Posted by Citrix

If you or someone else wants to modify SBPRP for PSP, we will for sure give it a try.

AFAI get it, SBprp can already be used for PSP...
That way, the project would have a new automatic client , as llrnet still uses the old algorithms (--> slower) - and the new version won't be residue-compatible with PRP.

[Ken_g6]

Originally Posted by Mystwalker

AFAI get it, SBprp can already be used for PSP...
That way, the project would have a new automatic client , as llrnet still uses the old algorithms (--> slower) - and the new version won't be residue-compatible with PRP.

It's easily usable by individuals. And it's faster, too!

You could try using it for the entire project , but I wouldn't recommend usage that heavy without certain improvements, most notably multithreading.

I may get around to some of these things. This is the most interesting personal project I'm working on right now. But don't bet the farm on it.

[Citrix]

Could some one write some directions, I or some one else will test it out to see if it is any better.

Thanks,
Citrix

[Ken_g6]

Instructions:
1. Start a PRP job, then stop it and delete the "zXXXXXXXX" file.
2. Extract the contents of my zipfile to the same directory.
3. Run "java SBprp" from a command prompt in that directory.
4. Connect all SB clients to that server, port 5911. (Why 5911? Because 5911 = 0x1717 )

But don't go overboard. I'm almost done with that multithreaded implementation. It will probably handle up to about 100 clients (which means maybe 25-50 users). More than that will probably need a database. I keep meaning to learn JDBC (Java DataBase Connectivity)...

[Mystwalker]

When it comes to JDBC, maybe I can help...

[Ken_g6]

New Version 0.11! It's got threading, plus a few other fixes and tweaks. Edit: Oh, and it respects intermediate blocks, too!

Myst, I think I can do it when and if it comes to that. But the first question would be whether a database is needed, and the second whether Citrix can set one up.

I tried this new version with three machines searching for low (k*2^10240) primes. They didn't skip any, and found two primes I found years ago, plus four more, in just minutes!

The next question is, is the new LLR35 faster than SB

[Mystwalker]

Originally Posted by Ken_g6

Myst, I think I can do it when and if it comes to that. But the first question would be whether a database is needed, and the second whether Citrix can set one up.

A DB would be useful with a lot of concurrent accesses. The application doesn't need to care about serialization (resp. serializability) then. In addition, you could do certain operations (stats etc.) faster than with scripts on flat files.

The next question is, is the new LLR35 faster than SB

I got the impression that it's slightly faster than PRP3 (I guess PRP3 and SB are quite close when it comes to speed on P4s, I haven't tested it, though; an answer to this topic would be very interesting - could you test it (especially for various k and n sizes)) - but ltd saw no improvement, so I'm not 100% sure.
Bad thing is that residues of PRP/SB and LLR are not comparable.

[Ken_g6]

A DB would be useful with a lot of concurrent accesses. The application doesn't need to care about serialization (resp. serializability) then. In addition, you could do certain operations (stats etc.) faster than with scripts on flat files.

Well, you still have to make sure you synchronize things; e.g. allowing operations between reading the next WU and marking it as in progress would be a no-no. But both writing and reading the results would be much easier. The biggest help would probably be not having to rewrite that entire ini file each time something changes.

It's been so long since my databases class, I don't even know where to begin anymore. So if you want to write something up, be my guest; that's why I made it open source.

I got the impression that it's slightly faster than PRP3 (I guess PRP3 and SB are quite close when it comes to speed on P4s, I haven't tested it, though; an answer to this topic would be very interesting - could you test it (especially for various k and n sizes)) - but ltd saw no improvement, so I'm not 100% sure.

I don't have a P4, but on my Athlon XP there was no significant difference in speed. If LLR is faster, it's less than 1% faster, and the new LLR is still a beta. The SB client has been pretty well tested ; OTOH this server is Alpha quality.

[Mystwalker]

Originally Posted by Ken_g6

Well, you still have to make sure you synchronize things; e.g. allowing operations between reading the next WU and marking it as in progress would be a no-no.

That's what transactions are for.

So if you want to write something up, be my guest; that's why I made it open source.

I'm afraid that has to wait until mid February, as my exams start soon.
Maybe I'll look into it within this time span, though. Could be a good distraction to get my head clear.