BOINC?

[victor]

Originally Posted by pacionet

Hardware requirements:

"Hardware: use a host with good CPU capacity (dual Xeon or Opteron), at least 2 GB of RAM, and at least 40 GB of free disk space. Do whatever you can to make it highly reliable (UPS power supply, RAID disk configuration, hot-swappable spares, tempurature-controlled machine room, etc.). If you anticipate a high-traffic project, use a machine whose RAM capacity is 8 GB or more, and that uses 64-bit processors. "

Of course. But you should keep in mind that the only data we got to carry are only a small amount of ASCII char...
Properly compressed, 2 month's work on a P4 weigh only about 100kb.
The server described could stand the load of 100'000 computers requiring 10 2Mb's work unit each day...

:)

[Brucifer]

A prime beef I have with the boinc world is that the points awarded is not even between linux and windows for the same work unit completed. Some projects have switched to granting points on a per work unit basis so that the stock boinc point computation code is bypassed. If this was done, then boinc would be a good move for you providing that the primary objective is for the project to gain more participants. It has really helped the RieselSieve sieving effort gain more participants. But again, many linux/unix users are turned off on boinc due to the points discrepancy.

[KEP]

Originally Posted by Brucifer

A prime beef I have with the boinc world is that the points awarded is not even between linux and windows for the same work unit completed. Some projects have switched to granting points on a per work unit basis so that the stock boinc point computation code is bypassed. If this was done, then boinc would be a good move for you providing that the primary objective is for the project to gain more participants. It has really helped the RieselSieve sieving effort gain more participants. But again, many linux/unix users are turned off on boinc due to the points discrepancy.

But even if going BOINC means a problem is created for our Linux users, there is really no trouble in giving them ranges, just like the few of us working for TPS is reserving ranges at the moment. Actually we could keep the site we have now, and maybe create another one, jsut to handle the BOINC community. Anyway we all have to face it, we will really never get anywhere, unless we developers an attractive BOINC community, because with the current pace, it will take approximately 3 years before we find a twin, and let me be straight out honest, thats just to long an effort for such a little accomplishment. Anyway I'll follow the developement, and the effort, and maybe if we someday go BOINC, I'll become a fulltime user :)

I suggest a fixed credit of approximately 0.25 Cobblestones, to be granted with BOINC, or maybe just 1 WU = 1 Result, which will give people the credit they claim, that should solve the issue with conflicting credit system, and grant the Linux users the credit they actually creates and crunches for the project :)

Regards!

[jasong]

Originally Posted by pacionet

But LLR BOINC client will be fast as our current LLR client ?

I think it's the same software, except it has the ability to ask for more work from a server, instead of having to be run manually.

One comment: If we do go to BOINC, it would probably be wise to make a work unit be comprised of significantly more than one test. I think 10-20 would be a good number.

[KEP]

Originally Posted by jasong

One comment: If we do go to BOINC, it would probably be wise to make a work unit be comprised of significantly more than one test. I think 10-20 would be a good number.

You're right, if we don't complete 10-20 results, will any server die because of the amount of handelings there need to be done, maybe as much as 500 results per minut. By 10-20 results, every single user will only make a request to the server every 20-40 minutes, which is a huge reduction. Still though we have to get a server that can handle 45,000 requests every 1,000 users each day. And with (never reachable) 100,000 hosts we need a server capeable of handeling 4,500,000 requests each day. But where there is a will there will come a way

[victor]

IMO, 10-20 results is a too small work unit. 1 test requires between 2 and 3 minutes. For a dial-up user, a connection each 20 (10*2) minutes is too often, when we could do a 250tests unit or 500tests unit.
I don't remember how long does a distributed.net or a seti@home work unit last, but it's way longer than 20 or 30 minutes :)

I tried something with this 'BOINC legacy wrapper', but can't get anything work, so I'm already giving up ;)

[ValerieVonck]

IMHO, you should first try get the goals of your project defined

Any primefinding project will stand or fall with the amount of candidates eliminated.

This we cold ask to Bryan over @ Riesel Sieve.
You should also do P-1 factoring, e.g. for my onw k @ Riesel Prime Search I have found 100+ factors that Proth sieve didnot found in normal time.

The LLR part, I think it is not possible to speed it up.
Only if you take into account the following:
- If _everyone_ is switching over to Core2 Duo/Quad/Eighty and/or 64 bit CPU's
- Make MAJOR speed ups in the code (if possible)
- Make it possible to share "intermediate" result files
- Make it possible to "continue" a test: if 3*2^120+-1 is not prime, make available the "calculated number", so that the next client can continue for 3*2^121+-1 if possible

My $0.02
Regards
Cedric

[jasong]

Originally Posted by CedricVonck

- Make it possible to "continue" a test: if 3*2^120+-1 is not prime, make available the "calculated number", so that the next client can continue for 3*2^121+-1 if possible

This has been discussed before in other Mersenne forums, the test results aren't transferrable to help with other tests.

[pacionet]

An alternative could be to write a simple client/server application where the client does something like this:

while (1) {

         system("llr.exe PARAMETERS");

         // parsing lresult.txt for new primes

        // send data to server through a socket

        // receive a new pre sieved file from the server

}

The server waits for clients' output data and send new pre-sieved files to clients (and also updates a database for stats).

[gribozavr]

Sounds like llrnet?

[pacionet]

Originally Posted by gribozavr

Sounds like llrnet?

I didn't know LLRnet , now I found some informations. Yes, I think it would be somthing similar but to reduce the traffic over the net it would be better to parse lresult.txt before sending it (sending only the primes found).

Where can we find LLRnet client and server source code ? (Maybe we can modify it )