distributed factoring

[scrap]

Is it possible to have a distributed client for factoring ?
It needs zero configuration stuff (or at least the same kind of configuration of SoB client).

Factoring is complex right know. I don't understand anything at hand made reservation of range, B1, B2, what CPU and memory for factoring, what to report, etc...
I think things must be more clear. In that way, I dream of a distributed client/server stuff for factoring.

Best regards,

scrap

[Frodo42]

I think it is planed to include the P-1 test into the V3 client that is being worked on right now.
I don't know if these plans are still relevant.

hc_grove have made a page to explain a bit more about factoring

But as it is now it does take a bit more to do factoring ... however resources are badly needed for this right now.

[Mystwalker]

Just look at my thread ECMNet Client/Server 2.5.6 for automated factoring?.

A distributed factoring with sbfactor seems to be possible - all we need is a server...

[Vato]

I'm game to host it, as long as I have a trusty volunteer to act as deputy on running the server app if I'm unavailable....

Say the word and I'll get it installed...
Of course, it may be better if this was done by the main SoB folks, since the
data will need to be updated from the main servers to remove things when
factors are found by sieving/PRP, or if PRP catches up.

Comments?

[Frodo42]

Hmm, well if it's only sbfactor that can run this there should still be room for run the much faster mprime.
But anyways could be interesting with distributed factoring.

[Vato]

It appears to compile/link just fine on debian box.

Just doing some checking to see if it really works.

[Mystwalker]

Originally posted by Vato
I'm game to host it, as long as I have a trusty volunteer to act as deputy on running the server app if I'm unavailable....

Excellent!
What exactly do you mean with this deputy?
Should he run the ECMserver himself or look after the ECMserver on your computer?
The latter is no problem, the former maybe possible (I just got an idea...).

Of course, it may be better if this was done by the main SoB folks, since the data will need to be updated from the main servers to remove things when factors are found by sieving/PRP, or if PRP catches up.

I guess keeping the server uptodate is a fairly easy work - some scripts should to it. After all, we have a regularly updated factoring list by MikeH.

What concerns me more is a way to put the found factors into the system while credit the resp. finders. One option could be to negotiate a new interface for factor submission with Louie & Dave - one with accepts a user name for every factor entered and needs no login.


Apart from that:
If anybody needs the Windows binaries (the archive only contains the source), I can provide them.

[Mystwalker]

*deletedSomeOverhastyAndUnsolicitousStuff*

ECMNet should work with expressions on SBfactor and (in principle though not tested so far) gmp-ecm*.

Rogue answered my question for a P-1 only server setup:

For your second question, not with the current code. The server is driven by ECM curve counts. A few changes would be needed in Candidate.SendWorkToDo to ignore ECM, although I don't know exactly what those changes would be without looking into it further.

*Does anyone know how gmp-ecm compares to SBfactor speedwise when it comes to P-1 factoring "our" numbers (--> Proth numbers with n > 8M & k < 70,000)?

[rogue]

Originally posted by Mystwalker
*Does anyone know how gmp-ecm compares to SBfactor speedwise when it comes to P-1 factoring "our" numbers (--> Proth numbers with n > 8M & k < 70,000)? [/B]

SBfactor is much faster since it uses GW's FFT multiply. BTW, I would recommend some testing of it first since it was added later in the development. I only had one person to help me test it and I haven't heard from him in months. I know he had some problems but AFAICT it had to do with running a CygWin build of the client outside of the CygWin.

[Vato]

I meant that I'd give the deputy a login to keep an eye on the server and fix things, though it looks like we could run a pair as mutual master/slaves for
resiliency.

I'm still trying to decode how we'd install the candidates into the server
in an automated fashion. MikeH, do you already do this sort of thing?
We'd also need to avoid assigning work that folks doing manual work
already have.

As far as scoring credit goes, my initial thoughts would be to have each user
submit manually using the existing process, though I reckon it should be quite easy to do this with a perl script - dunno about M$ though - maybe Cygwin.

[Vato]

I'm still game for this - just been busy recently.
I'm going to try working the server by hand for a while to get a feel for the care and feeding - then I'll automate it. I'll report back on progress shortly...