Before you start... - Page 5

Xyzzy · 2012-07-31, 02:36

Dumb question alert!

So, how do we start doing polynomial selection?

Dubslow · 2012-07-31, 03:39

Quote:

Originally Posted by Xyzzy

Dumb question alert!

So, how do we start doing polynomial selection?

jasonp needs to make some fairly small modifications to Msieve's poly select code, and more importantly, someone needs to BOINCify it.

frmky, yoyo, and debrouxl are among the people who have previously BOINCd stuff. debrouxl has already said he's not familiar enough with Msieve's code (and doesn't have time; see his post above yours). I haven't seen yoyo around here, and frmky hasn't really been around anywhere recently.

I would be willing to at least help, but I've no clue how much work it is. Does anyone know where I can find a short description of what needs to be added to an app to BOINCify it? There doesn't really seem to be any documentation. I've found this page, but all it says about doing it is "Download the source". The only decently helpful line is "we recommend that you start by building the BOINC sample application example_app". It also seems to me that the page is designed more for people who are building a page from the ground up, rather than BOINCing existing code. Anybody on that list, am I reading too much into it? Should I just download the source and get on with it?

Edit: I just started going through the source, and it's all C++. I don't know any C++, though it's possible I can pick up enough of it to be useful anyways.

Edit2: Perhaps it would be simpler to start with frmky's or debrouxl's BOINCd GGNFS apps, and change calls to GGNFS to instead be calls to Msieve. (There'd be some other tweaks necessary, but I suspect it'd be easier than BOINCifying (parts of) Msieve from scratch.)

Xyzzy · 2012-07-31, 04:22

Dumb question alert!

Why does it have to be BOINCified?

Dubslow · 2012-07-31, 04:38

Quote:

Originally Posted by Xyzzy

Dumb question alert!

Why does it have to be BOINCified?

Do you have the hardware to do

Quote:

Originally Posted by xilman

a few thousand core years

in less than a year?

Although, jasonp, AFAIK poly select stage 1 has been ported to CUDA; since you and Mr. Bai were talking about doing only a teenee bit of S2, could what we would do through BOINC be ported to CUDA? Given Xyzzy's GPU farm, and other forum efforts, that might be enough to do poly select in less than a year.

I'm not sure what CUDA/Msieve's efficiency ratio is (compared to mfaktc), so I'm going to guess that it's around 50%*; then Xyzzy has the equivalent of around 200-300 cores**. Everyone else might add up to 500 cores, if we throw all our GPUs at it, and perhaps include fivemack's big hitter. Even then, that's 1000 cores, or still 2-3 years of work.

That's why BOINC, Mike.

*Given the embarrassingly parallel nature of trial factoring, I'm saying that mfaktc is (almost) "100% efficient at fully utilizing a GPU". Using PrimeNet's GHz-Days metric (and my GTX 460) I estimate that CUDALucas (large FFT muls) is about 20% efficient relative to mfaktc; that is, equivalent wall time on CUDALucas gets about 20% of the CPU-based GHz-Days that running mfaktc would. Given that poly select S1 is also AFAICT embarrassingly parallel, I chose the conservative 50% throughput relative to mfaktc.

**Xyzzy averages ~1000 GHz-Days per day; divide that 2 for my poly select/mfaktc efficiency guess, means around 500 GD/d; the GD metric was established on Core2s averaging 2-3 GHz, so that's 500/(2-3) ~ 200-300 Core2 cores. Of course, one Sandy Bridge/AVX core is worth 3-4 Core2 cores, so this is a questionable metric; then again, this is all ballpark estimates anyways.

LaurV · 2012-07-31, 05:02

Now as an outsider, I still don't see why boincify** it? Just need many workers and some coordination so each workers takes the hatchet and start chopping its range of coefficients, without chopping fingers and toes of other players.

(** edit: I mean, I know that the boinc part will bring a lot of benefits and billions of cores ready to contribute, but as it is not available and there is not so much chance to have it soon, I want to point that the poly selection phase may be done without it too, with a good coordination - again, this is outsider talking, combat me if I am wrong but do not kill me for it!).

axn · 2012-07-31, 05:39

*Where is RDS when you need him*

What is this obsession with rushing into compute? AFAICT, this project's a joke. If you're going to *seriously* do this, you need to plan it out from start to finish.

1. Create a forum
2. ???
3. Profit!

^^^^ That is NOT a plan.

There is no evidence whatsoever that current state of msieve polyselect can produce decent polys for something this size. The siever is vaporware. Post processing???

Dubslow · 2012-07-31, 05:46

Quote:

Originally Posted by axn

*Where is RDS when you need him*

What is this obsession with rushing into compute? AFAICT, this project's a joke. If you're going to *seriously* do this, you need to plan it out from start to finish.

1. Create a forum
2. ???
3. Profit!

^^^^ That is NOT a plan.

There is no evidence whatsoever that current state of msieve polyselect can produce decent polys for something this size. The siever is vaporware. Post processing???

jasonp seems to be under the impression that his poly select code is up to the task (as does xilman). We are aware of the lack of software for the rest of the steps. Have you read the thread?

Batalov · 2012-07-31, 05:58

It is a bit too early... to... say...

debrouxl · 2012-07-31, 07:58

The changes made by squalyl to the GGNFS code for RSALS, and subsequently modified by frmky for NFS@Home, are only several dozens of lines, IIRC, yielding a diff (against the original HEAD revision of ggnfs-lasieve4e.c in July-August 2009) of up to several hundreds of lines. See the first post of http://www.mersenneforum.org/showthread.php?t=12458 .
The gist of the work was to call the functions of the BOINC C API (start, stop, heartbeat, etc. - they have different names, but that's what they do) at the appropriate places.

But as I hinted above, RSALS is going to be merged into NFS@Home, and shut down in the next few weeks; squalyl, frmky, potentially others and of course I need to spend time preparing the transition.

pinhodecarlos · 2012-07-31, 08:07

Quote:

Originally Posted by debrouxl

But as I hinted above, RSALS is going to be merged into NFS@Home, and shut down in the next few weeks; squalyl, frmky, potentially others and of course I need to spend time preparing the transition.

Sorry for the offtopic but please don't forget to warn the usual post-processing suspects how they will now help (newsletter to all). Thank you.

debrouxl · 2012-07-31, 08:23

Of course, but I didn't have time to write said newsletter yet

2012-07-31, 05:02	#49
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 3×23×149 Posts	Now as an outsider, I still don't see why boincify it? Just need many workers and some coordination so each workers takes the hatchet and start chopping its range of coefficients, without chopping fingers and toes of other players. ( edit: I mean, I know that the boinc part will bring a lot of benefits and billions of cores ready to contribute, but as it is not available and there is not so much chance to have it soon, I want to point that the poly selection phase may be done without it too, with a good coordination - again, this is outsider talking, combat me if I am wrong but do not kill me for it!). Last fiddled with by LaurV on 2012-07-31 at 05:06

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Where should I start?	christian_	Information & Answers	9	2016-01-22 19:28
Where to start	Jellyfish420	Homework Help	46	2013-02-06 13:51
How to start?	Thomas11	Lone Mersenne Hunters	29	2008-12-21 13:47
how to start with P-1?	ValerieVonck	Marin's Mersenne-aries	8	2006-04-29 22:21
How to start?	OmbooHankvald	Factoring	15	2005-09-03 13:42

2012-07-31, 02:36	#45
Xyzzy Aug 2002 7×1,223 Posts	Dumb question alert! So, how do we start doing polynomial selection?

2012-07-31, 04:22	#47
Xyzzy Aug 2002 10000101110001₂ Posts	Dumb question alert! Why does it have to be BOINCified?

2012-07-31, 05:39	#50
axn Jun 2003 2×2,719 Posts	Where is RDS when you need him What is this obsession with rushing into compute? AFAICT, this project's a joke. If you're going to seriously do this, you need to plan it out from start to finish. 1. Create a forum 2. ??? 3. Profit! ^^^^ That is NOT a plan. There is no evidence whatsoever that current state of msieve polyselect can produce decent polys for something this size. The siever is vaporware. Post processing???

2012-07-31, 05:58	#52
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 3·7·479 Posts	It is a bit too early... to... say...

2012-07-31, 07:58	#53
debrouxl Sep 2009 11·89 Posts	The changes made by squalyl to the GGNFS code for RSALS, and subsequently modified by frmky for NFS@Home, are only several dozens of lines, IIRC, yielding a diff (against the original HEAD revision of ggnfs-lasieve4e.c in July-August 2009) of up to several hundreds of lines. See the first post of http://www.mersenneforum.org/showthread.php?t=12458 . The gist of the work was to call the functions of the BOINC C API (start, stop, heartbeat, etc. - they have different names, but that's what they do) at the appropriate places. But as I hinted above, RSALS is going to be merged into NFS@Home, and shut down in the next few weeks; squalyl, frmky, potentially others and of course I need to spend time preparing the transition.

2012-07-31, 08:23	#55
debrouxl Sep 2009 11·89 Posts	Of course, but I didn't have time to write said newsletter yet