TF: A job half done?

[CRGreathouse]

Let me offer a different explanation.

The software lets you choose what kind of work you want to do. One of the choices is called "LL": this choice means that, if there are prepared exponents available, the server will try to send them to you os you can run a LL test on them. If not, the server will send you exponents that you will prepare and then LL. There's never any agreement that the server has to send you anything.

[cheesehead]

Originally Posted by chalsall

To what rules are you referring?

The included license.txt, or http://mersenne.org/legal/ ?

Those, plus the operating rules of GIMPS and PrimeNet, which I think are drawn in by this legal clause:

"Participation Constitutes Agreement. "Participant" is understood to be an individual person, or a single individual person designated as the authorized representative of any group, team, organization or legal entity, who personally, or whose computer(s), accesses and/or communicates with the Web Site to perform, or cause to perform, mathematical calculations that are, or typically, systematically organized by GIMPS. This includes, but is not restricted to, use of Prime95, MPrime, GLucas, or similar software, automatically over the Internet or using the Web Site (manual testing forms, reports, APIs, etc.), to get data or work assignments or to upload results or other data. Participation constitutes agreement with the TCU by that individual and any group, team, organization or legal entity the Participant represents.
"

I don't actually understand what you are trying to claim from this language and the above.

I am simply arguing that those who wish and ask to do LL testing might reasonably expect to actually be assigned (and the client software do (and only do)) LL testing.

They are assigned LL testing!! Sometimes the assignment includes finishing up the last parts of TF and/or P-1, but if those find no factor, the program proceeds directly to the LL test.

Do you really mean "LL testing without any prior TF or P-1" or "LL-only" when you write "LL testing"? If so, please make that distinction explicit.

Is that an unreasonable expectation?

If what you mean is that you're insisting that LL requests _must_ be fulfilled with LL-only assignments, not assignments of LL preceded by the last parts of TF and P-1, then, yes, that is unreasonable. There isn't enough to go around.

Added: If you think that somewhere GIMPS or PrimeNet promises or offers LL-only assignments, please tell us where that is.

Is the problem simply that you insist on interpreting "LL" as "LL, and only LL, with no TF or P-1" in the context of assignments?

[CRGreathouse]

We're making hammers. The prestigious job is attaching the hammer's head to the handle, but we have trouble getting the handles shaped fast enough to supply the handle-attachers' needs. Occasionally we send them blocks of wood with the hammer-heads instead of finished handles with the heads; in these cases we have them shape the handles before they attach them. It would be crazy to do otherwise!

[davieddy]

Originally Posted by CRGreathouse

Let me offer a different explanation.

The software lets you choose what kind of work you want to do. One of the choices is called "LL": this choice means that, if there are prepared exponents available, the server will try to send them to you os you can run a LL test on them. If not, the server will send you exponents that you will prepare and then LL. There's never any agreement that the server has to send you anything.

Speaking as an overage/computersavvy/nonstupid? average punter:

I wouldn't have joined GIMPS if it wasn't as simple* as it sometimes sounds.

David

*Or should that have been "user friendly"?

[Prime95]

Originally Posted by chalsall

But, empirically, not enough users are selecting TF and P-1 work.

So why then are those who choose LL work burdened with lessor work? Why not let them work what they choose?

I think we have an overabundance of people selecting TF and not enough selecting P-1.

Users who choose LL-only are
1) given LL tests that do not need TF or P-1, but only if any are available.
2) otherwise, they are given the next best thing: an LL-test that requires a day or two of preliminary P-1 and/or TF work.

GIMPS has operated this way for the last decade.

Originally Posted by petrw1

An interpretation of http://www.mersenne.org/primenet/ might help me and others: If I look at the current LL wavefront 50-53M and columns near the end I see:

----------Available
Exponent  P-1  LL
50000000 5484 
51000000 4977 15668
52000000 17651 3274
53000000 16981 3723

Am I incorrectly interpreting this as telling me that in the 51M range for example there are 4977 exponents that need P-1 (and then potentially LL) and 15668 that are available for LL (implying P-1 is already done for these)?

There are 4977 exponents that need P-1 and then the last bit of TF and then LL. There are 15668 that are ready to go for an immediate LL test.

And secondly, I take your comment to imply that we STILL DO NOT have enough people doing P-1?? Not to suggest any priorities for you but I can say that one reason I have reduced my P-1 participation (and one other has commented the same on this forum) is that I have two PC's; both E6550 Dual Cores for which Prime95 crashes at least half of the instances that either of these PCs finds a factor whether in Stage 1 or 2.

P-1ers have made GREAT progress. I suspect LL testers will overtake P-1ers, but maybe I'm wrong. Perhaps someone has saved the last few months of the status page and can tell if the exponents assigned for LL in the 50-51M range is going up faster than the exponents available for LL in the 51-52M range.

Please try version 26.2. I may have found the source of the factor crash bug! I'd like to know if I haven't.

[davieddy]

http://www.youtube.com/watch?v=Tpvpyb-PXM0

PS did those badgers have TB, road killed or culled?

Let me know how they taste.

David

[petrw1]

Originally Posted by Prime95

There are 4977 exponents that need P-1 and then the last bit of TF and then LL. There are 15668 that are ready to go for an immediate LL test.

Earlier posts have suggested that there are NOT enough LL-only tests to go around; yet in the range I listed (50-53M) there are 22,500. Isn't that a lot?

Interesting that of the last 50 or so LL tests I was assigned (anywhere from 38-50M) about half go straight to LL; about half to P-1 followed by LL.
I have NOT YET received an LL assignment that did P-1; TF and then LL.

Please try version 26.2. I may have found the source of the factor crash bug! I'd like to know if I haven't.

Okay I will....in fact I just upgraded one of these PCs.

[ckdo]

Originally Posted by petrw1

----------Available
Exponent  P-1  LL
50000000 5484 
51000000 4977 15668
52000000 17651 3274
53000000 16981 3723

Am I incorrectly interpreting this as telling me that in the 51M range for example there are 4977 exponents that need P-1 (and then potentially LL) and 15668 that are available for LL (implying P-1 is already done for these)?

No.

Originally Posted by Prime95

There are 4977 exponents that need P-1 and then the last bit of TF and then LL. There are 15668 that are ready to go for an immediate LL test.

No, not quite. You're right (both of you, in a way) about the 15668. However, as far as the 4977 are concerned, these exponents have had TF up to at least P-1 depth (they may or may not require an addional bit after P-1), period.

Not all of these 4977 will still require a first-time LL test. They may have had one already. The 40M-50M range has 7000+ such exponents. Actually, if memory serves me right, if an exponent has been double checked (or tested twice due to a suspicious result) and neither of the testers did the P-1 step, it will still be included in the count.

I've done P-1 on some 55,000+ exponents below 40M which had been LL tested (and many of them, double checked) without doing the P-1 step first, so far. I am not currently planning to do the same in the 40M-50M range - the "LL done but no P-1 done" quota is much higher there (like 5% if memory serves me right). I simply don't have the resources to do it.

As far as the quota is concerned, we're probably going to see an increase there as well. Higher exponents will require more memory for P-1. Sure, people upgrade their hardware all the time, but who thinks about increasing the memory allocation when he copies the data over (or simply installs more memory)?

[Uncwilly]

Originally Posted by chalsall

I am simply arguing that those who wish and ask to do LL testing might reasonably expect to actually be assigned (and the client software do (and only do)) LL testing.

If they really, really care about that (which most users truly don't even know about the difference), then they can set the program to only do the LL's. But, if they do and the exponent has a factor that would have been found, they wasted their cycles.

[Prime95]

Originally Posted by petrw1

I have NOT YET received an LL assignment that did P-1; TF and then LL.

This is primarily due to aftereffects of the v4 PrimeNet server and decisions I made in transitioning to the v5 server.

The v4 server and v24 clients did TF all the way. In fact, existing v24 clients that haven't upgraded are still doing TF the old-fashioned way (do all TF before any P-1). Your LL clients have not had to do any TF because the v4 server had done all necessary TF up to about 50M.

The v25 P-1 clients did some P-1 work below 51M, but several months ago were moved to the 51M+ range so that they could get ahead of the LL wavefront. This is why it is fairly common that your LL assignments would first do P-1.

I know the server prefers to give out P-1 assignments that have not had the last TF bit done. What I don't remember is which it prefers if its choice is between a 51M exponent that has had all its TF done and a 52M exponent where there is another bit of TF to do.

[ATH]

Originally Posted by Prime95

Perhaps someone has saved the last few months of the status page and can tell if the exponents assigned for LL in the 50-51M range is going up faster than the exponents available for LL in the 51-52M range.

Date (YYYY-MM-DD)	Assigned+Completed LL 50-51M	Available LL 51-52M
2010-01-01			79				 7894
2010-02-01		       100		        	10387
2010-03-01		      2795		        	12066
2010-04-01		      9211		        	13726
2010-05-01		     13141		        	14282
2010-06-01		     19681		        	14759
2010-07-01		     19123		        	15348
2010-08-02		     16455		        	15531
2010-09-01		     14415		        	15639

Here is the monthly data 49-54M since january: LL+P-1stats.txt

t's somewhat hard to interpret since alot of things is going on. Around april there was alot of TF work 51-53M, which interferes with available P-1. From Aug to Sep "Assigned LL" 50-51M went down about 3000 while Completed LL only went up 900, rest seems to have become "Available P-1".

Let me know if you want more from the old summary reports.