Newer milestone thread - Page 51

Brian-E · 2012-08-02, 07:38

Um, Dubslow, I'm a little confused.

Just a few days ago, in this same thread, you apparently wholeheartedly agree with patrik who points outs that there's no significance in the smallest exponent which has had no LL test (because many smaller ones will have had erroneous ones), but now you seem to advocate poaching in that area?

I must be missing some subtle humour here. Can someone explain the joke please?

chalsall · 2012-08-02, 15:16

Quote:

Originally Posted by Dubslow

So we really should wait for him?

I don't know who grew impatient, but this candidate was completed sometime in the last 12 hours....

NBtarheel_33 · 2012-08-02, 15:24

August 2, 2012. All exponents below 40 million have been tested at least once.

Dubslow · 2012-08-02, 16:49

Quote:

Originally Posted by Brian-E

Um, Dubslow, I'm a little confused.

Just a few days ago, in this same thread, you apparently wholeheartedly agree with patrik who points outs that there's no significance in the smallest exponent which has had no LL test (because many smaller ones will have had erroneous ones), but now you seem to advocate poaching in that area?

I wasn't entirely clear in my response to his post. I agree that more effort should be put into DC, but that doesn't mean we shouldn't try and shrink the LL wave a little. If you reread my post, I said "(excepting <43M tests)", for which a 39M expo certainly counts.

And as pointed out, that problem is now moot, having been poached by someone else while we were talking about it.

Brian-E · 2012-08-03, 00:45

So you disapprove of poaching ...

Quote:

Originally Posted by Dubslow

It's not just discouragement of the slow-but-steadies, but it actually reduces GIMPS' throughput. The sniper could have just let the others finish their job and been doing other, unassigned expos; those other unassigned expos would be done now, and the expos the sniper poached will done soon anyways, except without the other ones done now.

... except when the exponent is <43M and the first LL test has not completed, because we need to shrink the LL wave a little.

Interesting.

Dubslow · 2012-08-03, 01:44

...except when the exponent is so far behind the wave, not completed, and has little hope of ever being completed (by its assignee). If the due date was within 2012 (or within 365 days, I'm not really sure what my "border" is), I wouldn't have said a thing. 2014 was too long in my book, and as pointed out, it probably won't be a good test anyways.

petrw1 · 2012-08-03, 04:39

Quote:

Originally Posted by LaurV

Too late! Poached already. We were talking, he was working**...

No. we (Me and DB) were talking about the 38M one that is suspect.

Dubslow · 2012-08-03, 04:50

Quote:

Originally Posted by petrw1

No. we (Me and DB) were talking about the 38M one that is suspect.

Whoops

I was thinking of the 39M (that has been poached). (And DB confused me for a minute

)

Brian-E · 2012-08-03, 09:40

Quote:

Originally Posted by Dubslow

...except when the exponent is so far behind the wave, not completed, and has little hope of ever being completed (by its assignee). If the due date was within 2012 (or within 365 days, I'm not really sure what my "border" is), I wouldn't have said a thing. 2014 was too long in my book, and as pointed out, it probably won't be a good test anyways.

There are always a handful of jobs in progress which are a long way behind the wave. One of them will always be the one with currently the lowest exponent. The situation at the moment is perfectly normal.
Both your statements that this job "has little hope of ever being completed" and "probably won't be a good test anyways" are pure speculation. Regarding the second, I would be interested to see any evidence that long-running tests are usually incorrect. I wonder if there is even any evidence that they are generally less reliable than ones which take a week or two.
PrimeNet operates a good working system whereby inactive jobs are assumed dead and given to someone else. This can also happen with jobs which are proceeding so slowly that they are holding the general progress up. There is no need to pre-empt this system.
Poaching happens. Stragglers have their work pinched by impatient busy-bodies who don't respect the ideals of the project. But publicly advocating this behaviour is going too far in my opinion.

Dubslow · 2012-08-03, 09:52

Quote:

Originally Posted by Brian-E

There are always a handful of jobs in progress which are a long way behind the wave. One of them will always be the one with currently the lowest exponent. The situation at the moment is perfectly normal.

Why should it be normal? Just because that's how it is doesn't mean that's how it needs to be. GPU to 72 was created with the explicit goal to shrink the wave.

Quote:

Originally Posted by Brian-E

Both your statements that this job "has little hope of ever being completed" and "probably won't be a good test anyways" are pure speculation. Regarding the second, I would be interested to see any evidence that long-running tests are usually incorrect. I wonder if there is even any evidence that they are generally less reliable than ones which take a week or two.

http://www.mersenneforum.org/showthread.php?t=16499

Quote:

Originally Posted by Brian-E

PrimeNet operates a good working system whereby inactive jobs are assumed dead and given to someone else. This can also happen with jobs which are proceeding so slowly that they are holding the general progress up. There is no need to pre-empt this system.

PrimeNet's assignment rules have always been under suspicion by many forum goers, with many apparent examples of it not expiring assignments when it should.
http://www.mersenneforum.org/showthread.php?t=16617 -- post 5
Many have advocated a complete rewrite of PrimeNet's assignment code, including a more-than-two-tiered assignment threshold system where ANONs would get exponents ahead of the wavefront, etc. (I don't really feel like going post hunting though.)

Quote:

Originally Posted by Brian-E

Poaching happens. Stragglers have their work pinched by impatient busy-bodies who don't respect the ideals of the project. But publicly advocating this behaviour is going too far in my opinion.

In general, I agree, but I've already stated what I thought to be exceptional circumstances. If I wasn't worried about the ideals, I would have begged forgiveness rather than ask permission like I did.

ATH · 2012-08-03, 13:06

I refer again to George own idea for new recycling rules from 2 years ago:
http://www.mersenneforum.org/showpos...&postcount=443

I don't know why these are not working or not implemented, but when exponents are far above those limits, I have no problem with "poaching" them, unless they are moving and will be done in say 6months or less.

I already started 39286963 but Signals beat me to it, I think I'll finish it as a doublecheck.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Newer X64 build needed	Googulator	Msieve	73	2020-08-30 07:47
Performance of cuda-ecm on newer hardware?	fivemack	GMP-ECM	14	2015-02-12 20:10
Cause this don't belong in the milestone thread	bcp19	Data	30	2012-09-08 15:09
Newer msieves are slow on Core i7	mklasson	Msieve	9	2009-02-18 12:58
Use of large memory pages possible with newer linux kernels	Dresdenboy	Software	3	2003-12-08 14:47

2012-08-02, 07:38	#551
Brian-E "Brian" Jul 2007 The Netherlands 110011000101₂ Posts	Um, Dubslow, I'm a little confused. Just a few days ago, in this same thread, you apparently wholeheartedly agree with patrik who points outs that there's no significance in the smallest exponent which has had no LL test (because many smaller ones will have had erroneous ones), but now you seem to advocate poaching in that area? I must be missing some subtle humour here. Can someone explain the joke please?

2012-08-02, 15:24	#553
NBtarheel_33 "Nathan" Jul 2008 Maryland, USA 5×223 Posts	August 2, 2012. All exponents below 40 million have been tested at least once.

2012-08-03, 01:44	#556
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3·29·83 Posts	...except when the exponent is so far behind the wave, not completed, and has little hope of ever being completed (by its assignee). If the due date was within 2012 (or within 365 days, I'm not really sure what my "border" is), I wouldn't have said a thing. 2014 was too long in my book, and as pointed out, it probably won't be a good test anyways.

2012-08-03, 13:06	#561
ATH Einyen Dec 2003 Denmark 2×1,579 Posts	I refer again to George own idea for new recycling rules from 2 years ago: http://www.mersenneforum.org/showpos...&postcount=443 I don't know why these are not working or not implemented, but when exponents are far above those limits, I have no problem with "poaching" them, unless they are moving and will be done in say 6months or less. I already started 39286963 but Signals beat me to it, I think I'll finish it as a doublecheck.