Proposed DC assignment and recycle rules - Page 6

philmoore · 2014-01-28, 06:04

I've had some more thoughts on this, equally applicable to the LL assignment and recycle rules thread. First is that the poaching is often driven by the obvious milestones, so rather than create categories such as 1500-3000-12000 to work for all cases, we allow people such as George/Scott/Mike to define categories such as:

1) Extremely high priority assignments: to be assigned only to reliable participants who appear to be able to finish them within a few weeks.

2) High priority assignments: to be assigned only to reliable participants who appear able to finish them within a short period of time, say within 6 months, at which time they expire and are reassigned.

3) Medium priority assignments: to be assigned to participants who appear able to finish them within 1 year to a year and a half, maybe even 2 years with reports of adequate progress.

4) Low priority assignments: to be assigned to any participants, renewable indefinitely unless they eventually move into a higher priority. But perhaps even these should be expired after a few years if no progress is reported.

Currently, we would classify category 1 as any hypothetically expiring DC assignment below M43. Category 2 would be any DC assignment below M44, and maybe even below 10 million digits, and also any first time LL assignment below M48. Perhaps in a couple of months, exponents below M44 would be moved from category 2 to category 1. Category 3 would include most DC and LL assignments up to the leading edge, while category 4 would include any assignment considerably above the leading edge, including 100 million digit reservations.

The disadvantage of this method is that it requires someone to actually make a decision as to what the cutoff points should be, but I also see this as an advantage, as I don't think the automatic determination of these categories by some sort of 1500-3000-12000 rule can adequately accommodate the occasional pressure of "milestone achievement" which seems to drive poaching.

One last point: I don't mind if a poacher receives credit for completing work on an assignment for which progress was never made, but I like the idea of any participant who has been poached reporting any progress at all on a poached assignment to receive full credit with a message to that effect.

LaurV · 2014-01-28, 06:56

Grrr... When this thread started? I let you (all) alone for a week, and look what you are doing... How come you start interesting topics only when I am not around?

(some people know that I was in Germany last week)

Prime95 · 2014-02-01, 02:47

Updated web page for review:

http://www.mersenne.org/thresholds

I'll be tweaking it off and on, so don't be surprised if there are bugs

tha · 2014-02-01, 10:18

Looks very good.

The value 300 Mb for P-1 looks outdated to me. As the server has all the client information available how much would we win and loose if we would up that value to a higher level?

Xyzzy · 2014-02-01, 15:34

Quote:

The value 300 Mb for P-1 looks outdated to me. As the server has all the client information available how much would we win and loose if we would up that value to a higher level?

http://www.mersenneforum.org/showpos...5&postcount=10

Prime95 · 2014-02-01, 17:32

The new DC rules are now being enforced when getting an assignment. Let me know of any problems.

Implementing the new recycling rules is next. This isn't easy as I'm currently locked out of the server. Thanks, Microsoft.

cuBerBruce · 2014-02-01, 21:30

I would like some clarification about what "returned at least 2 results for each worker thread in the last 90 days" means.

First, does this mean Lucas-Lehmer test results (LL or DC), or results of any work type?

Second, what if a computer has been doing first LL tests and then gets a DC assignment. Will it be "penalized" for having been doing LL (which take longer) instead of DC?

It seems to me this requirement is a bit harsh for category 3 assignments. A core that can do a DC in 60 days could finish 3 assignments in the time that it is being allowed for just getting the assignments started (180 days). Yet, it would seem to be prevented from getting these assignments. This doesn't make much sense to me.

Overall, I agree with this effort to stop low-exponent assignments from getting held up for ridiculous amounts of time.

Prime95 · 2014-02-02, 02:18

Quote:

Originally Posted by cuBerBruce

I would like some clarification about what "returned at least 2 results for each worker thread in the last 90 days" means.

First, does this mean Lucas-Lehmer test results (LL or DC), or results of any work type?

Second, what if a computer has been doing first LL tests and then gets a DC assignment. Will it be "penalized" for having been doing LL (which take longer) instead of DC?

2 results means 2 LL results (LL or DC).

There is no "penalty" for having been doing LL. Yes, it is harder to complete 2 results in 90 days, but I think a Core2 computer (4 or 5 year old technology) could meet that requirement.

These rules are a first draft to see how well they work. The penalty for getting a category 4 vs. a category 3 is only 2 million (a 34M exponent instead of a 32M).

cuBerBruce · 2014-02-02, 04:00

Quote:

Originally Posted by Prime95

2 results means 2 LL results (LL or DC).

There is no "penalty" for having been doing LL.

NONSENSE.

I have a machine that started out getting DC assignments. It completed these in under 35 days each, easily meeting the 2 results every 90 days requirement. But for its last 3 assignments, Primenet decided to give it LL assignments instead of DC. The smallest of these took 88 days.

So YES, this machine definitely will be "penalized" under the new rules for it having been given LL assignments, assuming Primenet reverts to giving it DCs.

OK, the penalty will only be for the first new DC, assuming that I report that first new DC result before fetching more assignments.

Prime95 · 2014-02-02, 04:50

Quote:

Originally Posted by cuBerBruce

NONSENSE.

OK, I went back and checked. My 6 year old Core2 laptop did LLs in 2 months. So, the "penalty" would not apply to AVX machines but would apply to many Core 2 machines.

I could reduce the penalty by summing the exponents of recent LL results. This would give "number of reported LL iterations" the last 90 days. Advantage: makes it easier for an LL'ing machine to qualify for preferred DCs. Disadvantage: a bit harder to explain.

TheMawn · 2014-02-02, 08:21

Did the two-results-in-ninety-days thing have anything to do with the reliability of the computer? Or just it's short-term rolling average?

2014-02-01, 02:47	#58
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 10000001010111₂ Posts	Updated web page for review: http://www.mersenne.org/thresholds I'll be tweaking it off and on, so don't be surprised if there are bugs Last fiddled with by Prime95 on 2014-02-02 at 04:51 Reason: URL updated

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
PrimeNet Assignment Rules	S485122	PrimeNet	11	2021-05-20 14:54
Modifications to DC assignment rules	Prime95	PrimeNet	74	2017-01-18 18:36
Understanding assignment rules	Fred	PrimeNet	3	2016-05-19 13:40
Proposed LL assignment and recycle rules	Prime95	Data	156	2015-09-19 12:39
Proposed TF, P-1, ECM assignment and recycle rules	Prime95	Data	9	2014-02-27 23:52

2014-01-28, 06:04	#56
philmoore "Phil" Sep 2002 Tracktown, U.S.A. 1121₁₀ Posts	I've had some more thoughts on this, equally applicable to the LL assignment and recycle rules thread. First is that the poaching is often driven by the obvious milestones, so rather than create categories such as 1500-3000-12000 to work for all cases, we allow people such as George/Scott/Mike to define categories such as: 1) Extremely high priority assignments: to be assigned only to reliable participants who appear to be able to finish them within a few weeks. 2) High priority assignments: to be assigned only to reliable participants who appear able to finish them within a short period of time, say within 6 months, at which time they expire and are reassigned. 3) Medium priority assignments: to be assigned to participants who appear able to finish them within 1 year to a year and a half, maybe even 2 years with reports of adequate progress. 4) Low priority assignments: to be assigned to any participants, renewable indefinitely unless they eventually move into a higher priority. But perhaps even these should be expired after a few years if no progress is reported. Currently, we would classify category 1 as any hypothetically expiring DC assignment below M43. Category 2 would be any DC assignment below M44, and maybe even below 10 million digits, and also any first time LL assignment below M48. Perhaps in a couple of months, exponents below M44 would be moved from category 2 to category 1. Category 3 would include most DC and LL assignments up to the leading edge, while category 4 would include any assignment considerably above the leading edge, including 100 million digit reservations. The disadvantage of this method is that it requires someone to actually make a decision as to what the cutoff points should be, but I also see this as an advantage, as I don't think the automatic determination of these categories by some sort of 1500-3000-12000 rule can adequately accommodate the occasional pressure of "milestone achievement" which seems to drive poaching. One last point: I don't mind if a poacher receives credit for completing work on an assignment for which progress was never made, but I like the idea of any participant who has been poached reporting any progress at all on a poached assignment to receive full credit with a message to that effect.

2014-01-28, 06:56	#57
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 10100000110011₂ Posts	Grrr... When this thread started? I let you (all) alone for a week, and look what you are doing... How come you start interesting topics only when I am not around? (some people know that I was in Germany last week)

2014-02-01, 10:18	#59
tha Dec 2002 881 Posts	Looks very good. The value 300 Mb for P-1 looks outdated to me. As the server has all the client information available how much would we win and loose if we would up that value to a higher level?

2014-02-01, 17:32	#61
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 17·487 Posts	The new DC rules are now being enforced when getting an assignment. Let me know of any problems. Implementing the new recycling rules is next. This isn't easy as I'm currently locked out of the server. Thanks, Microsoft.

2014-02-01, 21:30	#62
cuBerBruce Aug 2012 Mass., USA 2·3·53 Posts	I would like some clarification about what "returned at least 2 results for each worker thread in the last 90 days" means. First, does this mean Lucas-Lehmer test results (LL or DC), or results of any work type? Second, what if a computer has been doing first LL tests and then gets a DC assignment. Will it be "penalized" for having been doing LL (which take longer) instead of DC? It seems to me this requirement is a bit harsh for category 3 assignments. A core that can do a DC in 60 days could finish 3 assignments in the time that it is being allowed for just getting the assignments started (180 days). Yet, it would seem to be prevented from getting these assignments. This doesn't make much sense to me. Overall, I agree with this effort to stop low-exponent assignments from getting held up for ridiculous amounts of time.

2014-02-02, 08:21	#66
TheMawn May 2013 East. Always East. 11×157 Posts	Did the two-results-in-ninety-days thing have anything to do with the reliability of the computer? Or just it's short-term rolling average?