Reliability, Confidence and exponent assignments
 

L.S.,

I have a pc with Reliability = 0.98 and Confidence = 10.0

On the assignments tresholds page it says:

Thresholds for double-check assignments
Force trial factoring if CPU reliability less than 0.5
and CPU confidence level is greater than or equal to 2.0
[B]Preferred exponents are below 31,068,721[/B]
Minimum Pentium 4 equivalent speed to get preferred exponent 1000 MHz
Required CPU reliability for preferred exponent 0.9
Required CPU confidence level for preferred exponent 2.0

The machine got assigned 4 exponents each about 32,000,000. I would expect the line [B]Preferred exponents are below[/B] to be higher than the assigned exponents. Anyone a clue?

I think the lower exponents may have been cleared. I have not checked PrimeNet, but GPU72 is showing DC assignments above 33M.
[url]http://www.gpu72.com/reports/available/[/url]

[URL="http://www.mersenne.org/report_milestones/"]Judging by the milestones page[/URL], everything has been DCd to [SIZE=2]26,414,867.
There are a lot of 'hogged" exponents after that one, which were reserved a year ago but never expired or got released.
A lot of them are hogged by the same guy.
Use [URL="http://www.mersenne.org/report_LL/?exp_lo=26414867&exp_hi=33000000&exp_date=&user_only=0&user_id=&exdchk=1&exbad=1&exfactor=1&dispdate=1&B1=Get+LL+data"]this link[/URL] to help push the lower milestone bound & DC hogged exponents.
[/SIZE]

[QUOTE=Karl M Johnson;356075][SIZE=2]
Use [URL="http://www.mersenne.org/report_LL/?exp_lo=26414867&exp_hi=33000000&exp_date=&user_only=0&user_id=&exdchk=1&exbad=1&exfactor=1&dispdate=1&B1=Get+LL+data"]this link[/URL] to help push the lower milestone bound & DC hogged exponents.
[/SIZE][/QUOTE]
Well, thankyou very much. That brings up, amongst several hundred others, the DC which I am currently working on. I was assigned it by GPUto72 two months ago, started work on it within a few weeks (as soon as my previous one reported) and now have it 83% complete. It is my sole work on my only machine, work which I have enthusiatically and continuously used for GIMPS work (and everyday computing) for more than 6 years. I am not "hogging" the work.

[QUOTE=Brian-E;356080]I am not "hogging" the work.[/QUOTE]
Unless you're Carsten Kossendey, of course you are not.

[QUOTE=Karl M Johnson;356110][...]of course you are not.[/QUOTE]
Right, but I guess my petulant, personally wounded tone was really designed just to express distaste for what you were suggesting while not really wanting to start yet another tedious row about poaching involving all the usual suspects.:max:
Agreeing to disagree is probably the only solution, but I still get off on expressing that disagreement.:smile:

Karl, you are new here. First is not smart to publicly advocate poaching. I am not a saint, and occasionally poach, too, but I don't push others to do it. Everybody does what he likes, as long as that is not bothering other people. Second, [URL="http://www.gpu72.com/reports/worker/89edb735f68ff3faa688634ac776d5f8/"]that guy is not hoarding exponents[/URL]. He is a long-time contributor, and one of the few guys who is doing TF in the DC range (and TF for exponents which had two LL tests done already - well people are still hunting for factors of cleared exponents, some of us feel happy just to find factors!).

Yes, those 1100 exponents are TF assignments, not DC assignments. But they may be seen as DC because from the point of view of PrimeNet, there was enough TF done for them - mind that PrimeNet is not aware of the invention of the GPU, nor about GPU factoring.

Generally would be no harm if you DC a couple of those exponents. The guy will release them when he finish his TF, and your work and his work is different, you won't step on his toes, nor him on yours.

In case I understood you wrong, and you want to do TF, be our guest, there are about 20 thousands exponents free there! CK is not bottlenecking anything.

Edit: BTW, Carsten is active user here on the forum, you can send him an ICQ (his id is public, if you go to his profile).

The only thing I can come up with myself is that the value of 31,068,721 is independent of the amount of available reliable clients. So when the server runs out of preferred exponents it continues to dole out regular assignments to all clients. What remains a mystery then is what serves as a base for computing the 31,068,721 value.

[QUOTE=tha;356465]The only thing I can come up with myself is that the value of 31,068,721 is independent of the amount of available reliable clients. So when the server runs out of preferred exponents it continues to dole out regular assignments to all clients. What remains a mystery then is what serves as a base for computing the 31,068,721 value.[/QUOTE]

Every night the limit changes. The smallest 1000 (I think) exponents that are not double-checked are considered "preferred".

[QUOTE=Prime95;356466]Every night the limit changes. The smallest 1000 (I think) exponents that are not double-checked are considered "preferred".[/QUOTE]

Smallest [B]3000[/B], IIRC. Yep, see [URL=http://www.mersenneforum.org/showpost.php?p=308213&postcount=22]here[/URL].

My guessed interpretation of the situation, after reading the above contributions, is the following:
[LIST][*]No machine which does not meet the speed, reliability and confidence requirements will be directly assigned "preferred" work by PrimeNet. But that does not say that a machine which does meet the requirements will necessarily get a preferred exponent.[*]We quite possibly have the ongoing situation that all preferred exponents have been assigned. In this case, the only way a client can get preferred work directly from PrimeNet is if is strikes lucky and requests it at a moment when a preferred assignment has just expired or been given back by its assignee.[*]Some of the preferred work is assigned to GPU72. So anyone who particularly wants this work can join GPU72 and have a reasonable chance of getting it that way.[/LIST]