[QUOTE=Prime95;427741]I think you are overthinking / fine-tuning this too much.

I'd suggest something simple such as either
1) Any machine that has contributed more than X GHz-days in the last N days is upgraded to cat 2 assignments.
or 2) Nightly sort cpus by GHz-days produced in the last N days and the top Y CPUs are automatically upgraded to category 2.

The two are similar, but the advantage of the second system is it auto-adjusts over time.

The rather minor downside to auto-cat-2 assignment upgrades is that a user will have only 150 days to complete an assignment where he may have expected 270 days.[/QUOTE]

What about using the existing "Top LL Producers" report to group users into percentiles based on work done (in GHz-days) in the last year, setting a minimum of say 1,000 GHz-days/year of LL testing for consideration? Anyone that is a new user or that has contributed less than 1,000 GHz-days of LL testing over the last year is assigned Cat 4 exponents (with an appropriate expiration date), while the Cat 1-3 exponents are assigned by percentile: the lowest [I]x[/I] percent of the Cat 1-3 exponents go to the top [I]x[/I] percent of contributors (based on the work done in the last year).

The mechanism could work as follows: when a user requests an LL assignment, PrimeNet calculates the user's rank, finds an exponent at or near that user's percentile level, and assigns it. If the user's rank is too low, the user is new, or the user is anonymous, a Cat 4 exponent is automatically assigned.

Such a scheme might even eliminate the need for expiration of exponents other than the Cat 4s. Right now, the 1,000 GHz-days/year cutoff for LL testers is at Rank # 1,239 who has completed five LL tests in the past year. This means that the worst case scenario would be a Cat 3 exponent requiring ~73 days for completion. The other end of the spectrum is curtisc who would likely turn in the plum Cat 1s within a day or two.

One consideration that would have to be made is how the LL rankings might fluctuate over time. How often might it occur that today's Rank # 400 is next month's Rank # 2,000? I imagine that this is perhaps not so much a problem in the top 100-200 users; below that point, things could get a little more muddled.

The other consideration is that this could be seen as punishing users wishing to remain anonymous. One approach to this could be to issue generic serial IDs (such as the S-series accounts in PrimeNet v4) to any user not choosing a user name upon sign-up, or simply adding a stipulation that users remaining completely anonymous may only receive Cat 4 assignments. (Considering that all of GIMPS' prime discoveries have been made on what would have been Cat 3 or Cat 4 assignments, I am not so sure that this is so onerous of a requirement!)

[QUOTE=NBtarheel_33;427913]What about using the existing "Top LL Producers" report to group users into percentiles based on work done (in GHz-days) in the last year, setting a minimum of say 1,000 GHz-days/year of LL testing for consideration?[/QUOTE]

1000 GH-day/year can be achieved by 1 computer producing 1000 GH-day/year, or 1000 computers producing 1GH-day/year. Clearly, the latter cannot be given any Cat 1 assignments.

Point being, it is the individual computer's performance (time to complete an exponent) that matters. Aggregate performance of a user is not relevant.

[QUOTE=axn;427914]1000 GH-day/year can be achieved by 1 computer producing 1000 GH-day/year, or 1000 computers producing 1GH-day/year. Clearly, the latter cannot be given any Cat 1 assignments.

Point being, it is the individual computer's performance (time to complete an exponent) that matters. Aggregate performance of a user is not relevant.[/QUOTE]

How would a user with 1000 computers each producing 1 GHz-day/year possibly complete a single LL test within a year, let alone get onto the LL leaderboard?

[QUOTE=NBtarheel_33;427915]How would a user with 1000 computers each producing 1 GHz-day/year possibly complete a single LL test within a year, let alone get onto the LL leaderboard?[/QUOTE]

Whoosh (as the point misses you).

Fine. One 4-core computer completing 4 LL-tests in a year (1LL test/core/year). Do you want milestones to be held up for a year?

[QUOTE=axn;427916]Whoosh (as the point misses you).

Fine. One 4-core computer completing 4 LL-tests in a year (1LL test/core/year). Do you want milestones to be held up for a year?[/QUOTE]

Minimum 1,000 GHz-days/year. 4 LL tests in a year would be a minimum of 250 GHz-days average for each one, i.e. tests up around 80 million. That won't endanger any milestones anytime soon.

Not to mention that even if such a user were on the first-LL leaderboard and eligible to get a Cat 3 assignment, it is quite possible now for a Cat 3 to take a year or more to be processed. The present minimum completion time is 270 days, not including assignment churn.

[QUOTE=NBtarheel_33;427917]Minimum 1,000 GHz-days/year. 4 LL tests in a year would be a minimum of 250 GHz-days average for each one, i.e. tests up around 80 million. That won't endanger any milestones anytime soon.

Not to mention that even if such a user were on the first-LL leaderboard and eligible to get a Cat 3 assignment, it is quite possible now for a Cat 3 to take a year or more to be processed. The present minimum completion time is 270 days, not including assignment churn.[/QUOTE]

Sigh...

Imagine the same user with 20 machines, which is 20,000 GHz-day/year. Not a single expo will be crunched in less than a year, but the user will be in top 100, eligible for the Cat 1, by your proposal (whatever cutoff you propose, just imagine that many more computers to make them qualify for Cat 1). Now what?

EDIT:- Or consider this. The user has a combination of fast and slow computers. User is eligible for cat 1, but the slow computer can't process them fast enough. How will your scheme work, without accounting for individual computer productivity?

NBtarheel_33, great, thank you.

Yep, I'll hang in the last boarding zone. Women and children can have all the Cat 1 assignments.

[QUOTE=axn;427921]Sigh...

Imagine the same user with 20 machines, which is 20,000 GHz-day/year. Not a single expo will be crunched in less than a year, but the user will be in top 100, eligible for the Cat 1, by your proposal (whatever cutoff you propose, just imagine that many more computers to make them qualify for Cat 1). Now what?

EDIT:- Or consider this. The user has a combination of fast and slow computers. User is eligible for cat 1, but the slow computer can't process them fast enough. How will your scheme work, without accounting for individual computer productivity?[/QUOTE]

What if we set a minimum for GHz-days production *and* number of completed LLs?

[QUOTE=axn;427921]Or consider this. The user has a combination of fast and slow computers. User is eligible for cat 1, but the slow computer can't process them fast enough. How will your scheme work, without accounting for individual computer productivity?[/QUOTE]

For what it's worth, the current setup and anything else going forward already considers per-CPU performance, not just the user account as a whole.

That's also one reason why manual assignments aren't eligible for cat 1 or 2, because there's no way to ascertain which machine will be doing the work and how reliable it is.

[QUOTE=NBtarheel_33;427972]What if we set a minimum for GHz-days production *and* number of completed LLs?[/QUOTE]

Maybe, but consider a hypothetical where some bloke just finished a 100M digit exponent that took a year+ to run. After all that time, he's thinking "whew, maybe I'll do some small cat 1 stuff for a change".

Their ghz-days will be pretty high after turning in a big assignment like that, but it will only show as 1 result in the past XX days.

Yet I'd argue this person has shown a certain commitment to follow through by sticking with a large test for so long.

Anyway, just something to keep in mind... which is why I'm not totally married to the idea of "X results in the past Y days", but more about their actual throughput.

Someone could churn through maybe 4 smaller double-checks in the past 90 days on a slower system, but if they wanted to do a first-time LL test they could maybe only muster a couple in that same time. Especially if they want to have 4 running at once... with small DC assignments that may have been fine, but with larger FFT first time checks, their system will suffer greatly and probably won't finish anything for over a year. It happens. :smile:

This is based upon the rate of change of the P90 years on the classic status page:

We are now in new territory. Never before[SUP]*[/SUP] has the estimated date of completion been so soon. The current outlook is for all the first time LL's below 79.3 million to be done by October 9th of 2017. Using the same measuring scheme, in late 2005, the estimated completion date was end of August 2039. It had been Nov 2050 in early 2003.
(*There was a period in 1999-2000, when there was so much of the early factoring work being done that the numbers were skewed. About 200,000 exponents were being factored out each month, from October 1999 through March 2000. By Sept 2000 the estimate was 2023 and climbing.)

Further if some of the recent trends keep up, we may see the completion (except stragglers) by June 2017. There has been a huge up tick in completed DC's since the announcement of M49. If a reasonable number of theses CPU's stick around and get promoted to first time LL's that would speed things along.

The dark blue line on the graph represents a rough moving average. The green lines are drawn in along recent peaks and bottoms. The other 3 lines are generated trend lines with projections.