How much is enough?

[kriesel]

https://www.mersenne.org/thresholds/?setting=1
cat 1 dc: Computer must have enough LL and DC GHz-days over the last 120 days to indicate the assignment will be completed in 30 days (including the setting for "days of work to queue").
Computer must have returned at least 5 results in the last 120 days.

cat 2 dc: Computer must have enough LL and DC GHz-days over the last 120 days to indicate the assignment will be completed in 60 days.
Computer must have returned at least 3 results in the last 120 days.

Cat 3 dc: Computer must have enough LL and DC GHz-days over the last 120 days to indicate the assignment will be completed in 90 days.

I have a newly deployed system that has completed dozens of LLDC, within the past 30 days, with none bad or expired or suspect.
It's getting only Cat4 DC assignments. Its workers completed cat4s in 17.4-18.5 elapsed days, and can if operating continuously complete them in ~8 days each.
System setting for days of work to queue is already <= 3 days.
The account is already set to get the smallest exponents.

How many quickly completed tests are enough?
What's the time threshold for "quickly"?
Must some be first-time tests?
What's stopping this system from getting lower cat DC that will help DC up to Mp48* sooner?

[petrw1]

Did you Change the Setting here to get the lowest exponents?

https://www.mersenne.org/thresholds/?setting=1

[kriesel]

Yes.

Originally Posted by kriesel

https://www.mersenne.org/thresholds/?setting=1
...
The account is already set to get the smallest exponents.

[tuckerkao]

I was wondering about the same thing. If I have earned over 2,000 total GHz days and finished at least 5 successful PRPs within the past 4 months, then I can be qualified for any category 1 assignments.

[retina]

Originally Posted by kriesel

... at least 5 results in the last 120 days.

... at least 3 results in the last 120 days.

... the last 120 days ...

I have a newly deployed system that has completed dozens of LLDC, within the past 30 days...

There's your problem.

Be more patient.

[kriesel]

|----30d----|----30d----|----30d----|----30d----|
|need at least 5 good results in this span |
system in question___________: |---17--|
Substantially exceeded the 5 good results within the past 120 days requirement as stated, very quickly. Soon it will be 30 completed in < 30 days.
If the system needs to also have a 120 day history with the project, the requirements ought to say so, clearly. No part of https://www.mersenne.org/thresholds/ does.
"in the last 120 days". The most recent 30 days or less is always in the most recent 120 days.
It could be completing ~2 LL DC daily toward showing MP48* the 48th Mersenne prime, for ~112 days, that the 120-day-history-required theory would prohibit. The same issue arises with a second box that arrived recently, so ~4/day.
Total elapsed time of the first assignments was ~18 days, well within Cat1's 30 day requirement. Raw compute rate indicates ~8 days if up continuously, which appears to meet even Cat 0 requirements.

[S485122]

I installed my current machine in May, I first got four CAT 4, followed by a CAT 3 and two CAT 2 before getting CAT1 assignments. The category shifting took 4 days from new to sufficiently trusted to get CAT1 exponents.

Or the rules have changed or there is something else at play. What is the "Reliability" and "Confidence" of those computers ? (After dozens of good results and no expiring of assignments they should be at 1,00 and 10,0 respectively.)

Which software are you using ? (I suppose mprime or Prime95, if it is another program it would be fed by manual assignments. I also suppose you are running the latest version.)

Jacob

[kriesel]

Win10 x64 Pro build 1909; prime95 V30.3b6 PrimeNet API connected.
Reliability previously indicated as 0.99, now 1.00; confidence 10.0
37 work units completed, 30 currently assigned.
2323 GHzD credited.
Rolling average 1173 (117%).
Cat4 followed by cat4. We'll see what round 3 brings. Maybe it just doesn't like the high worker count (that came from the default 4 cores/worker prime95 default self-setup, and is pretty good compromise between max throughput and acceptable latency) I could reduce to somewhat fewer workers without too much of a hit on throughput.
One possible difference is I have put a few manual assignments on, one per worker, and they are larger exponents so will have long run times.

Thanks for a useful response, S485122.

[S485122]

Originally Posted by kriesel

...
Maybe it just doesn't like the high worker count (that came from the default 4 cores/worker prime95 default self-setup, and is pretty good compromise between max throughput and acceptable latency).

That might be it, I think the number of work units done is computed by core.

Jacob

[Prime95]

Jacob is on the right track. The calculation is per-worker. If your machine has 16 workers, then your 17 results are barely over 1 per worker in the last 120 days.

I suggest grabbing a bunch of manual DCs and populating your worktodo.txt.

[kriesel]

Originally Posted by Prime95

Jacob is on the right track. The calculation is per-worker. If your machine has 16 workers, then your 17 results are barely over 1 per worker in the last 120 days.

I suggest grabbing a bunch of manual DCs and populating your worktodo.txt.

Thanks.
Sounds to me like what you're saying is

Computer must have returned at least 5 results in the last 120 days.

actually means (and if so it would be helpful to clarify it to)

Computer must have returned at least 5 results in the last 120 days per prime95 worker.

I could shrink the number of workers and hit that very soon. Unless it keeps track of which result was from or which assignment was to which worker number; then it would take weeks longer. The PrimeNet API and the setting to do LL DC on it will keep feeding it the fairly quick LL DC. Condensing the existing worktodo to put already-assigned work (current plus next, times lots of worker count) onto 4 workers should do the trick, or very close to it.