View Single Post
Old 2021-07-09, 14:26   #2210
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

3×52×79 Posts
Default

Quote:
Originally Posted by drkirkby View Post
The server is still issuing PRP test assignments with an entry in worktodo.txt that indicates 2 tests will be saved.
...

If my understanding is correct, this would only seem appropriate if the computer doesn’t have the resources to generate and upload the proof file, or someone has configured their software not to upload a proof. I am using a recent version of mprime (IIRC 30.6b4), and are generating and uploading proof files.
Yes, this is known behavior, and may be by design.
The value 2 is an approximation. There are several cases.
  • PRP with proof: needs a cert. Proof power typically 7 or higher, and failure rate is small but nonzero. So 1.00n. We hope this is the most common case, for efficiency and reliability.
  • PRP no proof; will need a PRP DC. PRP TC rate is in the PPM range. So 2.0000n. (There are older OSes/systems that can't run the proof-capable prime95 versions. Mlucas doesn't have proof-capable PRP released yet. Systems might lack the necessary disk space or be configured in prime95/mprime with too little space usage allowed.)
  • LL; will need an LL DC. (Yes, LL first test are no longer being issued by PrimeNet API, but there are some folks that may be converting PRP assignments to LL. Or obtaining them manually. Or running them without assignments. Or reporting in already-assigned work after a long slow run. Maybe run on an old slow GPU or CPU/OS that can't run Gpuowl or PRP-capable mprime/prime95 respectively.) Error rate 1-2% per run for wavefront exponents means ~2.02-2.04 per exponent. (And much higher error rate with higher exponent means higher numbers of tests saved. The sparse data for 100Mdigit and higher is consistent with error rate about proportional to run time, ~p2.1, so at 100Mdigit, ~2.5; 300Mdigit, ~17.)
For PrimeNet API connected mprime/prime95, the server may have the necessary info for the PRP first test assigned system to determine PRP proof capability and so choose and specify 1 not 2 tests saved. If there's an error of any sort in the first PRP test and proof generation that's unrecoverable, it can't predict ahead of time that a second test will or won't be needed, or whether the second will be performed by a PRP proof capable installation. I think that's covered by the 1.00n.

Last fiddled with by kriesel on 2021-07-09 at 14:38
kriesel is offline   Reply With Quote