Would it be sensible to upload PRP tests when 50% complete?

[drkirkby]

Looking at the exponents that are going to expire very shortly, I see a number have done a fairly large percentage of a PRP test, then stopped. For example,

This expires in 2 days
https://www.mersenne.org/M103499453
has completed 65% of a PRP test, but has not connected to the server since 12th May.

This has completed 82% of a PRP test, but will almost certainly expire
https://www.mersenne.org/M103610623

Given it is possible to stop and restart a PRP test from files on ones disk, it must be technically possible for someone to complete a PRP test that someone else has already done 50% of the work, if the 50% completed PRP test can be downloaded from the server. .

I can see obvious problems with this approach
1) People would be uploading files that might be wasted, since they would complete the test anyway.
2) The server would have to store the files.

If this was implement, two things could be done to reduce the impact on the server, and the amount of data people upload.
a) Ignore tests allocated in categories 0 and 1, as those are only assigned to people who don't let exponents expire.
b) Ignore very large exponents. Many people start the 100 M digit prime tests, then give up. Having completed one myself, I can understand why that happens.

Just a thought.

[kriesel]

Originally Posted by drkirkby

Looking at the exponents that are going to expire very shortly, I see a number have done a fairly large percentage of a PRP test, then stopped. ...

Given it is possible to stop and restart a PRP test from files on ones disk, it must be technically possible for someone to complete a PRP test that someone else has already done 50% of the work, if the 50% completed PRP test can be downloaded from the server. .

I can see obvious problems with this approach
1) People would be uploading files that might be wasted, since they would complete the test anyway.
2) The server would have to store the files.

If this was implement, two things could be done to reduce the impact on the server, and the amount of data people upload.
a) Ignore tests allocated in categories 0 and 1, as those are only assigned to people who don't let exponents expire.
b) Ignore very large exponents. Many people start the 100 M digit prime tests, then give up. Having completed one myself, I can understand why that happens.

Just a thought.

This has been discussed ad nauseum in the past, and a reference info entry written two years ago. Start reading there, or here.

[Uncwilly]

Very quick summary:

If you are willing to donate infinite bandwidth and storage to GIMPS and bandwidth to the users, sure, it is possible. Sensible? Effectively not with out those.

[kriesel]

Plus donate some design, documentation, and programming time, to modify and test the server, PrimeNet API, major applications, and client management software for GPU apps.