RDCF problem when leaving a project and returning to it

[Ageless]

Result Duration Correction Factor will go down to the lowest it can on a project, but when the project puts a new application in, the RDCF number is useless. It will ask too much work. The same happens when someone leaves a project when the results are at a low runtime and returns later, when the results are at a high run time.

I know the RDCF is to teach BOINC how long rsults run and thus how much work it can ask to top off the queue for that project, but this goes awry on multiple projects. Perhaps that when people detach from or don't ask for new work 4(+) weeks from the project, that their RDCF number is reset to 1 automatically?

It would help take down some confusion.

[Nicolas]

My opinions:

• DCF should slowly go to 1.0 while no work is crunched; similar to the way exponentially-decayed RAC works server-side, only that it would go to 1 and not to 0. Note that RAC is linear and DCF is a multiplier, so the math would probably be different from the one used for RAC decay.
• DCF should be tracked per application, not only per project.

[davea]

A reasonable request, but I'm not going to implement it right now. DCF is a kludge to compensate for bad FLOP estimates by projects. I don't want to make the kludge even more complicated.

BOINC provides a tool (job_times.php) for finding the precise FLOP count distribution; projects should use it. If you're attached to a project that has bad estimates, complain to them.

There may be an additional factor due to host characteristics, but this should be more or less constant across applications.