This is not a complaint, but a POSITIVE situation and question.


Looks like BOINC project Predictor has reached a point where returned results need to be analyzed before more WUs are created/distributed. THAT IS GREAT! see post

I try to keep my machines attached to and available for 3 projects so this isn’t a problem for me at all.

QUESTION: **IF** this no-work time is expected to be a long time (many weeks, or a few months, or more), is there a way to increase the backoff past 1-day or KEEP it at once-per-day (until more WUs are received of course). Maybe even have the server provide a ‘check back in xxx hours’ to the client (i.e. check back in 3 days).


I think something like this would help keep people/computers attached to and available (‘allow new tasks’) for a project during a WU hiatus.


Details:

BOINC has one (or more) backoff schemes.
When a project does not have any work to send out, BOINC automatically starts to backoff queries to the project.

The wiki implies an exponential increase with an upper limit of 1 day for backoff.

With project Predictor, it never reaches a 1-day backoff.
And it doesn’t seem to STAY at 1-day (or any high backoff)

The backoff I’m seeing is 1min, 1min, 1min, 1min, 2:14, 3:40, 15:26, 04:15, 1:12:55, 2:40:53, 1min then “Fetching scheduler list”. This seems to be repeated.

The backoff seems to get reset more frequently than every day when “Fetching scheduler list”/” Master file download succeeded”

These constant, excess, queries/request don’t hurts me or my machines. I’m not concerned with bandwidth usage. I doubt it is a problem for the project servers.

---

There are a few limits for maximum backoff and they vary between versions of the core client. The most common one is 4 hours. Some older clients would get up to 2 weeks when the could not find the project.

The projects can send any delay they wish to the client. So if the repeated queries become a problem they can fix it.

---

BOINC WIKI

BOINCing since 2002/12/8