I run off-line, except for occasional connects to the server. My cache size is currently specified as 6 days.

On Mar 12 I downloaded work, all of which would expire on Mar 26. On Mar 14 (running offline) my system had about 100 hours of work left in its "ready queue" for the project. The client went into 'earliest deadline first' scheduling mode.

Someone had posted: 'if a project deadline is less than double the cache size, the box will only run in EDF mode'. That's what seems to have happened here. (The hour on Mar 14, plus 12 days, amounted to more than the deadline.)

BUT the processing % currently allocated to the project would have completed the 100 hours still-to-do BEFORE Mar 20 !! I would have preferred the client's logic to be: "calculate when the last WU is estimated to complete; then (assuming the next report is not made for 6 days) will the result be reported in time?". In my case, the 100 hours still-to-do would have completed (at the project's allocated processing %) on Mar 19 !! EDF mode was *not* needed.


Let me suggest that the BOINC client have *two* specifiable values - one to limit the number of days of READY work to be kept at the client <ought to be __variable__ for different projects>, and the other to limit the number of days that COMPLETED work might be kept at the client <same for every project>. The first would govern the amount of work downloaded (I prefer to keep a sizeable queue, so as not to run out of work in case the server is down). The second would govern when EDF mode is entered (similar to today's "connect interval" - first the value would be subtracted from the deadline, then the threshold for EDF would be calculated backwards assuming 100% processing for the project).
.