Often I abort a task which won't complete by the given deadline.
Q1. Why doesn't the system auto-delete these and switch to a valid task instead of wasting my computer time?
Q2. Does the system use GMT? I'm in Australia, so it could be important.

The deadline is represented in the BOINC Manager with the local [your] time.

Autodelete at least at WCG is only when the task is found to be redundant, when the server tells the host, that has an unreturned copy, it's no longer required, logged as 'server aborted'. Other than that, projects have their own policies, nothing the client forces.

How you get tasks that are not meeting completion deadline is a source of discussion. Reducing the buffer size is the easiest fix. Because of idiosyncrasies in task scheduling, recommend to never cache more than half the shortest deadline your active projects have. E.g. if the shortest deadline is 3.50 days, set the buffer to 1.75 days at most. Then processing is most always in FIFO order. Set a higher buffer and the client often will revert to Earliest Deadline First prioritization, high priority processing, which causes tasks that were running to be pre-empted.

---

Coelum Non Animum Mutant, Qui Trans Mare Currunt

The task deadline shown on a project's web page will always be in UTC (which is effectively the same as GMT). The deadline shown in BOINC Manager will be in the local time zone as set for the computer's clock - which you would normally set to be the local time zone for wherever you're using the computer.

When a task is issued to your computer, the deadline is always a fixed number of hours or days from the time it's allocated. So it doesn't matter where you are in the world - Australians get just the same length of time from from allocation to deadline as residents of any other country.

If a task is still unstarted on your computer when the deadline is reached, then your local BOINC client will abort it. But if the task has started, and run for even one second before being suspended, it won't be aborted automatically.

"If a task is still unstarted on your computer when the deadline is reached, then your local BOINC client will abort it."

As of which version is this a client policy to abort unstarted+overdue tasks? There are the red line messages if running overdue tasks to say there is a chance of not getting credit, but never observed the client having initiative... it's at least with WCG the 'server aborted' notice sent, and only if any of the two [the extra copy or the original] has returned with success.

To add: Seen on part time clients and dual boots, where seriously overdue tasks, even when running, were being aborted. In all those instances the canonical result was already migrated off from the active Result viewing pages.

Found one thread where it's said to have been in place from 6.6.12 [auto delete], just not seen it for tasks still on the Result pages of the project and not having reached a validated state. [WCG, which is 99% what my hosts do].

---

Coelum Non Animum Mutant, Qui Trans Mare Currunt

"If a task is still unstarted on your computer when the deadline is reached, then your local BOINC client will abort it."

As of which version is this a client policy to abort unstarted+overdue tasks? There are the red line messages if running overdue tasks to say there is a chance of not getting credit, but never observed the client having initiative... it's at least with WCG the 'server aborted' notice sent, and only if any of the two [the extra copy or the original] has returned with success.

To add: Seen on part time clients and dual boots, where seriously overdue tasks, even when running, were being aborted. In all those instances the canonical result was already migrated off from the active Result viewing pages.

Found one thread where it's said to have been in place from 6.6.12 [auto delete], just not seen it for tasks still on the Result pages of the project and not having reached a validated state. [WCG, which is 99% what my hosts do].

It's "auto abort" by the client, acting on its own initiative.

The web site will still say 'no reply', until a separate, later, scheduler RPC to report the changed state (after all, the task might have started running a few seconds before deadline, and not been aborted at all).

Whatever the proper semantic descriiption, dug a little further and found this snip from 2012 http://boinc.berkeley.edu/dev/forum_thread.php?id=7704&postid=44809.

Ever since BOINC v6.6.12 (March 2009), the client will self-abort a task in the cache which hasn't been started before the deadline is reached. In that case, the client returns

#define EXIT_UNSTARTED_LATE 200

Never ever seen it happen.

---

Coelum Non Animum Mutant, Qui Trans Mare Currunt

Saw it happen on one of my computers a few months ago. WCG, in fact. It was the first set of runs for (I believe) the Ebola project when the tasks were estimated around a day, with a 3 day deadline for resends, on my slow computer that took up to 5 or 6 days to run. Only one or two were canceled due to not starting in time. It is no longer an issue, but remains one of my pet peeves against the "3-day resend" deadlines.

There is also Volpex, which sometimes (rarely) delivers WUs after the deadline. But then again, they have a 5 minute deadline.

In any case it is most likely to happen when a project sends WUs with deadlines that are shorter than the length of the cache.

BOINC is supposed to schedule those short-deadliners to the front, unless there are so many, they still are stuck in the traffic jam.

---

Coelum Non Animum Mutant, Qui Trans Mare Currunt