Thank you Jean.

As to the runtime required - I see that only ~1,64% of the work is being completed per calendar hour (not the run hours shown in Boinc) on this PC. The PC is a Core i7-860@2,80GHz with Linux Fedora 12, HyperThreading is on, CPU Frequency is On-demand, oscillating between 1,2 and 2,8 GHz. My settings limit CPU usage by 70% for Boinc. Given these numbers, approx. 60 calendar hours needed to complete a Type A unit.

BTW, if a deadline for DDD Phase 2 is missed and the result is returned after the deadline - how long the result will still be useful for the project after the deadline?

Last time this was discussed, it was decided to not just auto-abort unstarted tasks, even the estimates indicates can't meet the deadline.

Instead, the BOINC API includes a function that tells how long until the deadline, and it's up to the individual project if they wants to use this by adding something to their applications. This can example be a check on application-start, example for Rice:
if time_to_deadline < 7 hours, don't run the task but immediately abort it.

Another use would be something like:
if time_to_deadline < 1 hour, stop task and upload current progress.


BTW, while most users wouldn't hit this problem, one of the rules of the Scheduling-server is to always assign a single task to a computer that doesn't have any work in project, except if it's hard_app, even if the computer based on the various estimates has no hope of returning it by deadline. So, if client just auto-aborts any tasks that can't meet the deadline, these computers will basically download X tasks until hitting their daily quota, and continues this for each continuous day, and never doing any work.

If a project makes an "oopsie" and incorrectly sets the wu-estimates much too high, with client auto-aborting, fairly quickly all wu with the wrong estimates will error-out, and it's a good chance most of the computers on project is sitting with daily-quota of 1, and with upto a 24-hour deferral until can ask for more work... (more than one non-WCG-project has made this oopsie).

By not auto-aborting, client will over time learn the estimate is bogus, and the DCF drops to 0.01 or something...

sepmsk,
If this is what you want, fine.
But in case you don't know, with this power management setting your processor is slowing down when it has "nothing to do for you". The problem is that BOINC with its lowest priority is considered by the system as a background task, not "something for you", and if you leave your machine switched on running BOINC only, for example while you have dinner, then the speed will drop to 1.2 GHz. Combined with your 70 % setting and hyperthreading that makes every thread behave as a very slow machine, hence your estimated runtime for an A-type.

If this is not what you want and if you would like to be able to process an A-type within its deadline you might start with setting your power management to Performance (or something similar) and raise BOINC to 100 % (with such a processor and the way BOINC works you should never be disturbed by BOINC).

And if this is not enough you might limit BOINC to 4 threads only ("use at most 50 % of the processors") while the A-type is running. You would still benefit from hyperthreading for your own activity but BOINC would process each of its 4 tasks more quickly.

Has the deadline been extended? I thought Type As has 8 days but its not.

ts01_ c441_ ps0000_ 1-- 612 Pending Validation 24/03/10 10:51:10 27/03/10 22:43:45 36.68 945.0 / 0.0 <- mine
ts01_ c441_ ps0000_ 0-- - In Progress 24/03/10 10:51:07 08/04/10 16:20:53 0.00 0.0 / 0.0

ts01_ c450_ ps0000_ 1-- 612 Pending Validation 24/03/10 10:51:37 31/03/10 20:13:56 33.41 982.3 / 0.0 <- mine
ts01_ c450_ ps0000_ 0-- - In Progress 24/03/10 10:51:36 05/04/10 06:18:24 0.00 0.0 / 0.0

I remember my deadlines were 1apr2010.

Or is the deadline is based on the capability of the machine?