Wildly different runtime are normal for this project as the tests being run just simply don't lend themselves to accurate runtime predictions. I will say the 18 hours seems too long, if you are using a 6.6.x version of BOINC then you are seeing wall clock elapsed time and not CPU time. There is a "deadman" switch at 4 hours to automatically stop unless you are at least 60% finished and a hard stop at 8 hours. They are in the process of upping both those limits to 6 and 12 hours respectively but even if you got a WU from a new batch you still should not be running 18 hours.

Thanks for the response guys.

I just looked at my BOINC Manager and there's still some showing as 14hrs in the time to completion field.



http://img297.imageshack.us/img297/3064/hcmdworkunit.png

The Time to Completion (TTC) is yet another item!
Even if your version of Boinc were showing CPU times you could see such fancy values or even worse, particularly if you run HCMD2.
This is because the TTC is computed from an estimation of the workload included in the WU through a somewhat strange algorithm and adjusted via a Duration Correction Factor (DCF).

The estimated workload is expressed as the number of floating point operations to do, and obviously for non-deterministic computations it is more or less approximative, below or above. The DCF is recomputed after each job completes, and if you had a series of jobs running longer than expected the DCF may be much too high. Add to this that the algorithm will never adjust the TTC quickly, even if a human observer can see after 10 % of the job that it will take 4 hours instead of the displayed 14. Instead the TTC is adjusted progressively and will match reality only at the end of the job. If you are in the oppposite case (an underestimated TTC after a series of jobs quicker than anticipated) the TTC will increase very little slowly from its start value until it matches a realistic value. Then it will behave normally.

Once you know this you will not care too much about the displayed TTC and you will make your own estimate from the percent done and the already used CPU time.
Note that if you use a "wallclock" version of Boinc you have one more correction to do considering what else your machine was doing before and what else it will do until the job completes.

Good luck! Jean.

Edit: Changed "very little" to "very slowly"

HCMD is more difficult to estimate when the WUs are created, but almost all projects are subject to this problem, depending on the cases which are computed in the WUs.
Only HCC is rather stable in duration because the nature and the boundaries of the work are always the same (image analysis).

Cheers. Jean.

Let me have the project credits per hour speak for them selves how odd that duck is:



It's on recourses the absolute lightest project at WCG. CEP is higher only because allot of clients are focused there the higher performance devices segment (quads on badge hunt), this project demanding quite a bit more resources for sure!