E@H Cache vs Deadline

> "What is considered SMALL??"
>
> Ran across this definition of resource share "When you join your second and
> subsequent projects this value and the value in the other project's
> preferences are totaled up and then used to create a proportion that indicates
> the amount of time to be allocated to that project."
> http://boinc-doc.net/site-common/oman-web/preferences-project-cah-view.php
> Resource share controls time, so if you set the share too low, your computer
> won't have enough time to finish a wu for some project. What value that is
> depends on the speed of your machine, the deadline, and how many projects you
> are attached to. At least that's my take.
>
> I guess 'enforce_delay_bound' must be enabled from the server side. Don't see
> it in the command line options http://boinc.berkeley.edu/client_unix.php
> unless it is a compile option?

Correct: enforce_delay_bound is a SERVER option. I should have enabled it three months ago, but didn't realize it was OFF by default.

This does the following. When the core client requests work from the scheduler, it also reports how many seconds of work it still needs to do. Before sending work, the scheduler asks if the sum of the execution time of this work and the amount of pending work on the host would give a projected finishing time after the deadline. If so, the work is not sent to the host.

To compute the expected execution time on the host, the scheduler takes into account the measured floating point speed of the host, plus the resource share given to the project, and the number of CPUS. Roughly speaking

expected exec time = (WU FLOPS * seconds/FLOP)/(RESOURCE FRACTION * NUM_CPUS)

I expect that in many cases, the problem we have been having is not that 'the deadlines are too short'. The problem is 'the scheduler sends more work to the host than the host can expect to finish before the deadline!'

Cheers,
Bruce