What happens to over deadline work units ?

[Geoff_A_Jones]

I usually complete uniprocessor units with 90% of the deadline remaining so I decided to try the SMP client however it is evident I will not manage to complete the unit within the 6 day deadline.
So I was wondering what happens to overdue work units, are they of any use if submitted late, or is it best to delete them immediately if it is obvious that the deadline cannot be met?

[uncle_fungus]

If you submit results after the "preferred deadline", you will receive the points, but the work will normally be reassigned to someone else resulting in duplication of work.
If you submit results after the "final deadline", you won't receive any points.

If your machine cannot meet the SMP deadlines, I suggest you continue to run multiple uniprocessor clients on your machine instead.

In general it is a bad idea to delete workunits (known as dumping) since the assignment server logic is highly likely to resend you the same unit the next time you request one since the system hasn't received the results from the original WU.

[metal03326]

Read this - http://fahwiki.net/index.php/How_Import ... adlines%3F
It's not exactly what you are looking for, but it's good to know.

[Geoff_A_Jones]

If you submit results after the "preferred deadline", you will receive the points, but the work will normally be reassigned to someone else resulting in duplication of work.
If you submit results after the "final deadline", you won't receive any points.

If your machine cannot meet the SMP deadlines, I suggest you continue to run multiple uniprocessor clients on your machine instead.

In general it is a bad idea to delete workunits (known as dumping) since the assignment server logic is highly likely to resend you the same unit the next time you request one since the system hasn't received the results from the original WU.

Disregarding the points scoring system my question was aimed at discovering if a work unit was received back late is the work unit just discarded or is the data used.

for example if a work unit has a final deadline of 6 days and is reallocated to another folder on day 7 but the original work unit is received on day 8, the resent unit being returned on day 11.
Would the data received back first be used?
Would either the first or second set of data be discarded and why?
what would happen if the replicate results were not identical.

[anandhanju]

I thought the client killed the WU once it passed the final deadline, which meant there was no way one could send results on day 8 when the final deadline was on day 6. However, I'm not certain if this is the same behavior with the SMP client. Has your client been folding the same WU after you have noticed the expiration of the final deadline?

[MstrBlstr]

I am thinking that it only does that check when you first start up the client. I could be wrong however.

It may very well do it at the 6 hour mark, when it checks to see if you have a unit to upload as well. I am not certain.

If the unit is uploaded, it is kept AFAIK.

[codysluder]

The client does check for overdue WUs on restart but it's supposed to end a WU if it's working on it and the deadline passes. I don't know if that works correctly, though. I think it did with the cpu client but may be broken with -smp.

If the WU has been finished but not uploaded, it will be deleted from the queue.

[John Naylor]

for example if a work unit has a final deadline of 6 days and is reallocated to another folder on day 7 but the original work unit is received on day 8, the resent unit being returned on day 11.
Would the data received back first be used?
Would either the first or second set of data be discarded and why?
what would happen if the replicate results were not identical.

1) No.
2) The work unit should self-terminate when the final deadline passes, but if it doesn't (say if the option is checked to ignore the local clock) then the work unit may be used when received, we don't know. The user would not get any points for it however. The second set would be used so long as it is returned within the deadline.
3) Cross-checking of completed units is not performed, this is only done for EUEs. Whichever result is used would be what goes, as the software has enough data-accuracy checks that it can be assumed that all completed units will complete with the same results (unstable machines trigger EUEs [Early Unit End], unit restarts, or UMs [Unstable Machine, but this reason is only used on the GPU client] so bad overclocks can be ignored as a possible problem.)