Not all CPU's are equally capable in the FPU/Integer unit department, but offsetting for cycle efficiency, 3 times speed difference is too steep for same platform comparison for not far apart CPU's. Sure they're not switching up or down [stepping] due localized heat spiking [They run ultra cool compared to any WCG science on my laptop I7-2670QM].

Nothing much happens in the immediate term per uplinger. Each connect there's a counter being set of 3 days or some other "if there is no N percent completed/reported by sent date plus X, consider the task as discontinued and generate new task from last good trickle."

Theoretically there could be duplicate steps computing, but do not know if the intermittently connecting hosts' [later] result trickles are acknowledged. So i.e.

1) Yours report 50K trickle
2) Offline computing, trickles accumulating in the transfer queue
3) Offline to beyond the critical 'soft stop' point
4) New task generated and sent from 50001 to 150000 which start trickling
5) Original host connects again after 4) and sends more trickles.

The fun starts when WCG goes offline for longer maintenance which can cause one or the other not managing to get the N percent in by the soft stop point.

On a soft stop, we do not create the next work unit at that point. We let your client stop running at the next checkpoint and return the work unit. Once we get that back, then we create the next work unit.

On hard stop, we start from the last know spot immediately, we do not give your client the option to complete successfully. If you have some trickle messages returned by this point, you will get credit for those during the validation of the result. Anything returned after validation is unfortunately not used as another host is already computing those steps.

Thanks,
-Uplinger

yesterday just installed 1 more machine with Ubuntu...lots more FAHB available on Ubuntu, 'cause of a smaller number of machines!

also, why r those deadlines so short? all my WUs crunch in "high priority" 'cause of that...Win & Linux variants!

Soft Stop if minimum number of trickles are not returned by a certain percent of total allowed deadline time. It's good to know the process actually waits for that next trickle before a follow on task is generated... minimizing chance of duplication.

Hard Stop is when a trickle is bad. Any additional trickle is also considered bad [which is logical as when the trajectory goes off the narrow path, the following steps will be off too].