I'm a recent joiner to MW, though I've been running BOINC since it was just SETI. I'm a member of 6 active projects now.

I have noticed that MW is serving up WUs that are due within a few days. That is not a generally a problem. But it is a problem when it serves them up on a Thursday or Friday such that those WUs are due at the weekend or early Monday. My machine is a work machine; it is generally up 24hrs a day but only during weekdays. BOINC only seems to run MW at high priority shortly before the deadline; but by then my machine is off.

So, I find BOINC only completing those WUs first thing Monday and they are rejected on return. Obviously that's a waste of both our times.

I cannot find anything in BOINC preferences that allows me to mitigate this. Is there something that could be done on the MW side?

Perhaps not have deadlines for Sat-Mon inclusive?

Thanks, Simon.

3 day turn around is what it is here. If you aren't able to run the computer(s) over the weekend so they can contact the server when work is done then your best bet is to set MilkyWay to NNW (No New Work) on Friday morning and upload the completed work when you leave and just start up again on Monday. MilkyWay won't be changing the turnaround time anytime soon.

---

The reason for that is that your results influence the rest of the search - if results are handed in too late, the search will have likely have moved on and will be negatively influenced by them.

The reason for that is that your results influence the rest of the search - if results are handed in too late, the search will have likely have moved on and will be negatively influenced by them.

I'm intrigued... If I understand you correctly, the results (the actual conclusions not just when they are reached) of MW depend on whether participants power up their PCs or pay their BB bills or just detach from the project. That seems... odd?

But, thanks for your explanations. It's a bit of a shame; I doubt whether I'm going to remember to do NNW for MW every week or even to abort those that are still around at the end of the week to give someone else a chance with them. For me, I estimate that some of the WUs I get on thurs and all WUs I get on fri are not completed by the deadline due to this. That's ~30% of WUs you send out I send back late.

My explanation may be out of date, I'm not sure. It's easy to understand with a genetic search: your results are combined with someone else's to get the next WU, so you don't want really old WUs coming in and messing up the 'gene pool'. (minor variations in turnaround are okay since they help you avoid local maxima)

The current search method uses particle swarm, that is: each WU tries out a set of parameters, and their results are used to tweak the next set that is sent out. If you return an old WU, chances are it'll either be deemed an 'outlier' (too far off from the norm to be used) or be too bad to effect results much. You won't be helping the search, but I don't think you'll slow it down the way you would for a genetic search.

Adding to that, there's already a large difference in turnaround between all CPUs and the GPUs crunching for the project - if the CPUs' comparative slowness was hurting the search, I suspect it would have been mentioned by now. (of course the GPUs are going to be split off into a daughter project, but that's due to the strain they put on the server)

So maybe the deadlines could be raised somewhat? It would be great to hear from someone who actually knows, I'm just speculating..

So maybe the deadlines could be raised somewhat? It would be great to hear from someone who actually knows, I'm just speculating..

Some time ago there was general complaining about the deadlines being too short and it was said that 3 days was already fairly lengthy because the searches had already moved in another direction sometimes by the 3rd day... The suggestion to not get work would be the appropriate thing to do given the circumstances.