|
I hate to say this, but I'm going to ask whether the even the currently very long 28-day deadlines are long enough. We seem to be doing higher n's now than a few months ago, and it seems as if runtimes have gone up substantially (in my case, from about 12 days to 18 days.)
That's with a middle of the road Q6600 crunching 24/7 with 100% share.
Throw in BOINC scheduling silliness, less than 24/7 computing, resource sharing between PG and other projects, or a slower processor and you could easily blow a 28 day deadline.
From the cruncher's perspective, missing a deadline isn't really a big deal -- I understand that the purge times are set so that you'll still get credit unless you're VERY late returning the result -- but it's bad for the project since extra and unneeded results will be generated for the WU, thus wasting processing capability redundantly computing the same result.
So, there's two sides of this, as I see it. The first is whether or not SoB is seeing a lot of results being returned late. (That's different from missed deadlines, which also include results that will *never* be returned.) Are a lot of results being returned late? This is something the admins should be able to pluck out of the database pretty easily.
The more difficult question is whether raising the deadline would get more slow computers working on SoB. I've got a few perfectly usable computers that could not possibly compete an SoB within the 28 day deadline. (However, even if the deadline was longer, I'd most likely keep them crunching something shorter.)
There are, of course, disadvantages both to the crunchers and the projects of raising the deadline.
I'm not advocating that the deadline be lengthened, but I did want to point out that the WUs (or at least some of them) seem to have gotten longer.
Mike
____________
My lucky number is 75898524288+1 | 
