Hi Sporally,
Sorry so long to reply; I've been very busy...
Many people are indeed unhappy will the long WU times, judging by the E@H forums.
The reasons for it, AFAIK, are that the previous run was coming to an end, LIGO had a bunch of data that needed processing, and a new app was finished just in time to use on that data -- an app which was more sensitive, and so inherently took longer.
And, since the app was rushed, it did not include the code optimization that had sped up the previous few production apps. These two things combined to make WUs take much longer, particularly on Windows, and most particularly on Windows/AMD, if I understand the posts correctly.
From the point of view of the project, this is not all bad, since the science is getting done: the S5R2 search is now 34% complete.
From the point of view of crunchers, the effects can be bad, as you have seen -- with missed deadlines being one of the worst. We all take pride in returning work on time. And of course slower computers are punished most.
And there are also credit issues, I don't follow the discussions about this too closely, but as best I understand, Einstein@Home was overclaiming in relation to other projects, but the now app has overcompensated, and on some platforms gives less than it should.
As far as deadlines go, I don't understand why they have not been extended: it's an easy code change, I think? I know it means it will take longer for all the results to come in, but would a week's extension affect the final time so much? I don't know the answer.
So to me, it certainly makes sense to stop crunching E@H for a while if you are missing deadlines, or you have to micromanage BOINC to avoid missing them. There are plenty of good projects to try out, which is the beauty of BOINC.
But there is hope!
Akos is on the case!!
Quote:
|
Originally Posted by akosf
My Core2 runs an SSE optimized version of XLALComputeFaFb subroutine. It shows about 70% performance improvement. Bernd will implement it into the source code and compile for all x86 based platforms.
|
So if anyone decides to put E@H on hiatus, please check back here when there are new posts, as I'm sure the first person to get the faster app will post...
___
As far as the far, far, future goes, there may be one day a project I've been wishing for:
cosmology@home.
At the moment, the project is nothing more than this website. From his postings on the BOINC projects mailing list, it appears the developer has a science app, but is having is having problems BOINCifying it: creating WUs, figuring out the scheduler, validator, and all the machinery it takes to get a project started. (Many projects seem to founder at this stage.)
So it is very much a pre-pre-pre-alpha dreamy vapor-project so far, but if it does happen, and does worthwhile simulations, I for one can't wait to crunch for it.
___
Oh, by the way, congrats to the team for going over
10,000,000 in total credit...
