On Sep 30, 2009, at 12:24 AM, Raistmer wrote: > >> I suppose that you would vote to just start using Hubble to snap >> photos without calibrating the instrument ... your way we get lots >> more pretty pictures ... my way we get far more usable pictures ... >> more accuracy, more science > > Well, will try to extend this example. > Let's say we calibrate Hubble once. Then meteor hits and calibration > broken. > And we even don't know in what part it broken. > If we have only single device we have no choice to put it down and > try to repair. > > But if we have ability to replicate it, we launch many "Hubbles" and > will compare results they return. > Those that mostly agree with each other will go. > Sometimes having such redundancy is cheaper than repair itself. > Sometimes repair just impossible. > That's why space devices very often carry redundant systems. Just > because you can't put it back and recalibrate after some random event. > > Very similar situation with BOINC itself. Your CAN'T ensure that > user PC will work OK even right after completion of your artifical > calibration task. So you should implement redundancy. But when > redundancy implemented, why you need calibration on participants > PCs ???
No, but next month I will catch on to the fact that they are broken then ... and then I have the choice to invalidate the work done between the last known good and the broken system ... The flaw in your comparison is that both with adaptive replication and single task validation more and more projects are going to single task validation... I remember when we thought that min quorum of 3 was too low ... to be honest I still feel that way ... and with the speed of GPUs, I don't see that it would hurt ... of course, I am for accurate results not just numbers of results ... Yes you can find flaws in any system ... and this one is not going to catch every error ... then again not looking for errors does not catch any either ... _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
