How does F@H protect against errornous completed WUs?

[user123]

I'm not a new user but am curious to know how F@H sieves out completed WUs that are bad.

There are computer users out there who push their CPU, RAM or GPU overclock a little too far, which can result in a WU being completed incorrectly.
(Need not be overclockers but defective CPU, cache and RAM can result in errornous completed WUs)

I've read of WUs ending in EUE (early unit end) etc. in unstable PCs, but do improperly completed WUs always end in EUE?

[7im]

Usually, yes. However there are additional data integrity checks after the WU is uploaded. See the FAH WIKI entry on Common Errors, for the message Server reports problem with this work unit.

[Jesse_V]

The core does more than just simply run calculations. It also checks to see if the values it's working with makes sense. Are these numbers legal and in their proper range? Are the atoms and forces reasonably configured? Do the integrity checks pass? Are the checkpoint hashsums correct? etc etc. As 7im said, some of this is also done server-side.

There's also a function in place to send out WUs again when someone encounters a problem with it. If others successfully complete it, then it's likely a hardware problem for that first user. If not, then it's a bad WU and a researcher needs to debug the problem.

[bruce]

When a DC project is originally designed, they can take one of two opposing approaches: They can spend a lot of time and effort on error checking or they can waste an appreciable fraction of the donor's efforts by reprocessing everything and confirming the data is repeatable. FAH put their money into error checking as a way to minimize (as much as possible) the number of WUs that need to be reprocessed. To make that work, they have to have validation tests that can be trusted to detect errors earlier rather than later (i.e.- a bad WU should NOT be completed) so that good results only need to be processed once. This will also minimize the number of errors that make it all the way to the server-based checking stage.

Unfortunately, in addition to hardware errors and a small percentage of WUs which are inherently unstable, there is also the possibility of hacking. FAH points have an intrinsic value even less than bitcoin, but that doesn't guarantee that all donors are ethical. Extra levels of error checking are also a lot like locking your house when you're away. If everybody is ethical, that wouldn't be necessary. If a occasional thief happens to visit, the person in the neighborhood with the best security system will have the lowest chance of being broken into.

[user123]

Thanks for the informative replies.