The following 3 mechanisms work together:
1) adaptive replication (to reject wrong results with high probability)
Possibly with an added app-specific consistency check as John suggests.
2) enhanced host punishment (to prevent cherry-picking)
3) the new credit scheme (to prevent credit cheating)
These will provide reasonable levels of validation and
cheat-resistance with a minimal performance penalty
(i.e. a replication factor only slightly over 1).
Hopefully they'll become the defaults.
- DPA
John Keck wrote:
> I doubt it, at least at the BOINC level. Individual projects may be able
> to with sanity/consistency checks embedded into the tasks though.
>
> john
>
>
> On Fri, 2009-11-06 at 20:28 +0000, James Wanless wrote:
>> Indeed! Come to think of it, will it ever, logically, be possible to
>> avoid cheating w/o replication???
>> [that's a genuine question, rather than a rhetorical one...]
>
_______________________________________________
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.
