[RC5] RE: Do not forget about the cheaters :)

groovyr at comcast.net groovyr at comcast.net
Thu Jan 8 08:08:21 EST 2004


>Indeed, but there is an anti-cheating system in place which should allow, by
>means of some very simple statistics (merely looking at the data, or if you
>want to be scientific, apply the chi-square test), to detect these cheaters
>and invalidate their accounts.


Yes but it is failing and failed when a few on the German team were cheating
back in September and is failing now for the polish team which has obvious
cheating going on.   I even went to their forum and some team members
questioned other team members on HOW THE HELL are they doing so many
blocks and the replies were everything from wink wink to "offline machines
doing a huge dump and it will be back to normal tomorrow" then of course
the next day comes and STILL mega blocks being submitted.   It is so bad you
have polish team members lying to other team members.


>
>Unfortunately the system is far from fool-proof. Certain shortucts are
>possible which allow short-circuiting the work done by more than half (I'd
>say ~70% of work can be evaded without allowing detection even by the
>chi-square test).

yup


>
>But cheaters don't even need to be that sophisticated: AFAIK, there's no
>automated system to check for cheats, so unless some very blatant cheating is
>going on, it won't be noticed.

well we noticed it because it is blatant on the polish team.  1/3 the members
of the dutch team doing double to triple the blocks daily is nonsense. 
We tracked some of the polish members and they went from a few hundred units
day to 30K a day in 1 huge jump.    Not just 1 member but within 2 weeks a half dozen
team members jumped like that. My guess it the hacked client was passed around to
a few buddies.

30K units would take like 300  3GHz P4's to crank that out a day AND doing nothing else.



>It needn't be that way. Some Stanford researchers published a paper at the RSA
>Conference 2001 (which, by the way, was already noticed by the guy who
>released the hacked client back then). It's available at
>
>http://crypto.stanford.edu/~pgolle/papers/distr.pdf
>
>for anyone who cares about the gory details. For those who don't, in short,
>it's possible to detect cheaters with very high probability and minimal
>overhead. And it's easy to implement too, particularly considering the system
>that is currently in place.
>

very interesting stuff but if each client marked the blocks at the client level
then any client doing 2000+ a day would be checked.  no CPU time
used at all until then and you would know exactly were to narrow the search
and thus could actually do some blocks for real to make sure they are
bogus blocks.

Besides you need this to find a cheater on a sub team since you can not
penalize the whole sub team.   Each client should me identifiable.





More information about the rc5 mailing list