[viff-devel] Player thresholds

[Håvard Vegge]

Thanks for clarifying!

Guess there are a lot of different variations of the definitions, 
notions, etc. regarding secret sharing and MPC...
One of my problems is that when running the included sort.py protocol 
with n=5 and t=2, the sorting of the array fails:

Original array: [{33}, {61}, {95}, {67}, {37}]
Sorted array:   [{6692966529242218069}, {35128728804386641877}, 
{6621921405115695795}, {27880759555216652877}, {34356088148296101764}]
Made 9 comparisons

Works with n=5 and t=1 though...


Mikkel Krøigård wrote:
> Citat af Håvard Vegge <havardv at stud.ntnu.no>:
>
>> Hi!
>>
>> I'm trying to do some basic benchmarks with different number of 
>> players. A few questions:
>>
>> 1. Why have VIFF defined the threshold t to be the number of corrupt 
>> players, while the classical literature defines it as "t participants 
>> can reconstruct the message"?
> Well, in what I have read, it is most often referred to as the 
> threshold for the number of tolerated corruptions, though yes I have 
> also seen it described as the number of shares needed to reconstruct. 
> I don't think there's anything wrong with the way we do it, as long as 
> it is clearly defined.
>
>> 2. Say I create the config files this way:
>> python generate-config-files.py -n 5 -t 2 localhost:9001 
>> localhost:9002 localhost:9003 localhost:9004 localhost:9005
>>
>> Would this indicate that the polynomial is quadratic (of degree 2) 
>> and that 3 players in theory could reconstruct some secret?
> Yes.
>
>
>> 3. Assume three of these five players provide input, while the last 
>> two players do not. Will all players still participate in the actual 
>> computation, or are the last two just passive spectators and receives 
>> only the output?
> -n 5 indicates that 5 servers will participate, regardless of how many 
> will provide input. Think of it from the perspective of security. If 
> the input was shared to 3 servers, we could not tolerate 2 corruptions.