[viff-devel] FW: Bug in ViFF

[Hoogh, S.J.A. de]

Thanks for your early feedback.

I am running ViFF 0.6 so I will upgrade right now and tell you whether
the problem is still there. 

Sebastiaan

-----Original Message-----
From: Martin Geisler [mailto:mg at daimi.au.dk] 
Sent: woensdag 1 oktober 2008 9:39
To: Hoogh, S.J.A. de
Cc: viff-devel at viff.dk
Subject: Re: FW: Bug in ViFF

"Hoogh, S.J.A. de" <S.J.A.d.Hoogh at tue.nl> writes:

Hi Sebastiaan,

Thanks for giving VIFF some exercise! :-)

> Tomas Toft and I are using ViFF to analyze Toft's secure Linear
> Programming protocol. At the first sight everything seemed to work
> nicely, but when trying to increase the number of participants and
> the threshold we've found an annoying fact about ViFF:
>
> The threshold should be given explicitly when creating the runtime
> in any ViFF program. ViFF doesn't check whether this threshold
> matches with the parameters given in the configuration files.
> Wouldn't it be nicer if the threshold is read by "create_runtime(id,
> players, threshold, options, Toft05Runtime)" automatically?

You mean that the threshold should be put in the player-X.ini files? I
guess that is a good idea... As it is now, VIFF simply trusts you to
get everything right -- it lets you pretend that you know what you are
doing :-)

The threshold used when generating config files determines the
threshold for PRSS, whereas the threshold given to create_runtime
determines the threshold for Shamir sharings. Now that I think of it,
it seems natural to put both thresholds in the config files and unite
them at the same time.

> Now doing this consistently and setting #players =3 and threshold =1
> works and setting #players=5 and threshold=2 also works. However,
> there seems to be some bug somewhere when trying #players=5 and
> threshold=1. I've altered the millionaires.py such that it deals
> with 5 players instead of three and it provides correct output if
> threshold=2 but correct output for the participants 1, 2 and 3 only
> (and junk for players 4 and 5) if threshold=1...

Hmm, that is bad... did you generalize millionaires.py so that there
can be 4, 5, ... millionaires? How did you sort the inputs -- like it
is done in sort.py?

> In addition, if I create the config-files with 5 players and
> threshold=1 and if I put threshold=2 in the millionaires program, it
> works again, i.e., provides correct output....
>
> Hopefully, this information helps you finding the bug. If anything
> needs some clarification I'm happy to provide you more details...

Is this with version 0.7, which was just released? Also, can you send
us the modified millionaires.py program?

I have recently tested multiplications with (n, t) = (25, 8) and it
ran fine, but I must admit that I did not check that the results were
correct, I only made sure that the benchmarks could be run.

-- 
Martin Geisler
-------------- next part --------------
A non-text attachment was scrubbed...
Name: millionaires-5.py
Type: application/octet-stream
Size: 6587 bytes
Desc: millionaires-5.py
URL: <http://lists.viff.dk/pipermail/viff-devel-viff.dk/attachments/20081001/e84d0cfb/attachment-0001.obj>