VIFF has been used for several small and some larger applications. The largest applications are listed below. Please see the apps/ directory in VIFF for more examples of small programs using VIFF.

Nordic Sugar

In Denmark, the production of sugarbeet is managed by sugarbeet contracts. A sugarbeet contract determines the quantity of sugarbeet that a farmer is allowed to produce. Traditionally, sugarbeet contracts have been traded between individual pairs of farmers. This has been done in spite of the fact that trading in a central market was known to increase the overall profit. A central market has, however, not been possible due to conflicting interests and lack of trust between the parties.

In January 2008 the first large scale secure multiparty computation was carried out, effectively solving this problem. This was done by the SIMAP research project as reported in “Multiparty Computation Goes Live” (also published at Financial Crypto 2009). In the summer of 2009 the same computation was successfully repeated, this time using VIFF.

The computation was a double auction in which the production rights for several thousand tons of sugarbeets were traded. During the first weeks of the auction, several hundred Danish sugarbeet farmers submitted their encrypted bids to a central database. Then the actual computation took place between three players:

  • Nordic Sugar, the Danish sugar company
  • DKS, the consolidation of Danish sugarbeet farmers
  • Partisia, a Danish company specialized in secure multiparty solutions

The computation took about 15 minutes using three laptops on a LAN. Most of the computation time was spend converting the encrypted bids to secret sharings. The actual multiparty computation took only a couple of minutes. As a result, the sugarbeet contracts could be traded at an optimal price without any sensitive information being disclosed.

Using secure multiparty computation, trading sugarbeets using this kind of auction was possible without finding and paying a trusted third party to manage the auction. Such a trusted party would—if it could be found at all—probably have been quite expensive.

Distributed RSA

Atle Mauland from the Norwegian University of Science and Technology (NTNU) used VIFF for his Master’s Thesis titled “Realizing Distributed RSA using Secure Multiparty Computations”. The code is available for download.

The private key from a RSA key pair must be kept in a highly secure location (to prevent unauthorized persons from stealing it) but because we want to use the key, we cannot just write it on a piece of paper and store that in a safe.

This tension between high availability and high security makes a distributed solution attractive. Atle Mauland implemented a protocol by Boneh and Franklin for generating RSA keys in a distributed fashion. The protocol ensures that the private key is never available in the clear to any given party and an attacker must break into all machines to learn the private key. Meanwhile, the parties can use their shares of the private key to securely decrypt messages encrypted under the public key.

Generating a 1024-bit RSA key using VIFF took about 30 minutes on average (the time varied between 7 seconds and 2.5 hours). These results can likely be improved by using the GMPY library more aggressively.

Distributed AES

The Advanced Encryption Standard (Rijndael) block cipher turns out to have nice arithmetic properties which makes its computation by arithmetic circuits relatively fast. Marcel Keller from the University of Aarhus has implemented a multiparty version of AES for VIFF.

Using the viff.aes module, it is possible to securely compute a secret shared AES encrypted ciphertext of a (possibly) secret shared plaintext with a (possibly) secret shared key. The inputs have to be given either as a list of shares over GF256 (byte-wise) or as a string. The runtime has to be able to handle shares over GF256.

Encrypting a 128-bit block using a 128-bit secret shared AES key takes about 2 seconds using three machines. Decryption is not implemented yet.

Secure Voting

Typical Internet voting systems store all votes in a single location. Håvard Vegge from the Norwegian University of Science and Technology (NTNU) used VIFF for his Master’s Thesis titled “Realizing Secure Multiparty Computations” to implement a distributed voting system. The code is available for download.

The system removes the single point of failure by storing the votes in secret shared form between three servers. The voters will do the secret sharing on their own machine, encrypt the shares, and send the encrypted shares to a database. Each share is encrypted under the public key belonging to the computation server that will do the actual multiparty computation.

This project shows how VIFF can be integrated with many other technologies. The user creates a vote on a website programmed in PHP and the voting is cast using a Java applet. The applet has the responsibility of encrypting the votes for the computation servers. When all voters have cast their vote, a XML-RPC message is sent to the Python program running on the servers. That program decrypts the shares and uses VIFF to compute the result.

Table Of Contents

Previous topic

Installation Guide

Next topic


This Page