APPLICATIONS

A team of French researchers from the Delegation Générale de l’Armement (DGA/CELAR and DGA/SPOTI) and the University of Versailles-Saint Quentin en Yvelines has set a new record for the discrete logarithm problem. The computation was performed on TeraNova, a supercomputer clustering of Bull NovaScale servers. TeraNova demonstrated its power and reliability by computing a discrete logarithm in GF (2 613) which is the largest computation ever done. The ability to compute discrete logarithms is critical to finding the cryptographic key used in the Diffie-Hellman algorithm. This public key algorithm is commonly used when two persons want to exchange confidential messages without sharing any secrets. To build their common cryptographic key, they exchange over a public network two numbers that are exponentiations using information that needed to be kept secret. If someone can compute the discrete logarithm of one of these numbers, they can find the key that will be used. This work shows that discrete logarithms can be calculated with reasonable computing power for very large numbers having 613 bits. As a result of this new record, the cryptography community has a better knowledge of the robustness of algorithms like the Diffie-Hellman one, and can set recommendations on how to use these algorithms without risking security in the exchange of information. The record was set using the TeraNova system which is a cluster of 16-way NovaScale servers. Despite the challenge of this computation, it took only 13 days for 64 processors to find the solution. This shows the efficiency of Itanium 2 processors for this kind of cryptographic calculation. Delivering 2 teraflops, the TeraNova cluster of 20 servers, each equipped with 16 Intel Itanium 2 processors, operates within the TER@TEC HPC project part of the SYSTEM@TIC cluster. TER@TEC was created to promote the use of digital simulation and high-performance IT through scientific co-operation between partners in research and industry worldwide, among them Bull and the University of Versailles–Saint Quentin en Yvelines. “Thanks to TeraNova, we were able within a reasonable time frame to show that for such large fields, the general function field sieve algorithm outperforms its ancestor: Coppersmith's algorithm,” said Antoine Joux, Scientific Advisor to DGA/SPOTI (Délégation Générale pour l’Armement) and Associate Professor at University of Versailles Saint-Quentin. “The computing power of TeraNova considerably increased our research capabilities,” underlined Professor William Jalby. “The cooperation between the researchers and the Bull experts was key to establishing this new record.” “We are proud to see how NovaScale servers can help researchers solve key technology challenges, which are at the heart of world-wide concerns in terms of security,” said Jean-François Lavignon, HPC General Manager. “The cooperation between Bull and research laboratories is becoming more and more successful within TER@TEC and will continue to expand in the future.”