Algorithm-Based Secure and Fault Tolerant Outsourcing of Matrix Computations

Amrit Kumar (1) , Jean-Louis Roch (1)

1 MOAIS - PrograMming and scheduling design fOr Applications in Interactive Simulation

We study interactive algorithmic schemes for outsourcing matrix computations on untrusted global computing infrastructures such as clouds or volunteer peer-to-peer platforms. In these schemes the client outsources part of the computation with guaranties on both the inputs' secrecy and output's integrity. For the sake of efficiency, thanks to interaction, the number of operations performed by the client is almost linear in the input/output size, while the number of outsourced operations is of the order of matrix multiplication. Our scheme is based on efficient linear codes (especially evaluation/interpolation version of Reed-Solomon codes). Confidentiality is ensured by encoding the inputs using a secret generator matrix, while fault tolerance is ensured together by using fast probabilistic verification and high correction capability of the code. The scheme can tolerate multiple malicious errors and hence provides an efficient solution beyond resilience against soft errors. These schemes also allow to securely compute multiplication of a secret matrix with a known public matrix. Under reasonable hypotheses, we further prove the non-existence of such unconditionally secure schemes for general matrices.

Mots clés: Algorithm-based fault tolearance; Matrix computations; Linear codes; Perfect secrecy

Domaines: Cryptographie et sécurité [cs.CR]

Amrit Kumar, Jean-Louis Roch. Algorithm-Based Secure and Fault Tolerant Outsourcing of Matrix Computations. 2013. ⟨hal-00876156⟩