International Association
for Cryptologic Research

Key-policy attribute-based encryption scheme (KP-ABE) uses a set of attributes as public keys for encryption. It allows homomorphic evaluation of ciphertext into another ciphertext of the same message, which can be decrypted if a certain access policy based on the attributes is satisfied. A lattice-based KP-ABE scheme is reported in several works in the literature, and its software implementation is available in an open-source library called PALISADE. However, as the cryptographic primitives in KP-ABE are overly involved, non-trivial hardware acceleration is needed for its adoption in practical applications.

In this work, we provide GPU-based algorithms for accelerating KP-ABE encryption and homomorphic evaluation functions seamlessly integrated into the open-source library with minor additional build changes needed to run the GPU kernels. Using GPU algorithms, we perform both homomorphic encryption and homomorphic evaluation operations 2.1× and 13.2× faster than the CPU implementations reported in the literature on an Intel i9, respectively. Furthermore, our implementation supports up to 128 attributes for encryption and homomorphic evaluation with fixed and changing access policies. Unlike the reported GPU-based homomorphic operations in the literature, which support only up to 32 attributes and give estimations for a higher number of attributes. We also propose a GPU-based KP-ABE scheme for publish/subscribe messaging applications, in which end-to-end security of the messages is guaranteed. Here, while the exchanged messages are encrypted with as many as 128 attributes by publishers, fewer attributes are needed for homomorphic evaluation. Our fast and memory-efficient GPU implementations of KP-ABE encryption and homomorphic evaluation operations demonstrate that the KP-ABE scheme can be used for practicable publish/subscribe messaging applications.