FHE-Booster: Accelerating Fully Homomorphic Execution with Fine-tuned Bootstrapping Scheduling

Abstract

Fully homomorphic encryption (FHE) allows a user to outsource computation-intensive tasks to a cloud server witheut providing plaintext values or decryption heys to the server. A major drawback of these encrypted operations, however, is that they can be orders of magnitude slower than their plalintext counterparts. Moreover, because each ciphertext can only tolerate a llmited number of operatlons before the accumulated nole renders decryption impossible, an operation known as bootsirapping is needed to reduce such nolse and allow for unilimited computations. Notably, bootstrapping is signincantly slower than encrypted arithmetic operatlons, thus becoming a main performance bottleneck while evaluating FHE programs So far, the allocatlon and scheduling of bootstrapping operations has not been well Investigated, In part due to the complexity of the probkem and the difinculty in finding an optimal solution. To bridge thls gap, in thls work we formulate the bootstrapping scheduling problem and develop two Integer Programming (IP) modek. The first minimlies the number of bootstrapplng operations in an FHE program, while the second optimines the evecution time of the FHE program. We further develop two heurlstics for mapplng a target FHE program to a multi. core system in polynomial time. Our evaluation with a reallstic benchmark shows that our heuristic provides a 1.86x speedup compared to the baselline method.