Parallelizing matrix multiplication for Intel SGX

DegreeBachelor
StatusAvailable
Supervisor(s)Alexander Schlögl, MSc

Description

We have built a framework that partially black-boxes machine learning inference using Intel SGX. It has been stated in the discussion after the Foreshadow presentation that “hyperthreading [in SGX] is side-channel by design”. However, we are interested in the theoretical performance of parallelized matrix multiplication in an SGX enclave.

The project would implement matrix multiplication, parallelize it, and evaluate the performance on multiple problem sizes and numbers of threads. A comparison between regular untrusted execution and execution inside the SGX enclave would be useful to gain insights into the impact that running inside a trusted enclave has on performance.

Prerequisites

Knowledge in parallel algorithms, no prior SGX knowledge necessary

References

  • Schlögl, A. and Böhme, R. eNNclave: Offline Inference with Model Confidentiality. In 13th ACM Workshop on Artificial Intelligence and Security (AISec’20). 2020. [PDF] [Video]
  • Bulck, J.V., Minkin, M., Weisse, O., et al. Foreshadow: Extracting the Keys to the Intel SGX Kingdom with Transient Out-of-Order Execution. In USENIX Security Symposium. 2018, pp. 991–1008.