Title:Benchmarking Analytical Query Processing in Intel SGXv2

Abstract:The recently introduced second generation of Intel SGX (SGXv2) lifts the memory size limitations of the first generation. Theoretically, this promises to enable secure and highly efficient analytical DBMSs in the cloud. To validate this promise, in this paper, we conduct the first in-depth evaluation study of running analytical query processing algorithms inside SGXv2. Our study reveals that state-of-the-art query operators like radix joins and SIMD-based scans can indeed achieve high performance inside SGXv2 enclaves. These operations are orders of magnitude faster than joins optimized for the discontinued SGXv1 hardware. However, substantial performance overheads are still caused by subtle hardware and software differences influencing code execution inside an SGX enclave. We investigate these differences and propose new optimizations to bring the performance inside the enclave on par with native code execution outside an enclave.