Performance Analysis of Scientific Computing Workloads on Trusted Execution Environments

Abstract

Scientific computing sometimes involves computation on sensitive data. Depending on the data and the execution environment, the HPC (high-performance computing) user or data provider may require confidentiality and/or integrity guarantees. To study the applicability of hardware-based trusted execution environments (TEEs) to enable secure scientific computing, we deeply analyze the performance impact of AMD SEV and Intel SGX for diverse HPC benchmarks including traditional scientific computing, machine learning, graph analytics, and emerging scientific computing workloads. We observe three main findings: 1) SEV requires careful memory placement on large scale NUMA machines (1×-3.4× slowdown without and 1×-1.15× slowdown with NUMA aware placement), 2) virtualization-a prerequisite for SEV-results in performance degradation for workloads with irregular memory accesses and large working sets (1×-4× slowdown compared to native execution for graph applications) and 3) SGX is inappropriate for HPC given its limited secure memory size and inflexible programming model (1.2×-126× slowdown over unsecure execution). Finally, we discuss forthcoming new TEE designs and their potential impact on scientific computing.

Trustworthy scientific computing

“…Researchers at the Berkeley Lab and UC Davis have been empirically evaluating Intel SGX and AMD SEV TEEs for their performance under typical HPC workloads. Our results 1 show that AMD's SEV generally imposes minimal performance degradation for single-node computation and represents a performant solution for scientific computing with lower ratios of communication to computation. However, Intel's SGX is not performant at all for HPC due to TEE de-identification does not necessarily address the risk and trust issues involved in data sharing because reidentification attacks can still result in significant embarrassment, if not legal sanctions.…”

“…9 There are also issues with low-latency communication between TEEs, and also the cost of virtualization, that must be addressed to enable HPC at scale. 1 In addition, promising RISC-V efforts such as Keystone 4 exist that carry both the promise of broadening the scope of processors that contain TEEs, while also being open source and possible to formally verify. However, RISC-V based TEEs have not yet been developed that target scientific computing.…”