TeeBench: Seamless Benchmarking in Trusted Execution Environments

Trusted Execution Environments (TEEs) have enabled building secure systems that operate on untrusted machines. However, TEEs' architecture questions previous performance findings. The existing relational algorithms have been designed for traditional CPUs. Prior work has shown that these algorithms underperform in TEEs and, in most cases, can not be easily reused. Moreover, they frequently used benchmarks pertinent to CPUs and ignored TEE-specific metrics essential to understand the performance differences. Therefore, there is a need for a fair benchmarking approach for TEE algorithms. In this demonstration, we showcase TeeBench, a unified benchmarking framework for relational operators across TEEs. TeeBench focuses on TEE-specific hardware metrics. It enables a comprehensive performance analysis that helps researchers to evaluate their advances. It comes with an interactive web browser tool that allows the users to upload their implementation of a relational algorithm and seamlessly benchmark it across different TEEs. In addition, it introduces a novel TEE-Analyzer that hints the users about performance bottlenecks and suggests possible code improvements. Users receive instant feedback if changes to their algorithm improve the performance through an interactive, human-friendly web interface. We expect TeeBench to encourage the usage of TEEs and to advance the study of privacy-preserving systems.