SGX Full DBMS Experiments using Hyrise and Gramine

This is the repository containing our experiments investigating the Hyrise in-memory DBMS in SGXv2 using Gramine for the paper "Towards High-performance and Trusted Cloud DBMSs".

Abstract

Cloud Database Management Systems (DBMSs), such as cloud-native analytical or serverless databases, are experiencing rapid growth in adoption due to their flexibility and scalability. However, recent incidents with cloud providers show that the traditional model of a trusted provider/admin no longer applies to protect the customers' data. One promising solution that can prevent a sole reliance on cloud and database service providers are trusted execution environments (TEEs). While past TEEs had many limitations and caused high performance overheads, recent work shows that the support of TEEs like Intel SGX for DBMS workloads improved significantly. Thus, it is time to actively integrate TEE technologies into cloud DBMSs to achieve better security that does not rely on the cloud provider. In this paper, we discuss directions for how recent TEEs can be used to build efficient and secure databases. We summarize the recent results on Intel SGX's performance for DBMS workloads and lay out the remaining research challenges that must be addressed to achieve optimal performance and thus minimize the performance cost for additional security.

Requirements

• Ubuntu 24.04
• All requirements for compiling Hyrise
• All requirements for compiling Gramine
• Python package requirements

  sudo apt install build-essential libcap-dev

• Python packages from requirements.txt

Running the benchmarks

• Clone the repository and both submodules (slightly customized versions of Hyrise and Gramine)
• Move configs/machine-config.json to your home directory and adapt the configuration to your desired run directory, compiler versions and linker.
• Create a virtual environment, activate it, install requirements.txt into it, deactivate it
• Do not activate the virtual environment! Instead, set an alias to the python executable in the environment:

  alias vpython=$VIRTUAL_ENV_DIR/bin/python

  Do this, because the meson --install of gramine somehow gets confused by activating the virtual environment and then does not install the python packages of gramine to the correct directory.
• Switch to the scripts directory with cd scripts
• Run the experiment and plot scripts:

  vpython paper-exp-1-ootb.py
  vpython paper-exp-2-profiling.py
  vpython paper-exp-3-remaining-overheads.py
  vpython paper-exp-4-revision-operator-optimization.py
  ./paper-exp-2-perf-extraction.sh
  vpython paper-plot-1.py
  vpython paper-plot-2.py
  vpython paper-plot-3.py
  vpython paper-plot-4.py

  The scripts automatically store all results and figures under the run directory specified in machine-config.json. Compiled versions of gramine are stored under $HOME/gramine-bin.