Skip to content

Latest commit

 

History

History
76 lines (58 loc) · 2.92 KB

README.md

File metadata and controls

76 lines (58 loc) · 2.92 KB

SGXv2 Benchmarks

This directory contains the column scan, histogram, and read/write benchmarks used in the paper. They can be found inside the microbenchmarks directory.

Prerequisites

  • Intel SGX SDK for Linux (tested with version 2.24)
  • Intel SGX PSW for Linux (tested with version 2.24)
  • Linux OS (tested with Ubuntu 22.04.3 LTS, Kernel 6.5)
  • Dependencies: cmake, GCC and G++ 12, Python 3, pip, Git
  • Python package as defined in requirements.txt

Build

To build the code, run the following commands in the top-level directory:

  1. Configure the build files and put them in folder cmake-build-release: 
    cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=gcc-12 -DCMAKE_CXX_COMPILER=g++-12 -DSGX_MODE=PreRelease -S . -B cmake-build-release
    To show a list of available targets use the following command: 
    cmake --build cmake-build-release --target help
  2. Build all benchmarks using the configuration files in the folder cmake-build-release 
    cmake --build cmake-build-release --parallel --target all

Run

Each subproject inside the microbenchmarks folder is compiled into its own binary available under cmake-build-release/microbenchmarks/[benchmark_name] after build. They all use gflags and a --help flag is provided. For ease of use, we created shell scripts that run an experiment 10 times and store the output in a CSV file. They are automatically copied to the binary directory and are named with an integer followed by a short experiment name. For example, cmake-build-release/microbenchmarks/SimdScanMulti contains the experiment scripts 0_single_thread.sh and 1_scale_up.sh. The experiment scripts expect that the working directory is their local directory.

In order to plot the results. Copy the result CSV files to the corresponding results/data directory and run results/plot.py. Like the experiment scripts, the python scripts expect that they are executed in their directory.

Example

To reproduce the column scan benchmarks, first switch to the binary directory after building:

cd cmake-build-release/microbenchmarks/SimdScanMulti

Then run the experiment scripts:

./0_single_thread.sh &&\
./1_scale_up.sh &&\
./2_write_rate.sh &&\
./3_numa.sh

Then move/copy the result files to the source directory microbenchmarks/SimdScanMulti/results/data:

mv single-thread.csv scale-up.csv write-rate.csv cross-numa.csv ../../../microbenchmarks/SimdScanMulti/results/data

Finally, plot the results:

cd ../../../microbenchmarks/SimdScanMulti/results
python3 plot.py

The result figures are stored in img. The other results from RadixPartitioning and WriteBench can be reproduced with the same process.

Exception

To reproduce the SIMD scan experiments with 16-bit input values, checkout the branch 16-bit and run 0_single_thread.sh single-thread-16bit.csv.