Repository files navigation

Originally, these were detailed experiments done over the simdjson code base for the paperParsing gigabytes of JSON per second.

We use a simple Makefile to execute the docker scripts. If you have a proper configured Linux system, you should be able to generate all experimental results with a simple command:

make

The code for the experiments is in theexperiments directory. Each experiment is in its own repository with a docker file describing the container.

The experimental results are in theresult directory. They are organized first by node name (to distinguish results from different systems). When you typemake, it creates a new subdirectory and fills it with experimental results.

There is a distinct directory calledanalysis that contains the scripts processing the resutlts from theresult directory. The analysis scripts have dependencies: python, pandas, gnuplot. Doing the analysis is not fully automated, by design: a human being needs to look at the number.

Software:

• Linux
• Docker (you need privileged access).

Hardware:

• AVX2-compatible processing (Intel, AMD).
• We check that hyperthreading is disabled.

You need toinstall docker, make sure it is running, and make yourself a priviliged user which may involveusermod -G -a docker myusername or some similar command. You want to create a docker group, put yourself in the docker group and restarts docker.