Artifact Review Summary: Turbo: Effective Caching in Differentially-Private Databases
Artifact Details
Badges Awarded
Artifact Available | Artifact Functional | Results Reproduced |
Description of the Artifact
Turbo is a proposed caching system for linear query tasks in differentially-private databases. The artifact includes the source code and instructions to build docker containers for simulation experiments, which are hosted at Github.
Environment(s) Used for Testing
- Cloudlab (Ubuntu 20.04)
- Laptop with 16 cores and 32 RAM with Windows Subsystem Linux 2 and Docker Desktop
- PC with Intel(R) Xeon(R) E-2124G CPU, 64GB RAM, Ubuntu 18.04, and Linux kernel 5.4.0-150-generic
Step-By-Step Instructions to Exercise the Artifact
The reviewers followed the steps described in the README file in the repo https://github.com/columbia/turbo-artifact
How The Artifact Supports The Paper
The artifact is available on a public platform and functions properly.
The artifact reproduces all the results in the paper (Figure7a-7d, Figure 8a-8b, Figure 9a-9c, and Figure 10a-10c). While some minimal differences exist, the reproduced results are overall consistent with the results reported in the paper.
Additional Notes and Resources
We appreciate the author’s efforts in integrating all data-producing and figure-plotting work into one script, which is very easy to use. To improve the artifact, we encourage the authors to provide the basic hardware requirement for conducting all experiments in a reasonable time.