- A sample CloudLab parameter set consisting of 4 d430 nodes (k8s profile, with containerd as the container runtime) can be created using this link. Clicking the link allows a user to set up a 4-node Kubernetes cluster for running the experiments. Once all the nodes become
Readyand startup services haveFinished, SSH intonode-0and run Steps 0 through 2 above to evaluate the artifact.
- Clone artifact repository with:
git clone https://github.com/etclab/mazu-artifact.git - All scripts are run from inside the project root:
cd mazu-artifact
- First run
./scripts/bootstrap.shto install Go and Docker, and log in to Docker Hub.- Docker Hub login is used to push the built Istio images. Use
--skip-docker-loginto skip the login. - If you get a
permission deniederror with Docker, either runnewgrp dockeror log out and log back in.
- Docker Hub login is used to push the built Istio images. Use
- (Note: the next step can be skipped, as the required images have already been built and are publicly available under the Docker user
atosh502. Use the steps below to build the images from source and host them under a different Docker user's registry.) - To build the images and host them on Docker Hub:
DOCKER_USER=<user-id> ./scripts/setup-system.sh configure build-envoy build-istio- Duration: ~30 mins, Size: ~50 GB.
- Run
./scripts/benchmark.sh --setupto clone the required source files and benchmark scripts into./workspace. - Each figure (and the Section 7.4 results) can be reproduced in one of two modes:
--plot-only: plot the figures from the paper's data without running the experiment. Outputs are written to./plots/<name>/outputs/paper.--run: run a small-scale version of the experiment and plot its results. Outputs are written to./plots/<name>/outputs/run.- (Note:
./scripts/benchmark.shonly runs a scaled-down version of the benchmarks, as the complete runs (repeated 10 times) take ~48 hours.)
| Result | --plot-only (paper data) |
--run (small-scale run) |
Output files |
|---|---|---|---|
| Figure 2: crypto scheme microbenchmarks | ./scripts/benchmark.sh --plot-only --fig2 |
./scripts/benchmark.sh --run --fig2 |
fig2-crypto-scheme-microbenchmarks.pdf |
| Figure 3: RBE microbenchmarks | ./scripts/benchmark.sh --plot-only --fig3 |
./scripts/benchmark.sh --run --fig3 |
fig3-rbe-microbenchmarks.pdf |
| Figure 6: SocialNetwork latency, ready pods, CPU and memory | ./scripts/benchmark.sh --plot-only --fig6 |
./scripts/benchmark.sh --run --fig6 |
fig6a-end-to-end-latency.pdf, fig6b-total-ready-pods.pdf, fig6c-cpu-utilization.pdf, fig6d-memory-utilization.pdf |
| Figure 7: replica scaling p99 latency, CPU and memory | ./scripts/benchmark.sh --plot-only --fig7 |
./scripts/benchmark.sh --run --fig7 |
fig7a-p99-latency.pdf, fig7b-mazu-p99-latency-by-scale.pdf, fig7c-cpu-utilization.pdf, fig7d-memory-utilization.pdf |
| Figure 8: cold path latency, CPU and memory | ./scripts/benchmark.sh --plot-only --fig8 |
./scripts/benchmark.sh --run --fig8 |
fig8a-e2e-latency.pdf, fig8b-latency-breakdown-for-fortio-server.pdf, fig8c-cpu-utilization.pdf, fig8d-memory-utilization.pdf |
| Section 7.4: TPM microbenchmarks (printed to the terminal) | ./scripts/benchmark.sh --plot-only --sec7.4 |
./scripts/benchmark.sh --run --sec7.4 |
swtpm-bench.txt |