README.md

Elastic Scaling Micro-Benchmark (Fig.12)

In this experiment we measure the benefits of elastically scaling-up OpenMP applications to benefit from idle resources. We run a pipe-lined algorithm on a matrix with a varying number of threads, and at 50% of execution we scale-up to the maximum number of available threads. This plot is a best-case scenario for the benefits we can get by elastically scaling-up.

Granny

First, start the AKS cluster by running:

inv cluster.provision --vm Standard_D8_v5 --nodes 2 cluster.credentials

Second, deploy the Granny cluster:

faasmctl deploy.k8s --workers=1

Third, upload the WASM file:

inv elastic.wasm.upload

and run the experiment with:

# Without elastic scaling
inv elastic.run

# With elastic scaling
inv elastic.run --elastic

Plot

You may now plot the results using:

inv elastic.plot

the plot will be available in /plots/elastic/elastic_speedup.pdf, we also include it below:

Clean-Up

Finally, delete the Granny cluster:

faasmctl delete

and the AKS cluster:

inv cluster.delete