README.md

LULESH Experiment

This experiment is a single execution of the LULESH simulation using OpenMP.

Warning

LULESH does a lot of forking in a tight loop. This is highly inefficient in our current implementation, so the results are very bad until we do something about it. The first easy fix is caching scheduling results locally, avoiding a round-trip to the planner.

Start AKS cluster

In the experiment-base terminal, run:

(faasm-exp-base) inv cluster.provision --vm Standard_D8_v5 --nodes 1
(faasm-exp-base) inv cluster.credentials

Granny

Deploy the cluster:

(faasm-exp-faabric) faasmctl deploy.k8s --workers=1

Upload the WASM file:

(faasm-exp-faabric) inv lammps.wasm.upload

and run the experiment with:

(faasm-exp-faabric) inv lammps.run.granny -w compute -w network

To remove the cluster, run:

(faasm-exp-mpi) faasmctl delete

Native

Deploy the cluster:

inv lammps.native.deploy

And run:

inv lammps.run.native -w compute -w network

finally, delete the native cluster:

inv lammps.native.delete

Plot

To plot the results, you may run:

inv lammps.plot

which will generate a plot in ./plots/lammps/runtime.png, we also include it below:

Clean-Up

Remember to delete the cluster. From the experiment base terminal:

inv cluster.delete