README.md

dask-on-ray-blog

All benchmarks are run with the Ray cluster launcher on Amazon AWS. The included bash scripts will launch a cluster on EC2 and run the benchmarks in a remote screen session. You can check the status of the job by attaching to the screen session with

ray attach cluster.yaml --screen

Broadcast benchmark

Run:

bash run_broadcast.sh

This script should finish in <20 minutes. The output will be written to output-broadcast.csv and can be downloaded with:

ray rsync-down cluster.yaml /home/ubuntu/dask-benchmarks/output-broadcast.csv .

An example output is included at example-broadcast.csv.

Sort benchmark

The sort benchmark reads from S3, so you will first have to set a bucket for the data. Modify the run_sort.sh script to replace <s3_bucket> with the bucket name. You will also have to fill out the AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables in the same script with the appropriate values. Make sure that your AWS profile has the right permissions to access the S3 bucket!

Next, run:

bash run_sort.sh

This script can take several hours to finish completely since some of the larger datasets take some time to complete. To only run some of the trials, modify run_sort_trials.sh.

The output will be written to output.csv and can be downloaded with:

ray rsync-down cluster.yaml /home/ubuntu/dask-benchmarks/output-broadcast.csv .

An example output is included at example-sort.csv.