Hacker News Reader

What tutorial/paper/github repo/demo do you think shows the heart of why this is the best thing? I'm just not sure what this does better than some of the other strategies that are out there that are hadoop/HDFS based, S3/lambda based, from the privs/resources style k8s/cgroups/containers, or from the task based parallelisms that you can see in Apache Airflow, Jenkins, and other job runners.

View on HN

If you have a tolerance for video, I'd recommend the bootcamp [1]. Otherwise the tutorial [2].

I'm not familiar with all of the systems you mentioned, but I'd think of it more as a Spark or Dask or MPI competitor and less as a job scheduler. We have tasks, the runtime automatically identifies dependencies between tasks based on what data they use, and the support for describing different partitionings of data is very expressive.

[1]: https://legion.stanford.edu/bootcamp2017/ (updated source code here: https://github.com/elliottslaughter/regent-tutorial)

[2]: https://legion.stanford.edu/tutorial/ (there's also a parallel version in Regent which I think is easier to read, but has less descriptive text: http://regent-lang.org/tutorial/)

Reply on news.ycombinator.com