Explanation
Why use Ilúvatar for research?
Ilúvatar provides the features and guarantees we find to be necessary to conduct high-quality, reproducible, and open research.
- Load generation a first-class citizen of the system, making results easier to capture and analyze.
- Fast, predictable, jitter-resistant function execution for reliable and repeatable experimentation.
- An architecture designed to allow testing on a single worker node, or a cluster of load-balanced nodes.
- The ability to run a load scenario on a live system or as a simulation, using the same codebase.
- An extendible FaaS platform built for an open-source research community.
Details about the implementation can be found in the paper below. How to use the various features and jump-starting your research can be found in the Ilúvatar documentation.
A popular open-source platform used in research is OpenWhisk, but we have found it has high overheads under notable load, caused by a variety of factors. The scalability of OpenWhisk vs our Ilúvatar can be seen here.
We can ensure a constant 1-3 ms overhead on invocations at significant load, whereas OpenWhisk sees high and variable overheads, on the same hardware. It also does not have the research-first features implemented in Ilúvatar.
See Ilúvatar in action for research right here
More documentation / Questions
Ilúvatar supports a large variety of customization in configuration and setup, and methods of load generation and experimentation. It is easy to create custom functions that run new Python programs, information is in this document. Detailed documentation can be found closer to the code.
If you have questions, want help working with Ilúvatar, or anything else, we have a Discord server. Definitely gives faster answers than opening an issue for a question.
Why 'Ilúvatar'?
The name 'Ilúvatar' comes from J.R.R. Tolkien's mythology, published in The Silmarillion, a history of the world prior to the events of his Hobbit and Lord of the Rings books. Ilúvatar is the creator of the world, and orchestrated its form and development. We don't see our platform as being the ultimate, final, or conclusive FaaS platform. But as a FaaS platform, it controls and directs where and how Serverless functions are created and executed, across a possibly great variety of heterogeneity and network distance. Thus the inspiration for its name.