Alumet provides a generic measurement pipeline with three steps: poll measurement sources, transform the data, and write the result. It is designed to be able to ingest metrics from various sources without redundant work. Supported sources include RAPL domains, Nvidia's NVML, and Jetson INA sensors. The list of supported devices will quickly grow over time, thanks to the next feature of Alumet.
Unlike existing tools, Alumet is "adaptive" in two ways:
This new tool is flexible, yet efficient and relatively small. The measurement pipeline is asynchronous, powered by the Tokio Rust library. Plugins are written as static or dynamic libraries, with an emphasis on efficiency and performance on hot paths.
We have carefully selected the most efficient and most resilient low-level measurement mechanisms, in order to minimize the overhead of the tool while keeping it maintainable. Read more about our comparative analysis of energy measurement techniques on x86 CPUs (using RAPL at ~1000Hz on Intel and AMD processors). We are also working on HPC-specific deployment schemes.
Teams at Eviden (part of Atos group), the LIG (Laboratoire d'Informatique de Grenoble) and the Edge Intelligence research program of the MIAI institute are involved in the Alumet project.
The source code will be released under an open-source license, but we are not ready yet. In particular, we need to write more documentation and to polish the plugin API.
If you are interested, you can register with the form below. We look forward to working with the community!