Over the past decade, the bulk synchronous processing (BSP) model proved highly effective for processing large amounts of data. Today, however, we are witnessing the emergence of a new class of applications — AI workloads. These applications exhibit new requirements, such as nested parallelism and highly heterogeneous computations.
In this talk, Ion Stoica, Professor at UC Berkeley, discusses how his team developed Ray, a distributed system that provides both task-parallel and actor abstractions. Ray is highly scalable, employing an in-memory storage system and a distributed scheduler. Ion discusses some design decisions as well as early experiences using Ray to implement a variety of applications.