Abstract

RPC has been remarkably successful as a foundation for distributed applications because of its simple but powerful semantics. With no shared state between processes, RPC has produced highly efficient system implementations as well as flexibility and interoperability for applications. However, the lack of shared state also makes it difficult to build data-intensive applications directly on RPC, as all values must be copied between processes. Instead, these applications are often built on top of specialized distributed frameworks, such as Apache Spark for big data processing. These frameworks handle problems such as memory management on behalf of the application, but sacrifice interoperability as a result.

We argue that RPC itself should be extended with an immutable shared address space and first-class references [HotOS’21]. This preserves RPC’s original semantics while offering two key application benefits: 1) by allowing data-intensive applications to be built directly on RPC, we promote interoperability, and 2) by shifting functionality such as automatic memory management to a common distributed system, we can reduce duplicated work between specialized frameworks. A key challenge in this effort is ensuring fault tolerance without sacrificing performance or flexibility. In this talk, I will discuss our work on ownership, a method of guaranteeing transparent data recovery for an immutable shared address space [NSDI’21], as well as our ongoing work on persistence for stateful applications.

Bio

Stephanie is a final-year PhD candidate at UC Berkeley, advised by Ion Stoica. Her research focuses on the problem of designing and building a general-purpose distributed system. Towards this end, she is also a lead committer for the open-source Ray project and a software engineer at the startup Anyscale. She is a recipient of the UC Berkeley Chancellor’s Fellowship and was a participant of EECS Rising Stars in 2021. In addition, she holds a BS in computer science and math and an MEng in computer systems from MIT, where she was advised by Frans Kaashoek and Nickolai Zeldovich.