CAREER: OS-Managed Remote Procedure Call for Datacenter Applications

Remote procedure call (RPC) allows developers to build networked applications using a simple and familiar programming model, and it has become a fundamental building block for cloud applications today. Increasingly, application and service operations teams need a high degree of visibility and control over the flow of RPCs, including monitoring specific types of RPCs, and rate limiting to meet application-specific performance goals. This project aims to explore a new RPC system architecture that allows for efficient and flexible control of RPC traffic. The key insight is to rearchitect RPC as a managed service, decoupling RPC logic from individual applications. Applications specify type information to the RPC service, and the RPC service executes RPC policies and arbitrates resource use, and then marshals data customized to the underlying network hardware capabilities. The project entails four thrusts: (1) a safe cross-domain RPC memory management design, (2) a system design for efficient RPC policy support, (3) a system design for RPC live upgradability, and (4) techniques to use remote direct memory access hardware effectively in RPC. The new RPC architecture will improve the performance and manageability of cloud applications and reduce the cost of operating cloud applications. This project will develop new educational tools, including programming assignments and an RPC visualizer, to introduce a broader community of students to the standard practices of using RPC to construct distributed applications. This project will serve as a platform to engage undergraduate students in networking research. Software artifacts, papers, benchmarks, and tutorials developed as part of this project will be released on the following website https://mrpc.cs.duke.edu. This website will be regularly maintained for the next ten years. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.