Collaborative Research: PPoSS: Planning: Scaling Secure Serverless Computing on Heterogeneous Datacenters

点击翻译按钮获取中文摘要

Cloud computing has been a dominant computing paradigm that enables many important computing capabilities including large-scale (big) data processing, artificial intelligence, and scientific discoveries. A recent evolution of cloud computing includes the move to serverless computing, which simplifies the deployment of computation while enabling better scaling and higher resource utilization. Meanwhile, datacenters, the backbone of cloud computing, increasingly include heterogeneous compute and memory resources. The move toward serverless computing and heterogeneous architecture of datacenters produces a gap that unless addressed, results in inefficient use of resources. The project seeks to address this gap in order to enable new applications and new functionalities to be provided in the cloud, at lower cost and higher security, providing platforms for the advancement of science, engineering, and commerce. Future datacenters will consist of heterogeneous compute and memory. Applications in the cloud are increasingly varied in their requirements, such as degree and granularity of parallelism; memory latency, capacity, and bandwidth requirements; and security and privacy requirements. This project investigates serverless computing as a promising programming model for heterogeneous platforms. Serverless platforms decouple system management from application execution: applications provide functions that manipulate data, and leave it to the platform to determine when the function should run, with what input data, and on what physical machine. Current platforms, such as AWS Lambda, Google Compute Functions or Azure Functions do not fully implement this vision, as they do not expose heterogeneous resources nor manage all resources automatically. This project explores novel abstractions for compute that extend serverless functions to better leverage unique hardware characteristics, and for memory to allow more automated leveraging of workload characteristics such as locality and compute intensity.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.