CAREER: Harnessing Serverless Functions to Build Highly Elastic Cloud Storage Infrastructure

点击翻译按钮获取中文摘要

Crucial data-intensive applications such as big data analytics, artificial intelligence, and scientific computing exhibit highly dynamic and heterogeneous data access patterns. These applications require a large data storage system that is fast, elastic, cost-effective, and easy-to-use. Unfortunately, current cloud storage offerings are unable to satisfy all of these requirements: big data analytics applications can run up to 500X slower when they run as a cloud-native application using a cloud object store; provisioning more cloud storage resources does not solve the performance problem and storage over-provisioning leads to millions of dollars of extra cost. This project rethinks the design and implementation of next-generation, cloud-native storage infrastructures via a new approach. This approach is enabled by a novel take on the serverless computing paradigm -- serverless functions are elastic and truly pay-per-use, making them a fast and inexpensive storage medium. This project aims to build InfiniStore, a new cloud storage system, which elastically scales in response to constantly changing data access patterns by orchestrating the collective memory of serverless functions. Following an end-to-end, stacked approach, this project will develop an elastic and inexpensive memory store using serverless functions, techniques that provide strong consistency guarantees for data replicated in serverless functions, fault tolerance mechanisms to maximize the data availability in highly unreliable serverless environments, and a scalable serverless file metadata service to support file semantics.This project has the potential to redefine cloud storage pricing models -- InfiniStore enables memory-store-level pay-per-access that lets cloud users pay for the data they access. This project will enable, in a broad range of disciplines, the development of new, stateful, serverless applications that have not been previously possible. All artifacts will be open source to support the use and development both in academia and industry. This project has an integrated educational plan, which also harnesses serverless computing but uses it to develop a new serverless notebook cloud service, called InfiniCloud, for interdisciplinary education. InfiniCloud will allow educators and students to write and run serial or parallel Python programs at any scale, without needing to manage servers or clusters. Moreover, this project will use InfiniStore and InfiniCloud as a basis for various educational activities including curriculum development, diversity building, undergraduate engagement, and workshops.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.