Serverless computing platforms represent the fastest growing segment of cloud services, and are predicted to dominate cloud computing in the near future. These platforms run user-specified functions and automatically manage the underlying compute resources for the users. Serverless computing has the potential to impact fields as diverse as scientific computing, large-scale optimization, deep neural networks (DNNs), and video encoding in real-time due to its ease of management, inexpensiveness, massive scalability and flexibility in terms of compute power. In order to harness the full potential, however, several fundamental challenges must be addressed concerning several unique attributes of serverless computing, namely ephemeral machines, low memory footprint, heavy communication costs, and substantially different pricing policies. This project approaches these unique challenges by combining cutting-edge techniques from distributed computing with innovative concepts from coding theory, information theory, optimization, and game theory. The broader societal impact of this project is best appreciated by considering that nearly everyone, knowingly or unknowingly, benefits from the efficacy and ubiquity of cloud computing which has come to underly everything in modern digital life from web searches to online transactions. As serverless platforms are expected to dominate cloud computing in the near future, this project is expected to provide significant benefits by increasing efficiencies and reducing the user costs of cloud services.

This proposal takes a principled and foundational approach to minimizing latency and costs in serverless computing. This will lead to the development of theory and algorithms driven by theoretical principles which are informed by coding and information theory, resource allocation and optimization, randomized linear algebra, as well as mechanism design and game theory. The project consists of two major components: (a) Robust, efficient, and massively scalable distributed computing algorithms for latency minimization in serverless systems by integrating the power of coded computation into an optimization framework; and (b) Optimal pricing schemes for cost minimization, which leverage incentive mechanisms for pricing that maximize the total utility across customers, and enable cloud service providers to achieve favorable trade-offs between quality-of-service and revenue.

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.

Detailed Award Information

Award Information:
Title: CIF: Small: Foundations of Serverless Computing: Optimizing Latency and Utility
ID: 2007669
Effective Date: 10/01/2020
Expiration Date: 09/30/2023
Amount: $500,000

Institution Information:
Name: University of California-Berkeley
City: Berkeley
State: CA
Country: United States
MSI: Other Institution

Investigator Information:
Role Code: Principal Investigator
Name: Kannan Ramchandran
Email Address:

Organization Information:
Directorate: 4900
Division: NSF

Program Information:
Code: 4900
Text: Comm & Information Foundations