Application-specific hardware acceleration using Field-Programmable Gate Arrays (FPGAs) provide extremely high performance and power efficiency, but FPGA resources in currently available cloud platforms are used inefficiently because they are tied to physical host machines and cannot be allocated elastically, unlike general-purpose resources in the cloud such as CPU and system memory. This project, MintCloud, aims to design and implement a cloud platform for elastic allocation of FPGA resources, as well as a development and execution environment for distributed FPGA acceleration in such a cloud environment.

By elastically allocating FPGA resources, MintCloud brings the benefits of cloud architectures to high-performance FPGA-accelerated systems, including efficiency, manageability, and robustness. Cloud instances can easily obtain a large amount of FPGA resources, and the infrastructure is resilient to individual component failures. Furthermore, idle resources can be used efficiently by allocating them across the cloud where they are needed. For this goal, this project will develop components including secure access into remote FPGAs over the cloud network, a multi-FPGA programming abstraction and development environment, as well as an intelligent resource manager and allocator.

MintCloud provides a platform for industry and academia to rapidly develop and deploy complex, high-performance low-power applications at a low cost in the cloud, enabling the computationally intensive applications of the future. Furthermore, by lowering cost and improving manageability, MintCloud can deliver the benefits of efficient, FPGA-accelerated computation to resource-constrained entities and parties. The same benefits will also aid curriculum development efforts on the topic of FPGA-accelerated systems, by lowering the cost and barrier-of-entry of setting up and operating an educational environment.

All code, data, configuration information, simulation results, and other artifacts will be made openly available online until they are deemed no longer useful to outside entities. Depending on the type and size of the resources, they will be stored either in git repositories, long-term cloud storage, or as web pages. Links to all such repositories will be organized and listed in the project website, reachable at the following link: www.ics.uci.edu/~swjun/mintcloud.htm

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.

Agency
National Science Foundation (NSF)
Institute
Division of Computer and Network Systems (CNS)
Type
Standard Grant (Standard)
Application #
1908507
Program Officer
Erik Brunvand
Project Start
Project End
Budget Start
2019-10-01
Budget End
2022-09-30
Support Year
Fiscal Year
2019
Total Cost
$500,000
Indirect Cost
Name
University of California Irvine
Department
Type
DUNS #
City
Irvine
State
CA
Country
United States
Zip Code
92697