With increasing world-wide use of cloud-based computing resources it is more than ever necessary to have efficient, reliable, secured, and power-efficient datacenters. In order to cope with the versatile demand in social media, content streaming, banking, shopping, and high-performance computation, data centers must accommodate heterogeneous architectures to cope with the versatility in computing. FPGA devices have been recently introduced in cloud infrastructure to provide acceleration for critical applications in artificial intelligence, imaging, high-performance, and cryptography, with reduced power consumption. Unfortunately, current cloud providers are still assigning entire FPGAs to single customers, which leads to underutilization and increases in power consumption and costs. The primary reason is the lack of secured and stable approaches to enable the sharing of FPGA-resources among multiple cloud tenants, many of whom could be malicious.

This project aims at addressing the problem of multi-tenancy in FPGA-accelerated clouds and datacenters by providing the architecture and mechanisms for the efficient sharing of FPGA devices among multiple users. Specifically, the investigators will design and implement 1) hardware and software virtualization technologies for the integration of segments of the same hardware into cloud hypervisors, 2) isolation framework based on the NSA-developed Flux Advanced Security Kernel (FLASK) to extend domain separation from software threads and virtual machines down to provisioned hardware accelerators, and 3) temporal and spatial management approaches of the hardware resources, from the admission control to the execution on the FPGA device. The resulting cloud management framework will be made available to schools, colleges, communities and commercial entities to facilitate the adoption of heterogeneous cloud infrastructure that incorporates FPGAs. While the 20-node FPGA-accelerated cloud installed at the University of Florida will be used as testbed throughout the project, it will also serve as a vehicle to develop new classes and provide training to students, both on and off-campus. In this regard, the investigators will increase efforts to recruit and engage under-represented groups using existing resources such as the George Washington Carver Research program and the Florida-Caribbean Louis Stokes Regional Center of Excellence.

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 #
2007320
Program Officer
Erik Brunvand
Project Start
Project End
Budget Start
2020-10-01
Budget End
2023-09-30
Support Year
Fiscal Year
2020
Total Cost
$495,963
Indirect Cost
Name
University of Florida
Department
Type
DUNS #
City
Gainesville
State
FL
Country
United States
Zip Code
32611