Enabling Secure and Efficient Sharing of Accelerators in Expeditionary Systems
Abstract
Hardware accelerators are heavily used in tactical operations and potentially with expeditionary forces due to their energy-efficiency and high-performance. Such accelerators vary from simple cryptographic accelerators, based on reconfigurable logic, GPUs used for image recognition and threat detection, and customized near memory accelerators. Accelerators provide orders of magnitudes speedups compared to conventional processors, while maintaining high energy efficiency. However, one major concern when used in expeditionary systems is how to better utilize such resources given the area, size, and power restrictions. Such accelerators need to be utilized and enable concurrent execution of several tasks to run on accelerators, which can also introduce security and safety (determinism) concerns given the nature of the applications run on such systems. In this project, we aim to dynamically manage the scarce hardware resources in expeditionary tactical systems. The focus of this project is hardware accelerators, such as GPU, FPGA and in memory accelerators. The goal is to allow expeditionary forces better utilize such accelerators and thus enable more operations and computations to occur concurrently, without sacrificing the strict SWaP, security and determinism requirements of expeditionary tactical systems. Our fundamental research project will lead to several innovations on how to securely, efficiently, and safely, maximize the resource efficiency of hardware accelerators in environments such as expeditionary systems. Key Intellectual Merits: To the best of our knowledge, this is the first work to consider both security and resource utilization in hardware accelerators. Moreover, our scheduling framework will allow consideration of security policies and timing guarantees of individual tasks. Moreover, to the best of our knowledge, this is the first work to explore multi-tasking in near-memory accelerators, and the first security work to explore the security aspects of such line of accelerators. Relevance to Navy: Navy s tactical systems and advanced platforms deployed with expeditionary forces, such as those of Marine Corps, are size, weight, area and power, restricted. With such stringent requirements, and the sensitive nature of workloads run on such systems, dynamic resource allocation securely and safely is a unique requirement for efficiently utilizing such limited resources. Moreover, the advances for using accelerators in defense-related workloads, such as threat detection, fast image recognition, and speeding up cryptographic functions, further motivate efficient utilization of them in warfare situations. We believe that in the future many accelerators, such as GPUs, FPGAs and UPMEM, will become more common in tactical systems, and thus understanding their security, safe concurrent execution, and performance predictability, is crucial for their adoption in an efficient and safe manner
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 20, 2021
- Source ID
- N000142112809
Entities
People
- Amro Awad
Organizations
- North Carolina State University
- Office of Naval Research
- United States Navy