Files in this item

FilesDescriptionFormat

application/pdf

application/pdfPOTHUKUCHI-DISSERTATION-2020.pdf (9MB)
(no description provided)PDF

Description

Title:Intelligent systems for efficiency and security
Author(s):Pothukuchi, Raghavendra Pradyumna
Director of Research:Torrellas, Josep
Doctoral Committee Chair(s):Torrellas, Josep
Doctoral Committee Member(s):Abdelzaher, Tarek; Hwu, Wen-mei; Kale, Laxmikant (Sanjay) V.; Kim, Nam Sung; Voulgaris, Petros G.; Ansari, Amin; Greathouse, Joseph L.
Department / Program:Computer Science
Discipline:Computer Science
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:Ph.D.
Genre:Dissertation
Subject(s):resource management
computer architecture
computer systems
control theory
processors
performance management
power management
thermal management
energy efficiency
power side channels
security
machine learning
adaptive systems
robust control
MIMO
distributed resource management
heterogeneous computers
computer science
Abstract:As computing becomes ubiquitous and personalized, resources like energy, storage and time are becoming increasingly scarce and, at the same time, computing systems must deliver in multiple dimensions, such as high performance, quality of service, reliability, security and low power. Building such computers is hard, particularly when the operating environment is becoming more dynamic, and systems are becoming heterogeneous and distributed. Unfortunately, computers today manage resources with many ad hoc heuristics that are suboptimal, unsafe, and cannot be composed across the computer’s subsystems. Continuing this approach has severe consequences: underperforming systems, resource waste, information loss, and even life endangerment. This dissertation research develops computing systems which, through intelligent adaptation, deliver efficiency along multiple dimensions. The key idea is to manage computers with principled methods from formal control. It is with these methods that the multiple subsystems of a computer sense their environment and configure themselves to meet system-wide goals. To achieve the goal of intelligent systems, this dissertation makes a series of contributions, each building on the previous. First, it introduces the use of formal MIMO (Multiple Input Multiple Output) control for processors, to simultaneously optimize many goals like performance, power, and temperature. Second, it develops the Yukta control system, which uses coordinated formal controllers in different layers of the stack (hardware and operating system). Third, it uses robust control to develop a fast, globally coordinated and decentralized control framework called Tangram, for heterogeneous computers. Finally, it presents Maya, a defense against power side-channel attacks that uses formal control to reshape the power dissipated by a computer, confusing the attacker. The ideas in the dissertation have been demonstrated successfully with several prototypes, including one built along with AMD (Advanced Micro Devices, Inc.) engineers. These designs significantly outperformed the state of the art. The research in this dissertation brought formal control closer to computer architecture and has been well-received in both domains. It has the first application of full-fledged MIMO control for processors, the first use of robust control in computer systems, and the first application of formal control for side-channel defense. It makes a significant stride towards intelligent systems that are efficient, secure and reliable.
Issue Date:2020-07-17
Type:Thesis
URI:http://hdl.handle.net/2142/108504
Rights Information:2020 Raghavendra Pradyumna Pothukuchi
Date Available in IDEALS:2020-10-07
Date Deposited:2020-08


This item appears in the following Collection(s)

Item Statistics