Record:   Prev Next
作者 Raghu, Arvind Kandhalu
書名 Energy-Aware Resource Management of Real-Time Mobile Systems
國際標準書號 9781303441714
book jacket
說明 227 p
附註 Source: Dissertation Abstracts International, Volume: 74-12(E), Section: B
Adviser: Raj Rajkumar
Thesis (Ph.D.)--Carnegie Mellon University, 2013
The number and variety of mobile systems have been rapidly growing. This trend in computing has shown a shift towards highly-connected multi-core systems. We see this shift in platforms ranging from smartphones and tablets to avionic and satellite systems. Across this wide design space, there is a growing need for energy-aware management of the system resources due to the direct impact of limited energy supply and heat dissipation. Furthermore, the real-time aspects of certain mobile systems impose timing constraints wherein the application tasks have to be completed within a given deadline. In this dissertation, we study the problem of efficiently managing the processor and networking resources of energy-limited, real-time mobile systems while meeting application timeliness constraints
For multi-core processors, we provide: (i) energy-efficient task-to-core allocation schemes and (ii) run-time operating frequency adaptation to application modes. Given the convex relationship between processor power consumption and operating frequency, a load-balancing approach has been widely adopted. We show that such an approach for reducing power consumption may not always perform well, and that clustering compatible tasks could prove beneficial. We design and analyze new schemes that exploit task compatibility (with and without task-splitting) and perform better than existing schemes. Then, we describe our mode change protocol, that adapts the task allocation and processor operating frequency based on application modes
With respect to network resource management, we consider (i) static, (ii) predictable and (iii) unpredictable mobility scenarios. For static multi-hop networks, we develop a time-synchronized access protocol that allows for controlled bandwidth allocation and bounded end-to-end communication delays. For predictable mobility scenarios, we develop dynamic graphs to represent the time-varying yet predictable network topology and perform resource allocation while meeting application timeliness and energy constraints. For unpredictable mobility scenarios, the presence of links between nodes at any time instant cannot be guaranteed. We design and analyze a neighbor discovery protocol that guarantees a bounded discovery latency for a given energy constraint
School code: 0041
Host Item Dissertation Abstracts International 74-12B(E)
主題 Engineering, Computer
Alt Author Carnegie Mellon University. Electrical and Computer Engineeering
Record:   Prev Next