Duke University
Department of Electrical and Computer Engineering
Investigation of Energy-efficient Task Scheduling for Real-Time Systems
Amil Patel
Advisor: Dr. Krishnendu Chakrabarty
A push for energy efficiency has been a predominant theme in electrical and computer engineering over the past few years. The motivation comes from an emergence in truly mobile computing (thanks to PDA’s and IEEE 802.11b wireless LAN standard) and embedded applications. The focus of my work has been embedded real-time systems. Real-time constraints create exciting new challenges in designing energy-efficient systems. A non-real-time environment such as a standard Windows operating system supports several user applications (processes) simultaneously (such as Winamp and Instant Messenger). The Operating System scheduler is responsible for juggling these different processes. The scheduler may immediately handle one application while ignoring others, depending on the scheme implemented. By contrast, in a real-time environment processes are characterized by deadlines. The operating system must guarantee that these processes finish before their respective deadline arrives. In applications such as robotics or nuclear power plant control, failing to complete a task by the deadline may have catastrophic consequences. However, the focus here is energy sensitive real-time systems, such as a remote sensor node running on battery power. Data acquisition and processing must be done in real-time as events in the environment unfold. The engineering challenge becomes balancing the tradeoffs between power consumption and real-time constraints.
The aim of the project is to determine the practicality of an energy-efficient real-time operating system by developing a prototype consisting of software and hardware components. This is accomplished by implementing the Low-energy Earliest Deadline First 1(LEDF) scheduling algorithm into the Embedded Configurable Operating System (ECOS) scheduler. ECOS is an open source project supported by Red Hat. ECOS was selected because of its highly configurable design and because of its small memory footprint, an essential requirement for embedded applications.
ECOS will be run on top of the AMD Athlon 4 2(Intel x386 architecture) with PowerNow! technology. PowerNow! is ideal for embedded and mobile applications because the CPU core voltage can be scaled dynamically by the operating system. Reducing core voltage yields a reduction in power consumption with a speed trade-off. LEDF attempts to select the minimum speed necessary such that a process is completed by its deadline.
The practicality of the prototype will be demonstrated by comparing power savings generated with LEDF with the predicted power savings generated from an LEDF simulator.
Sources