Energy Aware Computing in Cooperative Wireless Networks

PhD: Anders Brødløs Olsen

 Casestory

1. Purpose
For next generation (4G) mobile communication systems issues related to resource optimization is of great importance. This need is manifold; integration of multiple functional devices, support of multiple communication standards, multimedia applications, support of adaptive transmission schemes, new envisioned services, and so on. The consequence is increased computational complexity, where computational enhancement typically is obtained by faster/larger platform architectures (CPU’s with higher clock frequency, more memory) with the penalty of increasing energy consumption. For handheld portable systems resources are limited; both from a money perspective, but also energy wise. It is generally accepted that energy consumption is one of the most significant design constraint for handheld mobile systems. The main reason is the lacking improvement in battery technology, making the gap between required and obtainable battery capacity constantly wider.

2. Background
4G developments are driven by concepts like; anywhere and anytime, seamless access, adaptive air interfaces, adaptive quality of services, flexibility, and efficiency. Recently, concepts of cooperation are introduced. Cooperative networks are a type of network, where unselfish assignments are carried out on the various terminals. Cooperative principles are mostly seen for transmission schemes, where a group of cooperative network notes are used to make a higher degree of diversity. By working together group and individual overall QoS can be improved, imposing that a willingness to spend individual resources for group QoS enhancements is present, but obviously also an individual average gain is important. Utilizing the cooperative concept the subject of “Energy aware task allocation for cooperative wireless networks” is investigated, also illustrated in Figure 1.

Figure 1. Scenario of Cooperative wireless network notes.

Abstracting a cooperative network into a structure of processing units (the individual terminals) connected by a wireless network technology (a short range wireless network) a "traditional" energy aware task scheduling approach is assumed. Obviously, energy conservation for both the “processing units” and the interconnecting “short range technology” is essential. For processing units the method of Dynamic Voltage Scaling (DVS) is proven to be a very efficient method. Power dissipation of CMOS together with knowledge of software (task-set) is used to scale the “speed” of the processing units. CMOS power dissipation is a quadratic function on supply voltage, so reductions will cause large power reductions. However, voltage supply is also inverse related to the obtainable clock frequency, making an inherent increase of voltage supply when the frequency is increased. Such performance scaling technologies are however not present in wireless communication technologies (at least at present time). Anyhow, more traditional power management systems are used, placing the various components in power down modes, whenever they are inactive.

3. Perspective
As mentioned, the overall concept is energy aware task allocation among cooperative wireless network notes, where such are abstracted into a system of distributed (or parallel) processing units connected by wireless communication facilities, as illustrated in Figure 2.

Figure 2. Cooperative note abstraction into a distributed system.

In the project the following focus subjects are considered:

• Conceptual understanding of the proposed scheme of “energy aware task allocation in cooperative networks”
• Modeling (especially from a energy perspective) of the various system components
• Task-set allocation/scheduling mechanisms, taking energy profiles of processing unit and communication facilities into account.
• Designing a simulation environment, where energy models and scheduling mechanisms can be evaluated.

Overall is the fact of slow operating processing unit using less energy than a single fast one utilized. With the addition that task migration is costly and must be careful considered, using run-time scheduling mechanisms. Our previous conceptual work has shown that cooperative terminals, for an ideal case, will have an energy gain of approximately three times, compared to identical workload executed on a single processor. In Figure 3 an illustrative example of our proposed scheme is illustrated, showing the energy benefits.

Figure 3. Two cooperative terminals, illustrating the energy benefits of cooperative execution.