Power and timing requirements are becoming more and more stringent as
applications move from less mobile devices to more mobile ones. As such, it is
important to optimize these applications as much as possible in order to provide the
best solution that is low power and low latency. Although there are many different
techniques to achieve a low power, low latency solution, this thesis focuses
specifically on low power scheduling at the behavioral level where resource-constrained
scheduling is the technique of choice since it directly considers the
resource limitations of mobile devices. Conventional resource-constrained
scheduling schemes are concerned with minimizing the latency or improving the
speed of an algorithm--represented by a data flow graph (DFG)--given a
limitation on resources. However, these conventional resource-constrained
scheduling schemes are no longer applicable since power has grown to be a major
issue, especially in mobile devices. Hence, the conventional resource-constrained
scheduling schemes gave way to current resource-constrained scheduling schemes
that utilize multiple voltages, which work to find a balance between speed and
power. These current multiple voltage schemes use various techniques to balance
and meet the speed and power requirements. But while they do a good job of
meeting these requirements, they fail to address a new issue that is beginning to
surface the number of memory registers needed. Therefore, to address this new
arising issue, this paper presents a novel resource-constrained scheduling scheme
that balances the speed, power, and register requirements. This algorithm is
compared to both a conventional resource-constrained scheduling scheme and a
current resource-constrained scheduling scheme with multiple voltages to show that
it performs better in finding a scheduling solution. Benchmark results show that,
on average, our algorithm has a better power savings while keeping the maximum
number of registers needed and the latency low compared to conventional resource-constrained
scheduling schemes and current resource-constrained scheduling
schemes utilizing just multiple voltages. / Graduation date: 2004
| Identifer | oai:union.ndltd.org:ORGSU/oai:ir.library.oregonstate.edu:1957/30730 |
| Date | 30 May 2003 |
| Creators | Lee, Chee |
| Contributors | Shiue, Wen-Tsong |
| Source Sets | Oregon State University |
| Language | en_US |
| Detected Language | English |
| Type | Thesis/Dissertation |
Page generated in 0.0024 seconds