| dc.description.abstract | In this thesis, we develop algorithms that make optimal use of frequency scaling
to schedule jobs with real??time requirements.
Dynamic voltage scaling is a technique used to reduce energy consumption in
wide variety of systems. Reducing supply voltage results in a lower processor clock
speed since the supply voltage has a proportional dependency on the clock speed of
the processing system.
In hard real??time systems, unduly reducing the speed of processor could result
in jobs missing their deadlines. The voltage scaling in such systems should therefore
take into consideration the deadline of jobs. This thesis will address two questions:
First, given a set of discrete frequency levels, we determine an energy-optimal sched-
ule of a given set of real-time jobs. We model the problem as a network flow graph
and use linear programming to solve the problem. The schedule can be used on
processors with discrete frequencies (like Transmeta Efficeon Processor and AMD
Turion 64 Processor).
Second, given a set of real??time jobs, we determine a set of optimal frequency
levels which minimizes the energy consumption while meeting all the timing con-
straints. This can be used to model variable-capacity facilities in operations re-
search, where the capacity of the facility can be controlled at a cost. | en |
