An approach for dynamic and leakage energy reduction via combined supply
voltage scaling and body biasing in real-time multiprocessor systems was
presented. Discrete voltage modes and intra-task scaling are also
considered. The mapping and scheduling of the task sets are already
given. This approach achieves energy efficiency by exploiting dynamic
slack, which results at runtime, for example when the tasks do not
execute their worst case number of clock cycles. Dynamic slack is
exploited by using online voltage reduction techniques. Since these
algorithms are executed online, after each of the tasks finishes, they
must have a low complexity. Moreover the energy and time overhead for
changing the supply and body bias voltage is also considered. First, in
the offline part of the approach, an algorithm computes for each task a
LUT (look-up table). This LUT contains a number of possible start times
and the corresponding voltage mode setting. During run-time, the online
algorithm readjusts the voltage settings for the task by using the
information in the LUT and a low overhead online calculation.

Réalisation Axome - Création de sites Internet