Projekttitel: Multi-processor optimization using genetic algorithms
Kontaktperson: Jan Madsen,
bygning 322.221, jm@imm.dtu.dk
Beskrivelse: The advance in silicon technology which allows more than
400Mio transistors on a single chip makes it possible to integrate complete
systems consisting of multiple processors on a single chip (SoC: System-on-Chip).
These processors may be of different types, ranging from simple
micro-controllers over general purpose processors (CPU's) to specialized digital
signal processors (DSP's). Given such a heterogeneous multi-processor
architecture one of the key challenges facing a system designer, is how to map
the application, in terms of a description of multiple concurrent processes, on
to this architecture while optimizing a number of design metrics such as
performance, power consumption, memory utilization and size, reusability, and
flexibility.
The main objective of this project is to study the use of genetic algorithms to
solve the mapping problem. In order to restrict the problem size, we will focus
on a single design metric, i.e. performance. Further more, we will use a
simplified model of the application, i.e. a task graph representing a number of
tasks and their relations, and a simplified model of the architecture, i.e. a
graph presenting the different processors and their interconnect. An important
aspect is to incorporate the delay of transferring data over the interconnects,
when estimating
performance.
Forudsætninger: I- eller E-fagpakken samt 49142 Algoritmer og
Datastrukturer.