Design Space Exploration for Run-Time Management of a Reconfigurable System for Video Streaming - Multi-objective Design Space Exploration of Multiprocessor SoC Architectures - page 189

Chapter 9
Design Space Exploration for Run-Time
Management of a Reconfigurable System
for Video Streaming

Giovanni Mariani, Chantal Ykman-Couvreur, Prabhat Avasare,
Geert Vanmeerbeeck, Gianluca Palermo, Cristina Silvano,
and Vittorio Zaccaria

Abstract This Chapter reports a case study of Design Space Exploration for sup-
porting Run-time Resource Management (RRM). In particular the management of
system resources for an MPSoC dedicated to multiple MPEG4 encoding is addressed
in the context of an Automotive Cognitive Safety System (ACSS). The run-time man-
agement problem is defined as the minimization of the platform power consumption
under resource and Quality of Service (QoS) constraints.
The Chapter provides an insight of both, design-time and run-time aspects of
the problem. During the preliminary design-time Design Space Exploration (DSE)
phase, the best configurations of run-time tunable parameters are statically identified
for providing the best trade-offs in terms of run-time costs and application QoS. To
speed up the optimization process without reducing the quality of final results, a
multi-simulator framework is used for modeling platform performance.
At run-time, the RRM exploits the design-time DSE results for deciding an oper-
ating configuration to be loaded for each MPEG4 encoder. This operation is carried
out dynamically, by following the QoS requirements of the specific use-case.

9.1 Introduction

The amount of resources available on MPSoC platforms is growing following the
Moore's law. To fully exploit the potential capabilities of future MPSoC, program-
mers need to split applications in multiple threads in such a way that these can be
allocated to different processors to be executed concurrently. There is not a unique
way to parallelize an application but different parallel versions of the same appli-
cation can be analyzed and used to trade off the application performance and the
number of threads in the application. If every thread needs a dedicated processor
to be executed, the availability of different parallelized versions can provide also

G. Mariani ( )
ALaRI, University of Lugano, Switzerland
e-mail: [e-mail]