DOI: 10.1007/s10766-007-0032-7
International Journal of Parallel Programming (© 2007)
A Fast and AccurateTechnique
for Mapping Parallel Applications
on Stream-Oriented MPSoC Platforms
with Communication Awareness
Martino Ruggiero,
1
Alessio Guerri,
1
Davide Bertozzi,
2,3
Michela Milano
1
and Luca Benini
1
Received: 22 November 2006 / Accepted: 17 January 2007
The problem of allocating and scheduling precedence-constrained tasks on
the processors of a distributed real-time system is NP-hard. As such, it has
been traditionally tackled by means of heuristics, which provide only approx-
imate or near-optimal solutions. This paper proposes a complete allocation
and scheduling framework, and deploys an MPSoC virtual platform to val-
idate the accuracy of modelling assumptions. The optimizer implements an
efficient and exact approach to the mapping problem based on a decomposi-
tion strategy. The allocation subproblem is solved through Integer Program-
ming (IP) while the scheduling one through Constraint Programming (CP).
The two solvers interact by means of an iterative procedure which has been
proven to converge to the optimal solution. Experimental results show sig-
nificant speed-ups w.r.t. pure IP and CP exact solution strategies as well as
high accuracy with respect to cycle-accurate functional simulation. Two case
studies further demonstrate the practical viability of our framework for real-
life applications.
KEY WORDS: MPSoCs; allocation; scheduling; Integer Programming;
Constraint Programming.
1
University of Bologna, DEIS, Viale Risorgimento 2, Bologna 40136, Italy.
E-mails: {mruggiero; aguerri; mmilano; lbenini}@deis.unibo.it
2
University of Ferrara, via Saragat 1, Ferrara 40132, Italy.
E-mail: dbertozzi@ing.unife.it
3
To whom correspondence should be addressed. E-mail: dbertozzi@ing.unife.it
© 2007 Springer Science+Business Media, LLC