Low energy and high performance scheduling on scalable computing systems

With the fast development of supercomputers, energy
consumption by large scale computer systems
has become a major concern. How to reduce energy
consumption is now a critical issue in designing
high-performance computing systems. Moreover,
reducing energy consumption for high-performance
computing can bring various benefits such as, reduce
monetary operating costs, increase system reliability,
and reduction of environmental impacts. Therefore,
in this paper we address the problem of scheduling
precedence-constrained parallel applications on heterogeneous
scalable computing systems with the objectives
of minimizing finish time and reduce energy
consumption. We provide a scheduling algorithm
based on the best-effort idea that adopts dynamic
voltage scaling (DVS) to reduce energy consumption.
That is, the algorithm firstly looks for nearoptimal
solutions employing a list-based scheduling
algorithm to find the minimum finish time (besteffort).
Then, a fast random local search algorithm
that exploits voltage scaling is used to reduce the
energy consumption of the generated schedule without
any performance degradation. Simulation results
on structured graphs representing real-world applications
emphasize the interest of the proposed approach.