As new astronomy projects choose interferometry to improve angular resolution and to minimize costs, preparing
and optimizing schedules for an antenna array becomes an increasingly critical task. This problem shares
similarities with the job-shop problem, which is known to be a NP-hard problem, making a complete approach
infeasible. In the case of ALMA, 18000 projects per season are expected, and the best schedule must be found
in the order of minutes.
The problem imposes severe difficulties: the large domain of observation projects to be taken into account; a
complex objective function, composed of several abstract, environmental, and hardware constraints; the number
of restrictions imposed and the dynamic nature of the problem, as weather is an ever-changing variable. A
solution can benefit from the use of High-Performance Computing for the final implementation to be deployed,
but also for the development process.
Our research group proposes the use of both metaheuristic search and statistical learning algorithms, in order
to create schedules in a reasonable time. How these techniques will be applied is yet to be determined as part of
the ongoing research. Several algorithms need to be implemented, tested and evaluated by the team.
This work presents the methodology proposed to lead the development of the scheduler. The basic functionality
is encapsulated into software components implemented on parallel architectures. These components
expose a domain-level interface to the researchers, enabling then to develop early prototypes for evaluating and
comparing their proposed techniques.