In the era of Climate Change, being able to control the power consumption of systems such as a supercomputer is a huge asset. As the High Performance Computing (HPC) community is getting closer to develop exascale systems, it becomes more important to use those resources effectively. One approach to do so is to make a program able to dynamically spread its computation to neighbour processors. There has been efforts to build the required tools to apply this concept in the HPC domain. In particular, past projects have used a resource manager and MPI library combination to add resource-elasticity support for HPC applications. It is possible to use the tools named above to try to keep the power consumption of the system inside some boundaries. More resource could be assigned to more power efficient applications to decrease the whole power consumption. Modifications to the resource manager to achieved the former are proposed in this work. In particular, the resource manager is extended to predict the power consumption of the running applications. Additionally, a heuristic that tries to allocate the nodes to the different applications, while keeping the power consumption of the system inside some boundaries, is developed. Finally, a phase oriented programming model, called the Elastic Phase Oriented Programming model (EPOP), is extended to measure the power consumed by each job.      «

In the era of Climate Change, being able to control the power consumption of systems such as a supercomputer is a huge asset. As the High Performance Computing (HPC) community is getting closer to develop exascale systems, it becomes more important to use those resources effectively. One approach to do so is to make a program able to dynamically spread its computation to neighbour processors. There has been efforts to build the required tools to apply this concept in the HPC domain. In particula...     »

In the era of Climate Change, being able to control the power consumption of systems such as a supercomputer is a huge asset. As the High Performance Computing (HPC) community is getting closer to develop exascale systems, it becomes more important to use those resources effectively. One approach to do so is to make a program able to dynamically spread its computation to neighbour processors. There has been efforts to build the required tools to apply this concept in the HPC domain. In particular, past projects have used a resource manager and MPI library combination to add resource-elasticity support for HPC applications. It is possible to use the tools named above to try to keep the power consumption of the system inside some boundaries. More resource could be assigned to more power efficient applications to decrease the whole power consumption. Modifications to the resource manager to achieved the former are proposed in this work. In particular, the resource manager is extended to predict the power consumption of the running applications. Additionally, a heuristic that tries to allocate the nodes to the different applications, while keeping the power consumption of the system inside some boundaries, is developed. Finally, a phase oriented programming model, called the Elastic Phase Oriented Programming model (EPOP), is extended to measure the power consumed by each job.      «

In the era of Climate Change, being able to control the power consumption of systems such as a supercomputer is a huge asset. As the High Performance Computing (HPC) community is getting closer to develop exascale systems, it becomes more important to use those resources effectively. One approach to do so is to make a program able to dynamically spread its computation to neighbour processors. There has been efforts to build the required tools to apply this concept in the HPC domain. In particula...     »