Large scale datacenters manage applications as black boxes. Most of the time, they assume that application behavior is not linked to the state of the underlying hardware. When an applications runs on a hot processors, it can be slowed down arbitrary by the processor as it tries to protect itself.
The goal of this internship is to evaluate the impact of temperature on the speed of the code, the impact of the execution of the code on temperature, and the possibility to reduce the frequency of the processor to cool down the processor at key points to cool down the processor (and thus speed up the application)
Large scale datacenters manage applications as black boxes. Most of the time, they assume that applications have no cross impact. When multiple applications are using the memory, their speed is reduces because of the bottleneck of the memory bus. In the other direction, two applications on the same core might not go at half the speed each: if one uses only floting point operations, while the other only memory access for example.
Topic Avoiding the ecological catastrophe will require an joined effort from every actor in the society – the ICT industry included. We postulate that some environmental-aware individuals are willing to reflect upon and reduce the footprint associated to their usage of new technologies. Similarly to the Low-tech Magazine[1], a solar-powered and very lightweight website, this internship will study an off-grid “sufficient”[2] data center in which a part of the users accepts to delay, degrade or even cancel the execution of their tasks to reduce the overall footprint of the infrastructure
The position is offered in the framework of the French ANR-funded ENERGUMEN project and will take place at IRIT laboratory in Toulouse, France. This project aims at proposing and evaluating new scheduling heuristics for malleable/reconfigurable HPC tasks (i.e. able to change the number of resources at runtime), taking into account computing requirements but also data movement that occurs during reconfiguration. We intend to study bi-objective problems using simulation, optimizing both consumed energy and a performance criterion, e.
