Home > TERATEC FORUM > Workshop 7

Is IO the next major frontier for Exascale?
Chaired by Marie-Christine SAWLEY, INTEL and Hervé OHEIX, SEAGATE

Exaflopic machines will be operational around 2020. Until then, the workloads will be relying on a growing usage of data –simulated, experimental, observational- during computation; some tasks will be even centered exclusively upon performing very large data analytics tasks. As the requirements become more stringent, the careful choice of IO solutions will be a pivotal part of the Exascale capable systems.

Data traffic and workflows sustaining simulation, computation and data analytics are steadily growing in complexity, volume and requirements. Current IO technologies used in HPC until now will evolve, only new innovative solutions will be able to address the scaling needs.

In this workshop, we explored some innovative ideas about efficient access to data  for Exascale,  such as object storage; show use cases on how to efficiently run Hadoop jobs on the top of a Lustre system, or present the concept of a software layer  combining integrating more closely the capacities of SSD with a distributed parallel file architecture.