NEXTGenIO @ SC16: Exascale I/O: Challenges, Innovations and Solutions

Traditionally, the HPC field has treated I/O like an afterthought – highest compute performance and efficiency took precedence for developing systems and applications. With applications having become more diverse and complex and handling huge data amounts, HPC systems rapidly growing in size, and the drive to minimize energy use, this has fundamentally changed: providing highly scalable and efficient parallel I/O is now seen as a major challenge for HPC systems and system software (SW), and a prerequisite for achieving Exascale levels of delivered performance. At the same time, progress in network and memory/storage technology provides new opportunities to address this challenge. The session unites leading HPC experts to discuss their analysis of I/O related challenges that HPC and the rapidly emerging field of high performance data analytics (HPDA) face, present potential hardware (HW) and SW solutions for dealing with the extreme volumes of data involved in HPC and HPDA in a scalable and highly efficient way, and examine how emerging technologies (like non-volatile storage class memory) can best impact. The scope encompasses system architecture and components, the I/O system SW stack and APIs for HPC and HPDA applications. A panel-style discussion with the audience will contribute to a consolidated view of the field and its evolution. The presenters will represent significant I/O development efforts in Europe (DEEP-ER, NEXTGenIO and SAGE projects), and world-wide (Intel, Seagate).

Full details of the workshop schedule, and slides from each of the talks, are below.