Skip to content Skip to navigation

NEXTGenIO at SC19

NEXTGenIO will again be well represented at SC19, the International Conference for High Performance Computing, Networking, Storage and Analysis, which will be held in Denver, Colorado (USA), on 17-22 November 2019.

The project partners will present a research paper, An Early Evaluation of Intel’s Optane DC Persistent Memory Module and Its Impact on High-Performance Scientific Applications, a tutorial on Practical Persistent Memory Programming, and a Birds-of-a-Feather sesssion on Multi-Level Memory and Storage for HPC, Data Analytics, and AI.

Tutorial: Practical Persistent Memory Programming

Monday, 18 November 2019 8:30am - 12pm, location 407.

Tutorial organised by Adrian Jackson (EPCC) and Javier Conejero (BSC).

Persistent memory, such as Intel's Optane DCPMM, is now available for use in systems and will be included in future exascale deployments such as the DoE Aurora system. This new form of memory requires both different programming approaches to exploit the persistent functionality and storage performance and redesign of some applications to benefit from the full performance of the hardware.

This tutorial aims to educate attendees on the persistent memory hardware currently available, the software methods to exploit such hardware, and the choices that users of systems and system designers have when deciding what persistent memory functionality and configurations to utilize.

The tutorial will provide hands-on experience on programming persistent memory along with a wealth of information on the hardware and software ecosystem and potential performance and functionality benefits. We will be using an HPC system that has compute nodes with Optane memory for the tutorial practicals.

Information about this tutorial on the SC19 webpage.

BoF: Multi-Level Memory and Storage for HPC, Data Analytics, and AI

Wednesday, 20 November 2019 5:15pm - 6:45pm, location 607.

Birds-of-a-Feather session led by Hans-Christian Hoppe (Intel), with Michèle Weiland (EPCC) and Kathryn Mohror (Lawrence Livermore National Laboratory)

This BoF investigates the opportunities arising from the progress in storage class memory (SCM) technology and the parallel rapid emergence of data-intensive application in the HPC context, which increasingly combine simulations with data analytics, AI, or graph analytics techniques. Topics include the benefit of SCM for applications, integration with the system architecture, SW interfaces, and, of course, results from proof of concept projects.

The session brings together technology providers, application and system SW developers, and system operators to engage in a discussion with the audience.

Information about this BoF on the SC19 webpage.

Research paper: An Early Evaluation of Intel’s Optane DC Persistent Memory Module and Its Impact on High-Performance Scientific Applications

Thursday, 21 November 2019 2pm - 2:30pm, location 405-406-407.

Research paper in the Improving Next-Generation Performance and Resilience session.

Authors: Michèle Weiland (EPCC), Holger Brunst (Technical University Dresden), Tiago Quintino (ECMWF), Nick Johnson (EPCC), Christian Herold (Technical University Dresden), Antonino Bonanni (ECMWF), Simon Smart (ECMWF), Olivier Iffrig (ECMWF), Adrian Jackson (EPCC), Mark Parsons (EPCC).

Abstract: Memory and I/O performance bottlenecks in supercomputing simulations are two key challenges that must be addressed on the road to Exascale. The new byte-addressable persistent non-volatile memory technology from Intel, DCPMM, promises to be an exciting opportunity to break with the status quo, with unprecedented levels of capacity at near-DRAM speeds. Here, we explore the potential of DCPMM in the context of two high-performance scientific applications in terms of outright performance, efficiency and usability for both its Memory and App Direct modes. In Memory mode, we show equivalent performance and better efficiency for a CASTEP simulation that is limited by memory capacity on conventional DRAM-only systems without any changes to the application. For IFS, we demonstrate that a distributed object-store over NVRAM reduces the data contention created in weather forecasting data producer-consumer workflows. In addition, we also present the achievable memory bandwidth performance using STREAM.