JUNE 18–22, 2017

Presentation Details

Name: (PP05) CloudLightning & HPC: Heterogenous Computing in the Cloud
Time: Tuesday, June 20, 2017
03:15 pm - 03:45 pm
Room:   Booth #L-212  
Breaks:03:15 pm - 03:45 pm Coffee Break
Presenter:   Anne C. Elster, NTNU - Trondheim
Abstract:   As the computing resources become more heterogeneous adding accelerators such as GPUs, MICs ( e.g Xeon Phis) and FPGA-based systems such as DFE (Dataflow Engines), the task for domain scientists who want to utilize these resources become even more daunting. Cloud Computing has so far offered great tools for computing areas such as storage, hiding the complexities of underlying infrastructures from the users, providing an attractive model to extend to HPC. The ongoing Horizon 2020 Project CloudLighting proposes to create a new way of provisioning heterogeneous cloud resources to deliver services, specified by the user, through a service description language and plug-ins that interfaces with the cloud-based provisioning system. The goal is to enable improved use of compute resources, also from an energy efficiency point of view (accelerators typically are very energy efficient) by delivering heterogeneous HPC applications as a cloud service. This poster will give an overview of the CloudLightning project, and focus on describing the main use cases we selected for this project: Ray tracing, Upscaling for reservoir simulation and genomics, as well as the current testbed for the project that includes an SMP Numascale system, systems with GPU and MIC as well as a Maxler FPGA-based DFEs that has hardware that may be tailored to the algorithms for best possible performance and energy efficiency.  

PP05_Elster.pdf (1661 KB)