guest ::
| Home > Publications database > Enhancing Remote Sensing Applications towards Exascalewith the DEEP-EST Modular Supercomputer Architecture |
| Home > Publications database > Enhancing Remote Sensing Applications towards Exascalewith the DEEP-EST Modular Supercomputer Architecture |
| Poster (Other) | FZJ-2019-00219 |
; ; ;
2018
Please use a persistent id in citations: http://hdl.handle.net/2128/21170
Abstract: Due to the advancement of the latest-generation remote sensing instruments, a wealth of information is generated almost on a continuous basis and with an increasing rate at global scale. This sheer volume and variety of sensed data leads to a necessary re-definition of the challenges within the entire lifecycle of remote sensing data. Trends in parallel High-Performance Computing (HPC) architectures are constantly developing to tackle the growing demand of domain-specific applications for handling computationally intensive problems. In the context of large scale remote sensing applications, where the interpretation of the data is not straightforward and near-real-time answers are required, HPC can overcome the limitations of serial algorithms. The Dynamic Exascale Entry Platform - Extreme Scale Technologies (DEEP-EST) aims at delivering a pre-exascale platform based on a Modular Supercomputer Architecture (MSA) wherein each module has different characteristics. The MSA provides not only a standard CPU cluster module, but a many-core Extreme Scale Booster (ESB), a Global Collective Engine (GCE) to speed-up MPI collective operations in hardware, a Network Attached Memory (NAM) as a fast scratch file replacement, and a hardware accelerated Data Analytics Module (DAM). As partner in the DEEP-EST consortium, we aim at enhancing machine learning in the remote sensing application domain towards exascale performance. Several of the innovative DEEP-EST modules are co-designed by particular methods such as the clustering algorithm Density-Based Spatial Clustering (DBSCAN) and classification algorithms like Support Vector Machines (SVMs) and Convolutional Neural Networks (CNNs). We intend to present how the different phases of these algorithms (i.e., training, model generation and storing, testing, etc.) can be neatly distributed across the various cluster modules and thus leverage their unique functionality. The MSA will be used to not only improve the performance of these methods but also to serve as blueprint for the next generation of exascale HPC systems.
|
The record appears in these collections: |
PDF
PDF (PDFA)