Computer Science and Mathematics Projects
Software Defined Network Science Flows
A number of Department of Energy (DOE) science applications, involving Exascale computing systems and large experimental facilities, are expected to generate large volumes of data, which are then required to be transported over wide-area networks for the purpose of storage, on-the-fly visualization, analysis, and computational steering. For these science workflows, the underlying data flows must be composed from complex component systems, which require a concerted optimization of the local storage-to-host data and execution paths to match the edge and long-haul network connections. Currently, such data flows are manually composed, configured, and optimized, which require teams with multiple areas of expertise and significant setup times.
Our Software Defined Networking (SDN) frameworks offer the opportunities to provide these science flows directly and transparently to the science users and applications entirely through software, without requiring a detailed expertise in the constituent systems. This project develops SDN-HPON (SDN enabled High-Performance Networks) technologies to enable DOE science applications for realizing complicated, sophisticated data flows across the science complex of computing, storage, and experiment sites, which are connected via high-performance networks.
The orchestration software will support both on-demand and advanced scheduling of custom, dynamic data flows that help complex science workflows involving networked High-Performance Computing (HPC) facilities, storage systems, and science experimental facilities. This software defined solution will build upon the emerging Open Networking technologies (e.g., OpenFlow, OpenSM), and also will develop novel control plane extensions to encompass other components, including processing nodes, cross-connects, I/O systems, storage networks, and file systems. These end-to-end science network flows will be seamlessly composed and realized entirely in software to achieve unprecedented speeds and performances through dynamic configuration and optimal resource utilization. Their implementations will be “phased-in” to support a gradual integration into production environments with a clear adoption road map.
