Jump to Navigation

High-Performance Bulk Data Transfers in Shared Wide-Area Networks

Hamidreza Anvari

Hamidreza Anvari is a PhD candidate in the Software Systems Research Lab at the Computing Science Department of the University of Alberta, Canada. He holds the prestigious “President’s International Doctoral Award” from the University fo Alberta. He completed his Masters degree at Sharif University of Technology, Iran, and his undergraduate studies at the University of Kashan, Iran, in Computer Software Engineering. His research interests lie in High-performance Software Systems, ranging from Design to Implementation to Optimization, including networked systems, HPC models, Big Data Analytics, and Data Mining. He is the main instructor for the “Internet Protocol Suite” course and the assistant instructor for the “Parallel and Distributed Systems” course at the University of Alberta.

High-Performance Bulk Data Transfers in Shared Wide-Area Networks
مسئول برگزاری: محمد حسین منشئی
محل: سالن اجتماعات دانشکده مهندسی برق و کامپیوتر / تاریخ: 1396-09-13 / زمان: 12:15 بعدازظهر

Computational science, especially in the era of Big Data, sometimes requires large data files to be transferred and shared over computer networks. For example a research group from West Canada may transfer its collected data to a high-performance computing (HPC) cluster located in Europe for processing and then transfer the results back to Edmonton. Our study concerns the performance issues in data transfer over wide-area networks (WANs), which span long distances between cities and countries. Faster is usually better. However, different workloads and use-cases often motivate different strategies. An increasing number of world-wide research projects depend on large data transfers. Although high-performance tools and protocols exist for large data transfer across WANs, they mostly address the performance issues over dedicated networks, in which only a single user uses the link at a time. In this case, throughput (speed) is the only performance indicator, and faster is always better. However, considering the shared nature of many existing research networks, the performance of existing tools is still inefficient and problematic, both in terms of throughput and fairness. In this talk we will review the properties and performance characteristics of the existing high-performance data transfer techniques over shared networks. We will then continue with a discussion on the possibility of adopting existing tools for the shared networks.

Seminars | by Dr. Radut

ارتقاء امنیت وب با وف بومی