In recent years the growth of the amount of information collected and stored within enterprises, scientific laboratories, and the Internet as a whole, has fueled the need for high-performance network storage.
With increasing network transfer rates, network file system performance is often limited by end-system overhead, caused primarily by memory copying and network protocol processing. In this talk, I will evaluate alternative strategies for reducing overhead in such systems, focusing on optimizations to remote procedure call (RPC)-based data transfer using either remote direct memory access (RDMA) or network interface support for pre-posting of application receive buffers.
I will also argue that RDMA network interface system are sufficient for storage applications. Such an interface can reinstate the multiprogramming principles that are violated in early commercial RDMA systems.
Kostas Magoutis is a Research Staff Member at the IBM T. J. Watson Research Center in Hawthorne, New York. His research interests include modeling and management of distributed middleware systems, self-regulating, high-speed access to network storage systems, and more recently, the emerging field of IT Services. Kostas received his PhD and SM from Harvard University in 2003 and 1998 respectively, and his BS from Aristotle University in 1993.
