Slides available here!

Speaker:

Andreas Herkersdorf, TU Munich, Germany

Title:

Self-Aware Load Distribution for Software-Defined Network Nodes

Abstract:

Software-Defined Networking (SDN) enables programmable network control through a vendor-independent, spatial separation of control and data plane in state-of-the-art network elements. In the data plane, Network Function Virtualization (NFV) decouples network functions from the underlying custom hardware platform, enabling the execution of network functions in virtual environments. Depending on the performance and energy requirements, SDN and NFV network elements leverage a diverse set of hardware architectures. While a trend towards commodity, fully software programmable platforms can be observed, many devices make use of hardware acceleration to sustain high throughput rates and energy efficiency.

This talk presents our ongoing work targeting the cost and energy efficient exploitation of SDN and NFV processing resources by means of dynamic, self-aware traffic redistribution among multiple network nodes. Since datacenter traffic is highly dynamic, scaling the network's processing resources which are dedicated to a set of heterogeneous network tasks under varying load conditions is a great challenge. In the past, research on SDN/NFV network management primarily targeted long-term network scaling at the hour or minute timescales (e.g. time-of-day dependent, large news/sport events etc.). However, it is well-known that datacenter traffic shows rapid fluctuations in volume and composition even on milli- and microsecond timescales. Clearly, the control latencies in a network under (logically) centralized control prevent efficient in-time network scaling as a response to these fluctuations.

Rather than overprovisioning individual network elements to meet peak processing requirements, our approach aims at extending the network elements by a self-aware, wire-speed hardware support layer (HSL). Situation dependent, the HSL distributes traffic to surrounding network elements based on a mix of measured and estimated local resource utilization. Instead of pinning a processing action to a particular network element, we allow a subset of devices to jointly work towards an efficient execution of all required processing actions while complying with QoS demands. Thereby, self-awareness is exploited on two abstraction levels: By tracking the load in the data plane of the local node, the introduced HSL obtains local awareness and increases the degree of freedom with which the network elements may complete actions. The SDN controller remains in charge of determining the processing actions and QoS properties of each network flow and conveys availability of processing capacities in neighboring network elements (collective group awareness).

Bio:

Andreas Herkersdorf is a professor in the Department of Electrical and Computer Engineering and also adjunct to the Department of Informatics at Technical University of Munich (TUM). He received the Dipl.-Ing. degree from TUM in 1987 and the Dr. degree from ETH Zurich, Switzerland, in 1991, both in electrical engineering. Between 1988 and 2003, he has been in technical and management positions with the IBM Research Laboratory in Rueschlikon, Switzerland.

Since 2003, Dr. Herkersdorf is director of the Chair for Integrated Systems at TUM. He is a senior member of the IEEE, member of the DFG (German Research Foundation) Review Board and serves as editor for Springer and Elsevier journals for design automation and communications electronics. His research interests include application-specific multi-processor architectures, IP network processing, Network on Chip, system level SoC modeling and design space exploration methods, and self-adaptive fault-tolerant computing.



* If you wish to modify any information or update your photo, please contact the Publicity Chair at the following address.