Capture at Scale: Ultra-fast Wireless All-to-all Communication, seminar by Olaf Landsiedel, Chalmers
A fundamental building block for many low-power wire- less protocols and applications is to efficiently share and process information among all nodes, including distributed con- trol, network-wide consensus, and aggregate functions. This work presents CAOS, an all-to-all communication primitive that addresses these needs. The key idea is to let the nodes synchronously transmit the piece of data they want to share. Nodes overhearing these synchronous transmissions receive (typically different) packets due to capture, merge their own data with the received data using a user-defined merge operator, and transmit the resulting packets again synchronously. This process repeats until all nodes have the same data. By applying user-defined merge operators, CAOS effectively integrates in-network processing into the underlying communication support. Results from two testbeds demonstrate that CAOS outperforms LWB and CTP, two state-of-the-art com- munication protocols for low-power wireless, reducing radio duty cycles by up to 33× and latency by up to 24× at a reliability of 100% in most scenarios. For example, CAOS computes aggregate functions, such as average or maximum, in a 100-node multi-hop network in less than 90 milliseconds.
Olaf Landsiedel is an Assistant Professor in Computer Science at Chalmers University of Technology. From 2010 to 2012 he spent two years as Postdoctoral fellow at KTH and SICS. In 2010 he received his PhD from RWTH Aachen, Germany, his advisor was Klaus Wehrle.