Katana: Generalized Data Processing on Peer-to-Peer Overlays

The Katana framework is a novel peer-to-peer (P2P) based generalized data processing framework. It provides a programming model in which processing logic may be implicitly distributed across a cluster of processing sites much like the Map Reduce framework but with augmented universality and extended expressiveness. As a result, the Katana framework consistently outperforms Map Reduce in our experiments on a 65 machines cluster with up to 144 GB of test data. In addition, the Katana framework functions on top of a structured P2P overlay, which is generally known to provide better scalability as compared to the master/worker architecture adopted by most Map Reduce implementations.