Technical Reports
HPL-2011-44
Clusterken: A Reliable Object-Based Messaging Framework to Support Data Center Processing
Stiegler, Marc; Li, Jun; Kambatla, Karthik; Karp, Alan
HP Laboratories
HPL-2011-44
Keyword(s): data center, publication/subscription, reliable object-based messaging
Abstract: Hadoop enables high productivity in the development of MapReduce applications. However, Hadoop becomes less effective the further an application's natural pattern of computation is from MapReduce. This paper presents Clusterken, a reliable, object-based messaging framework to support data-center-based applications. Clusterken provides object-level virtual cluster management, exactly-once message processing, composable reliability, output validity, and authorization based access control. Together, these features simplify programming and improve productivity on distributed applications by enabling arbitrary interaction patterns. We compare two implementations of a specification for a publication/subscription system, one in Hadoop, the other in Clusterken. The comparison suggests that, for at least this one application, Clusterken can yield a fourfold increase
5 Pages
External Posting Date: April 6, 2011 [Fulltext]. Approved for External Publication
Internal Posting Date: April 6, 2011 [Fulltext]