A Mechanism for TCP-Friendly Transport-Level Protocol Coordination

In this paper, we identify an emerging and important application class comprised of a set of processes on a cluster of devices communicating to a remote set of processes on another cluster of devices across a common intermediary Internet path. We call these applications cluster-to-cluster applications, or C-to-C applications. The networking requirements of C-to-C applications present unique challenges. Because the application involves communication between clusters of devices, very few streams will share a complete end-to-end path. At the same time, network performance needs to be measured globally across all streams for the application to employ interstream adaptation strategies. These strategies are important for the application to achieve its global objectives while at the same time realizing an aggregate flow behavior that is congestion controlled and responsive. We propose a mechanism called the Coordination Protocol (CP) to provide this ability. In particular, CP makes fine-grained measurements of current network conditions across all associated flows and provides transport-level protocols with aggregate available bandwidth information using an equation-based congestion control algorithm. A prototype of CP is evaluated within a network simulator and is shown to be effective.