Shingo Kinoshita

Network-wide Cooperative Computing Architecture (NCCA)

A network-wide cooperative computing architecture (NCCA) is described that reduces communication-processing overhead in parallel and distributed processing systems. Network-wide load balancing and fault tolerance are achieved by interconnecting workstations. disks, and other nodes in a logical ring topology over a high-speed network; by circulating service and information request messages over this network; then having the messages accepted by the most suitable nodes that satisfy the requests. In NCCA, the cooperative processing that used to be executed by communications between applications is horizontally and vertically migrated to a lower-layer communication protocol. By the control of this communication protocol, system-wide cooperation is achieved through autonomous operation of each node. Load balancing is achieved as each node estimates its own load by monitoring received messages. compares that value with an updated load threshold for the overall system, and decides whether to accept processing request messages.

An enhanced communication protocol for flexible interconnection

We developed an enhanced communication protocol that connects nodes in a logical ring configuration via a high speed network, circulates service request messages on the network, and controls the fulfillment of these requests and their capture to optimum nodes. This protocol is based on horizontal and vertical migration, from the cooperative processing conventionally enabled by communication between applications, to a lower layer communication protocol. The protocol uses the network as shared memory, and each node can autonomously refer to and modify data shared in communication messages. Such a flexible interconnection like this between clients and servers results in system wide load balancing and enables fault tolerance.