Cluster-based router architecture in active network

Abstract

An active network has the advantage of being able to accept new protocols quickly and easily. Routers in the active network perform customized computations on packets. As a result, the performance of many network applications has improved. However, they are susceptible to suffer from a lack of computing re-sources due to the additional packet processing of customized computations. The cluster-based active router is a possible solution to this problem. The cluster system can be constructed rapidly and give adequate computing resources to process active packets at a moderate cost. In the cluster-based router architecture, load balancing is achieved by the efficient distribution of packets. When a packet comes into a node in the cluster, the packet is processed by the node or is forwarded to another node. The decision of packet forward is based on the load information of the nodes. Decisions made only by CPU load without considering the forwarding overhead can increase the packet processing latencies because the overhead is not negligible in the loosely-coupled cluster system. And the portion of the over-head is relative to the packet processing time. The overhead may be small in high-computation packets and large in low-computation packets. In this thesis, we propose a cluster-based router architecture to share the overall computational load in the active network and present a packet distribution scheme. We designed and implemented the active router by modifying an active router program on a conventional cluster system. And for the efficient distribution of packets, we propose a new packet distribution scheme according to the estimated processing time (EPT). We validated the equations of EPT by comparing the calculated EPT and the measured end-to-end latency. We measured the end-to-end latency of packet delivery and counted the number of forwarded packets. So we con-firmed that packet distribution according to EPT reduces overall packet latencies especially low-c...