大型集群JGroups

Adapt and modify JGroups to handle large cluster (several hundreds nodes) Motivation JGroups provides a reliable and in order multicast and groups memberships. However when we try to deploy to a large cluster of several hundred nodes, we encounter a large memory footprint, frequent GC and high CPU usage. In short it gobbles too much resource and makes underline application useless. After several conversation with Beal (Jgroups creator), I decide to implement two new protocol (FRAG3 and FC2) on top existing protocol. Use case One sender and a few hundred receivers. Large and disperse message size from 20 plus mb size to a few kb. Goals High availability - since only one sender (cluster version will be followed). We will trade scalability and reliability for high availability. High scalability - need to be able to scale to a few hundred nodes cluster and provides the foundation of hierarchical structure