ReadAny
This is the model most high availability clusters would support. Whenever a request comes, the load balancer select one target server from the server pool. This model is suitable for read service only since the state on all servers will be inconsistent if a write is issued on only one server. A back-end storage like file server or database for all servers would be need to solve this problem.
ReadOne/WriteAll
This is a model for read/write service. When a read request comes, SG load balancer select one target server from server group. When a write request comes, SG load balancer multicast this request to all servers in the server group to ensure state consistence of all servers.
ReadFirst/WriteAll
This model is also for read/write service. Its difference with ReadOne/WriteAll is the handle of read request. When a request comes, SG load balancer multicasts this read request to all servers in the server group. And SG cluster returns the first reply back to the client. This model is designed for data partitioned servers.
| SG Cluster by Distributed System Lab E.E. NCKU 2001 |