Abstract: A method, system, and medium are disclosed for performing transparent failover in a cluster server system. The cluster includes a plurality of servers. In servicing a client request, a primary server replicates session data for the client into memory space of one or more backup servers. The primary server sends a response to the client, wherein the response includes an indication of the one or more backup servers. When the client sends a subsequent request, it includes an indication of the backup servers. If the primary server is unavailable, the cluster determines a recovery server from among the backup servers indicated by the request. The chosen recovery server would then service the request.

Abstract: A method, system, and medium are disclosed for performing transparent failover in a cluster server system. The cluster includes a plurality of servers. In servicing a client request, a primary server replicates session data for the client into memory space of one or more backup servers. The primary server sends a response to the client, wherein the response includes an indication of the one or more backup servers. When the client sends a subsequent request, it includes an indication of the backup servers. If the primary server is unavailable, the cluster determines a recovery server from among the backup servers indicated by the request. The chosen recovery server would then service the request.

Abstract: A method, apparatus, and system are directed toward overlaying a cluster of nodes over peers in a peer-to-peer network using a peer identification and communication mechanism. A string based name for a resource in a peer-to-peer network is received and is used to generate a unique identifier using a secure hash algorithm. An advertisement specifying at least one address associated with the resource is associated with the unique identifier. The advertisement is broadcasted. The unique identifier and/or the resource in the peer-to-peer network is associated with a resource in the cluster. A network packet may be addressed to the resource in the cluster by the unique identifier. The network packet may be routed to the address associated with the resource in the cluster using an Open System Interconnection (OSI) Level 2 or above protocol.

Abstract: A method, apparatus, and system are directed toward managing cluster membership through a discovery process that enables selection of a master for the cluster. If a node on a network is the first to initiate the discovery process, it assumes the master role. Where two or more nodes enter the discovery process at the same time, or where a cluster is disjointed, views of the potential members to the cluster are shared among the nodes. Each node coalesces and sorts the views. The master may then be determined from the sorted views. In one embodiment, the master is that node in the results having a highest unique identifier. The identified master sends an assertion of its role to the other nodes. If no assertion is received, or multiple assertions are received, updated views are again shared and sorted. In one embodiment, the nodes may be within a peer-to-peer network.

Abstract: A cluster or group of cooperating systems may implement failure chain detection and recovery. The group may include multiple nodes and each node may include a group management services (GMS) module that in turn may include a group communications mechanism to detect cluster membership events. Each GMS module may maintain an identically ordered view of the current group membership. When a member of the group fails, the member that joined the group immediately after the failed member, according to respective join times, may be selected to perform recovery operations for the failed member. If a group member fails while performing recovery operations for another failed member, the next member in the group (according to respective join times) may be selected to perform recovery for that node and may also perform recovery operations for the original failed node as well.