SYNCHRONIZATION PROCEDURE IN A ROUTING NODE

Disclosed is a method for providing synchronization of a transaction in a data processing system where the transaction involves an initiator node, which starts said synchronization, and a coordinator node, which decides the outcome of said transaction. The initiator node and the coordinator node communicate through a routing node. The routing node has no resources of its own which are modified in the transaction. The synchronization is provided by establishing a first conversation between the initiator node and the routing node and a second conversation between the routing node and the coordinator node. A first commit request message is then sent from said initiator node to the routing node. A second commit request message is sent from the routing node to the coordinator node. An additional step is performed of storing asynchronously, checkpoint information to non-volatile storage in the routing node after establishing the first and second conversations and before sending said second commit request message. The routing node acts as just a routing node and does not save checkpoint state as if it were a node with a real need to do so. In the event of failure the routing node only recovers the routing information, that is the end points of the conversation it was routing when failure occurred. The resynchronization protocols are carried between the end points with the routing node not aware of their contents, except to monitor when it can safely forget its routing information.