A system and method is disclosed that provides fail-save, scalable and clustered correlation of transaction trace data. The transaction trace data is provided by a set of agents injected into application processes and processed by a set of clustered correlation servers. Each agent monitors parts of transaction executions performed by the application process into which it is injected. The agents generate transaction trace data fragments describing the transaction execution part and report those transaction trace data fragments to the correlation cluster. Routing of transaction trace data from agents to correlation servers is performed in a way that transaction data fragments describing a specific transaction are received by one correlation server regardless of the agent sending it. Intermediate nodes may be used to server as proxy between agents and the correlation server cluster to further improve the availability and failure tolerance of the monitoring system.