Overlay networks have emerged as a useful approach to providing a general framework for new applications and services that are to be implemented without significantly changing the IP-layer network infrastructure.Overlay routing has been used as an alternative to the default best effort Internet routing for the absence of end-to-end Quality of Service(QoS). While the former has recently been investigated, the conflict of QoS restraints and resource optimization remains unsolved. Recent studies have shown that overlay paths can give better latency, loss rate and TCP throughput. In this paper, a multi-dimensional QoS objective model based on the analysis of multiple QoS constraints has been presented, and a routing algorithm to optimise the overlay resource of its nodes and links is then proposed.In fact, the algorithm obtained multiple QoS values using probability theory to achieve the routing according to the multi-dimensional QoS objective vector of the QoS objective model. Simulation results reveals that the algorithm works better than other existing algorithms in balancing the network resources, and applications with stringent QoS requirements could be run. Overlay networks have emerged as a useful approach to provide a general framework for new applications and services that are to be implemented without significantly changing the IP-layer network infrastructure. Overlay networks has emerged as an alternative to the default best effort Internet routing for the The absence of end-to-end Quality of Service (QoS). While the former has recently been investigated, the conflict of QoS restraints and resource optimization remains unsolved. Recent studies have shown that overlay paths can give better latency, loss rate and TCP throughput . In this paper, a multi-dimensional QoS objective model based on the analysis of multiple QoS constraints has been presented, and a routing algorithm to optimize the overlay resource of its nodes and links is then proposed. values ​​using probability theory to achieve the routing according to the multi-dimensional QoS objective vector of the QoS objective model. Simulation results reveals that the algorithm works better than other existing algorithms in balancing the network resources, and applications with stringent QoS requirements could be run.