To design an effective distributed parallel server, the two principal metrics of networking should necess- arily be considered: throughput and delay of both network layer and application layer to improve the original power of the network. Distributed Parallel Server (DPS) is characterized by the response characteristic and service capability of distributed parallel server. This paper makes an in-depth study on DPS, and based on the analysis and experiments it introduces an intermedium in DPS, namely distribution device, to partition flux in network layer according to working ability of application server. Consequently performance is improved.