Key to automated network management lies in the leveraging of the network knowledge resources to reduce the communication complexity. This paper proposes a generic framework for representing the knowledge semantics of multi-agent based data communication Network Management Systems (NMS). We focus our attention to the fault diagnosis functional area which is considered as the most crucial regarding the management of networked systems. The proposed network knowledge model adopts the holistic approach based on the CommonKADS methodology  for modularizing the elusive structures of network knowledge to facilitate its reusability and share- ability. The network static configurational knowledge is identified as domain knowledge ontology, whereas the static status inference knowledge is represented as multi-agent based fault-state causal reasoning models. The complex experiential knowledge of network management is formalized as task knowledge ontology. We envisage this model-based diagnosis approach as the first step towards the comprehensive knowledge acquisition, representation and dissemination in the network management domain.