Diagnosing the anomalies of inter-AS (autonomous system) routing and flexibly controlling its behavior to adapt to environmental changes are difficult, because this information changes spatially and temporally over different administrative domains. Multi-agent-based systems that coincide with this control architecture have been applied to these domains, but the number of deployed agents is small and more accurate analysis taking into consideration the actual Internet structure is desirable. To enable better analysis, cooperation among tens of thousands of agents is needed. This paper proposes a cooperative routing management system, called NetManager-M, which enables detailed analysis by using massively deployed agents on the Internet. NetManager-M can diagnose the routing flows around suspicious areas through cooperation among dynamically organized agents. This cooperation, which is conducted based on the current and previous routing topology, enables monitoring of routing update messages at neighboring observation points and the identification of the causes of problems in more detail. This system is thus an effective means of detailed analysis for typical scenarios of inter-AS routing failures.