Traffic anomalies and distributed attacks are commonplace in today's networks. Single point detection is often insufficient to determine the causes, patterns and prevalence of such events. Most existing distributed intrusion detection systems (DIDS) rely on centralized fusion, or distributed fusion with unscalable communication mechanisms. In this paper, we propose to build a DIDS based on the emerging decentralized location and routing infrastructure: distributed hash table (DHT). We embed the intrusion symptoms into the DHT dimensions so that alarms related to the same intrusion (thus with similar symptoms) will be routed to the same sensor fusion center (SFC) while evenly distributing unrelated alarms to different SFCs. This is achieved through careful routing key design based on: 1) analysis of essential characteristics of four common types of intrusions: DoS attacks, port scanning, virus/worm infection and botnets; and 2) distribution and stability analysis of the popular port numbers and those of the popular source IP subnets in scans. We further propose several schemes to distribute the alarms more evenly across the SFCs, and improve the resiliency against the failures or attacks. Evaluation based on one month of DShield firewall logs (600 million scan records) collected from over 2200 worldwide providers show that the resulting system, termed Cyber Disease DHT (CDDHT), can effectively fuse related alarms while distributing unrelated ones evenly among the SFCs. It significantly outperforms the traditional hierarchical approach when facing large amounts of diverse intrusion alerts.