A fault-tolerant strategy for hierarchical control in distributed computing systems

The authors describe a practical method for realizing fault-tolerant global control of resources in distributed computing systems. The method is particularly suitable for systems that are based on a centralized arbiter for making control decisions. Many applications in LAN-based computing, online transactions, and telecommunication systems fall into this category. The method exploits the inherent physical separation of distributed computing systems to achieve high reliability in the face of decentralized arbiter failures. A significant feature of the method is that the fault-tolerance mechanisms are imbedded in the normal control signal flow so that the overhead is practically negligible in the absence of faults. The principles behind the method, its internal structure, and its operations are explained. Also, the experience gained through its application is discussed.<>