A robust system is one that can ensure essential services in spite of faults and failures in useful but non-essential components. Unless we can ensure that critical services can only USE but not depend on less critical components, a seemingly minor fault can propagate along complex and implicit dependency chains and bring down the system. Modern real time systems are often developed concurrently by multiple teams. A team typically only knows the dependency relations between their components and neighboring components. In addition, dependency relations will change as software components and their interactions are being modified. Therefore, how to automatically track and analyze the system wide dependency from local information is important for the development of robust real time systems. This paper presents dependency algebra - a unified theoretical framework plus a prototype toolkit for dependency management in real-time systems.