Design of deadlock detection and prevention algorithms in distributed systems

A distributed system consists of a collection of processes which communicate with each other by exchanging messages to achieve a common goal. One of the key problems in distributed systems is the possibility of deadlock. Processes are said to be deadlocked when some processes are blocked on resource requests that can never be satisfied unless drastic systems action is taken. Two distributed deadlock detection algorithms handling multiple outstanding requests is proposed and are proven to be correct: it detects all cycles and does not detect false deadlocks. The algorithms are based on the concept of chasing the edge of the waitfor graph (probe-based). Simulation results show that the proposed algorithm performs very well compared to some existing algorithms. A deadlock prevention algorithm based on the notion of coloring the nodes of the waitfor graph is also proposed. Rollback is quite less compared to some existing algorithms