Abstract

Freedom from store-and-forward (S/F) deadlocks in a packet-switched network can be guaranteed with the use of deadlock avoidance protocols. However, these protocols put so many restrictions on the use of buffers that even under normal circumstances the buffer utilization is small. We propose instead a deadlock detection and resolution algorithm that is completely invisible under normal circumstances. As soon as certain channels in the network have trouble in accepting and transmitting packets due to the lack of buffers, the deadlock detection phase of the algorithm is invoked. When a deadlock is identified, the deadlock resolving phase of the algorithm is executed. Once the deadlock is resolved, the control is removed. The algorithm can be used in conjunction with either the complete partitioning or the sharing with maximum queue lengths output buffer allocation strategies. A proof on the correctness of the algorithm is given. Simulation results show that the network can maintain a relatively high throughput even when deadlocks are being detected and resolved. In addition, several properties of deadlocks are shown: i) deadlocks start to increase abruptly once the network operates beyond its capacity; and ii) under heavy load conditions, increasing the buffer pool size will not delay the occurrence of deadlocks.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call