Deadlock Detection

• Permit first three necessary conditions for deadlock
• For greater efficiency, tolerate occasional deadlock
• Require early detection to avoid 'doing nothing'
• Take recovery action

Algorithm for this :

1. Maintain information on current allocation of resources to processes + outstanding resource requests
2. Check for circular wait via resource allocation graphs, petri nets etc

Invoking the algorithm :

• Each time resource is requested - may lead to very high overheads
• Periodically
• When resource utilisation drops below a limit