The dynamic tree protocol: avoiding ´graceful degradation´ in the tree protocol for distributed mutual exclusion

A modification of the tree protocol for distributed mutual exclusion is presented. The existing tree protocol is very efficient as long as all nodes in the system are accessible, but if suffers from performance degradation during the time when some nodes are down or partitioned. Although failures may be infrequent, once a node is down, it may remain down for a relatively long time. In the modified protocol, a node requesting mutual exclusion may experience performance degradation at most once after some failures occur but then it adapts to a new system topology and its performance returns to normal. In the case when all nodes in the system are accessible, the protocol exhibits the same performance as the existing tree protocol.<>