Distributed Search Trees: Fault Tolerance in an Asynchronous Environment

We propose a distributed dictionary that allows insert and search operations and that tolerates arbitrary single server crashes. The distinctive feature of our model is that the crash of a server cannot be detected. This is in contrast to all other proposals of distributed fault-tolerant search structures presented thus far. It reflects the real situation in the internet more accurately, and is in general more suitable to complex overall conditions. This makes our solution fundamentally different from all previous ones, but also more complicated. We present in detail the algorithms for searching, insertion, and graceful recovery of crashed servers.