Font Size: 

Our paper describes an alternative approach to address translations based on Distributed Hash Table (DHT) usage for Web Application. Identification record is formed of all identifcations that the network device has, and all these identifications are used to create keys for storing the record to a DHT. As a result, the record is stored in more copies and each copy provides access to others, allowing us to create efficient update and recovery mechanisms. Storing collections of network identifications has some advantages: straightforward implementation of all possible translations at the same time (provides a way to location services which are unavailable in current systems), and fault-tolerance to DHT's node and network failures due to inherent replication of identification records.
The main contribution of this paper entails the illustration of the simplest mechanism that exploits multiplicity of location records in Generic Network Location Service (GNLS) reacts to the detected node failure as follows: Together with starting a stabilization mechanism (i.e. an update of node links and ?nger tables) lookup request is broadcast using the ?nger tables’ tree. For a single node failure, the worst case analysis shows that in at most O(2.logN) steps all replicas located out of the failing node will be found. That means, the lookup request can return the result, and, since the failing node is identi?ed, lost replicas can be reproduced at the failing node surrogate. The advantage of the reactive strategy to replication is that it bene?ts from essence of location records: multiple copies distributed among more nodes, i.e. the mechanism needs no extra memory.

The risk of placing all copies to a single failing node is acceptably low for higher number of DHT nodes. Moreover, such a situation can be simply detected and the additional copy of the location record can be created elsewhere, e.g. at the node’s successor. Any node may be used as a place for such an additional copy, linkage of this backup copy (found by lookup broadcast in the case of failure detected during regular lookup) to the true location record can be maintained without difficulties. The in?uence of multiple failures is lower than in proactive methods assuming the number of ?elds in location records is greater then number of copies in proactive methods (including the original).

This service can be exploited by various Web applications designers implementing overlay networks running on top of the existing Internet network topology including lookup services embedded in search engines.