Freshness Control of XML Documents for Query Load Balancing
Résumé
We present an approach for controlling the freshness of replicated XML documents. The main idea is that read-only transactions may accept to read stale data, provided they can express an upper bound on the staleness of the data they read. Controlling the freshness of data accessed by read-only transactions greatly improves load balancing since it allows for choosing a node for executing the transaction even if it is not perfectly fresh. Such a routing is based on detecting which parts of a document are likely to be updated by a given transaction. Due to the rich nature of XML, the problem is quite more complex than for relational SQL data. We present a new algorithm for conflict detection between transactions, needed to estimate freshness of data according to the missing transactions on a node. We also present new freshness measures, in order to take into account the structure/content nature of XML.