Restructuring View Maintenance Plans for Large Update

Materialized views defined over distributed data sources are a well recognized technology in data integration, e-business and semantic web. Due to the constantly increasing size of the information sources and the rapid rates of change, there comes an increasing pressure to reduce the time taken for refreshing such integration views. State-of-the-art incremental view maintenance literature requires O(n) (batch view maintenance) or more (i.e., sequential maintenance) maintenance queries with n is the number of data sources. In this work, we optimize the maintenance performance by restructuring the batch view maintenance plan to reduce the number of maintenance queries to remote data sources when maintaining a large set of updates. We first propose an adjacent grouping strategy which exploits the regularity in the batch maintenance plan. This solution reduces the number maintenance queries by sharing the common accesses to data sources. Then we propose a conditional grouping approach which reduces the number of remote queries to O(n) by unifying heterogeneous subexpressions (deltas). A cost model to analyze these approaches is provided. The proposed maintenance strategies have been implemented in our TxnWrap system. Experimental studies illustrate that our conditional grouping algorithm has about 300% performance improvement in terms of total processing time compared with existing batch algorithms in a major part of cases. Our experiments also reveal an additional dimension of this design space, namely the impact of the cooperation of the remote sources on maintenance performance.