Load-Adaptive Indexing Method for Schema-less Searches

There is a growing need for a horizontal-integrated platform for machine to machine (M2M) applications. To facilitate high reusability of the various data stored on the platform, the platform's search function must be able to locate all data useful to a particular application. We refer to this cross-schema-manner search as a schema less search. We have proposed a tree-structured indexing method, called UBI-Tree, for efficient schema-less searches. UBI-Tree predicts usage frequencies of each query based on characteristics of stored data. Moreover, UBI-Tree classifies incoming data to improve the search performance of frequently used queries. One disadvantage of the conventional UBI-Tree is decreased search performance as a result of differences between predicted and actual frequencies. Therefore, we propose a new method to improve search performance. This method utilizes the query history and is based on the real search load. We show experimentally that our proposed method improves the average latency and throughput of search processes.