Saving Space and Time Using Index Merging for Main-Memory Databases

Managing information stored in computers is an integral part of the society we live in. Efficient access to such data is usually supported through the use of index structures. Although indices can significantly reduce the cost of answering queries, they have two major drawbacks. First, they take additional space, which can more than double the size of the database. Second, their maintenance can be a bottleneck. The reason being that every time a table is updated, all indices defined over the table need to be updated accordingly. We address these two challenges by using a mechanism that reduces the need for storing redundant data among indices. The experiments that we have conducted show that our approach can reduce by half the amount of space taken by the search structures and can improve performance for workloads with high update ratios. These results are achieved by merging indices that contain redundant data into extended indices. The later artifact extends a search tree index in two ways: (1) the ordering condition on the index does not need to be linear (e.g., different ordering can be specified on different disjoint subsets of the indexed objects), and (2) efficient (i.e, logarithmic time) search in predefined subsets of the indexed elements is supported. Our current implementation manages extended indices stored exclusively in main-memory, where evaluating the impact of storing extended indices on secondary storage is an area for future research.