An index structure for parallel database processing

In relational database management systems, indexing is a popular and powerful technique to accelerate database processing. There are two main research issues on indexing; index structures and access controls. The studies of index structures, which are usually trees or hashing, aim at lower access cost and smaller storage overhead. On the other hand, the studies of access controls focus on concurrency of transactions accessing the same index. The root node of a tree-structured index, is a hot hot spot accessed by several transactions concurrently. If index keys are inserted or deleted frequently, many conflicts occur and the performance of the system is degraded. This problem is serious in multiprocessor environments where several processors access the same database in parallel. Similar problems occur in intra-transaction parallelism. The response time of a transaction cannot be reduced enough if parallel portions of the transaction conflict with each other while updating an index. A new index structure and its access control are discussed to achieve inter- and intra-transaction parallelism on shared memory multiprocessors. They dominate the utilization of processors in the systems.<>