HCB-Tree: A Height Compressed B-Tree for Parallel Processing

B-tree type indexes are popular in database applications because they provide a fast access path to large databases. In this paper we present a new storage structure which is suitable for fast parallel searching by using B-tree like indexes [I]. We call this modified B_ tree structure the Height Compressed B_ tree (HCB_tree). The main results presented in this paper are that parallel processing of HCB_ trees reduces access time, and minimizes the frequency of restructuring considerably. Since the I/O processing time is dependent on the number of disk accesses, i.e., the height of the index tree, reducing the number of levels of the index tree is very important for secondary database storage. One way to reduce the height of B_ trees is to use a large index node. But we shall show through queueing models that this approach alone may not improve performance. And in certain circumstances this approach may even degrade performance because of increased I/O transfer time and processing time required for large nodes. The logical construction of HCB_trees is based on conventional B_ trees, but the overall philosophy is different. The goal of the HCB_tree is to