Adapting the B +-tree for asymmetric I/O

Abstract

We present an indexing structure for secondary storage based on the B +-tree, but tailored for asymmetric i/o. In contrast to the traditional B +-tree, the tree is not kept balanced at all times; rather, it is kept unbalanced to cope with the i/o asymmetry commonly found in contemporary systems. We identify when imbalance is beneficial and present the unbalanced B +-tree, or the uB +-tree, its maintenance algorithms and its variants. We evaluate the uB +-tree by comparing it to the B +-tree. Our results show that the uB +-tree is highly competitive and significantly outperforms the B +-tree in environments of asymmetric i/o. © 2012 Springer-Verlag.