Effect of Node Size on the Performance of Cache-Conscious Indices

In main-memory environments, the number of processor cache misses has a critical impact on the performance of the system. Cache-conscious indices are designed to improve the performance of mainmemory indices by reducing the number of processor cache misses that are incurred during a search operation. Conventional wisdom suggests that the index’s node size should be equal to the cache line size in order to minimize the number of cache misses and improve performance. As we show in this paper, this design choice ignores additional effects, such as instruction count, which play a significant role in determining the overall performance of the index. Using analytical models and a detailed experimental evaluation, we investigate the effect of the index’s node size on two common cache-conscious indices: a cache-conscious B+-tree (CSB+-tree), and a cache-conscious extendible hash index. We show that using node sizes much larger than the cache line size can result in better search performance for the CSB+-tree. For the hash index, reducing the number of overflow chains is the key to improving search performance, even if it requires using a node size that is much larger than the cache line size. Extensive experimental evaluation demonstrates that these node size choices are valid for a variety of data distributions, for range searches on the CSB+-tree, and can also be used to speed up the execution of traditional hash-based query operators that use memory-resident indices internally.