The HV-tree: a Memory Hierarchy Aware Version Index

The huge amount of temporal data generated from many important applications call for a highly efficient and scalable version index. The TSB-tree has the potential of large scalability due to its unique feature of progressive migration of data to larger mediums. However, its traditional design optimized for two levels of the memory hierarchy (the main memory and the hard disk) undermines its potential for high efficiency in face of today’s advances in hardware, especially CPU/cache speed and memory size. We propose a novel version index structure called the HV-tree. Different from all previous version index structures, the HV-tree has nodes of different sizes, each optimized for a level of the memory hierarchy. As data migrates to different levels of the memory hierarchy, the HV-tree will adjust the node size automatically to exploit the best performance of all levels of the memory hierarchy. Moreover, the HVtree has a unique chain mechanism to maximally keep recent data in higher levels of the memory hierarchy. As a result, HV-tree is several times faster than the TSB-tree for point queries (query with single key and single time value), and up to 1000 times faster than the TSB-tree for key-range and time-range queries.