Heat Balancing for Btree Indexed Database over Ring Configured Shared Nothing Machines

In shared nothing machines, data are typically declustered and indexed across the system processing elements (PEs) to achieve efficient query and transaction execution. Since the access pattern is inherently dynamic, therefore, there is no one placement of relations that is optimal for the lifetime of the database system. Whenever data is moved across PEs, the indexes need to be modified too. Consequently reorganization based on the access history (heat) of the data or its corresponding index is essential and should be online, in addition, it should satisfactorily deal with the index modification, as data is moving across the PEs. In this paper, we study the heat balancing strategies that have the capability of minimizing the index modification. Meanwhile, it has been observed that the system performance during reorganization is greatly affect by the cost of the balancing decisions (migration cost) and the location of the hot spot in the system. In this paper, we propose a new migration configuration, namely the ring configuration, it has the capability to reduce the migration cost and to absorb the dependency on hot spot locations. It has been shown there is a considerable reduction in the system migration cost, which mainly improves the system performance during the reorganization process. We also propose a new online heat balancing algorithm, namely two bags of heat, that utilizes the ring configuration with additional advantages of distributing the given heat as evenly as possible across the system PEs and the capability to minimize the balancing convergent time. We evaluate the performance of the proposed strategy in comparison with the linear configuration; the conducted simulation shows its efficiency in correcting the system-performance