An Adaptively Hierarchical Framework for Remote Paging on Grids

The performance of remote paging on grids is strongly dominated by the network latency of data transfer. However, the memory clients in the previously proposed remote paging systems do not always put/get pages to/from the memory servers with the least network latency. Consequently, the communication cost of remote paging has not been reduced effectively. To address this problem, we propose an adaptively hierarchical framework for remote paging on grids in this paper. The memory servers in this framework are organized as a hierarchical list in an ascendant order based on their network latencies. Only the head server in the hierarchical list can accept pages from the memory client while the others can receive pages only from the upper-level server. When a memory server uses up allowed memory space, it will move a number of the least recently used pages to the lower-level server, and will release the memory space occupied by these pages for storing new arriving pages. Moreover, the order of memory servers in the hierarchical list is dynamically adapted according to their real-time network latencies. We have compared the hierarchical framework with the peer-to-peer one which is popularly adopted by the previously proposed remote paging systems through theoretical analysis and performance evaluation. The results show that the proposed framework indeed is more effective than the P2P one for improving the performance of remote paging.