Co-allocation strategy with variable block size in grid environment

Abstract

A co-allocation strategy in data grids uses parallel data transfer from multiple servers to improve the transfer completion time. A co-allocation strategy has good performance when every available server can fully utilize the connection between the server and the client to transfer data. In existing co-allocation strategies, the time which is not spent in data transmission is constituted of (1) the waiting time after the transmission of one block until the transmission of the next block for each server and (2) the time other servers wait for the last server to complete the transmission of the last block. The strategy proposed in this paper tries to reduce the number of blocks, in order to reduce the waiting time after the transmission of one block until the transmission of the next block, by increasing the block size. However, the larger block size increases the waiting time for the last block. Thus, we use variable block size for data transfer. Our strategy starts sending larger blocks first, and then gradually decreases the block size. We run a grid simulation to measure the data transfer completion time. From the experiments, we found that data transfer completion time is reduced comparing to the existing scheme. Furthermore, an advantage of our approach is that our algorithm is not sensitive to the variance of the transmission rate.