Analysis And Improvement Research On File Transfer Protocol GridFTP

Grid is a secure collaboration across geographically distributed resources. The widely distributed nature of these environments makes the performance of data transfer increasingly important. Also, applications of Grid are based on the data transfer technology because all the service process in Grid requires fast and secure data access.GridFTP that based upon the Internet FTP protocol is a high-performance, secure, reliable data transfer protocol for Grid network. GridFTP provide significant improvement in data transfer performance due to the parallel transfer in wide-area environment. With this feature, a single file can be transferred between a pair of hosts with multiple TCP streams to utilize the high bandwidth in Grid network.However, how to configure the optimal TCP parallelism in GridFTP to utilize the bandwidth effectively according to need is a questionable problem at present and the unfair utilization of the network resource caused by high parallelism is also should be paid attention. In this paper, the performance of GridFTP is quantitatively evaluated and the parallel transfer in GridFTP is primarily analyzed. By performing experiments on testing the parameters such as bandwidth, transfer time, throughput and self-similarity in transferring with different parallelism, we discuss the performance improvement for different TCP parallelism and the defection and limit that need to be improved. Furthermore, the queue delay and packet lost under different parallelism are also analyzed by performing NS2 simulation.Based on the previous analysis, this paper promotes an automatic parallelism configuration mechanism for GridFTP which optimizes the number of parallel TCP connections according to the available bandwidth. The proposed technique first measures the network status(the throughput and the round-trip time)at the GridFTP client.Then, based on the GridFTP throughput model, the number of parallel TCP connections are adjusted with a new AIMD(Additive Increase and Multiplicative Decrease) algorithm. The performance of the proposed automatic parallelism configuration mechanism through simulation experiments is evaluated and it can be indicated that the automatic parallelism configuration mechanism can tune the GridFTP parallelism at an optimal number according to the network status to maximize the throughput. With this function, it can help to avoid the network congestion and unfair share of network resource by excessive parallelism in GridFTP.