| ¡¡ | Chinese Journal of Computers Full Text |
| Title | Study on Transfer Model of Deployment System for High Performance Server Cluster |
| Authors | XUE Zheng-Hua1) DONG Xiao-She1) HU Lei-Jun2),3) WU Wei-Guo1) |
| Address | 1)(Department of Computer Science and Technology, Xi¡®an Jiaotong University, Xi¡®an 710049) 2)(State Key Laboratory of High-end Server & Storage Technology, Jinan 250013) 3)(Langchao Electronic Information Industry Co., Ltd, Beijing 100085) |
| Year | 2008 |
| Issue | No.11(1956¡ª1964) |
| Abstract & Background | Abstract This paper proposes a multi-branch tree based TFTP transfer model to relieve the performance bottleneck of TFTP server. According to the fact that average bandwidth of cluster system decreases with the increase of the size of a cluster, this paper proposes bandwidth loss-based dynamic pipeline image transfer model which reveals the relation between node arrival and deployment performance. The proposed model is validated by mathematical analysis, simulation and real measurement. To improve the scalability of the proposed model, the cluster system is divided into partitions, and each partition has only one switch. All partitions concurrently work. A comparative measurement is performed among multicast, dependable multicast, BT and the proposed model. The results show that multicast and reliable multicast have better performance, however with worse reliability. The performance of the proposed model outperforms that of BT, and it just takes 17.2 seconds to deploy a 596MB system image to 48 servers. Keywords server cluster; deployment; dynamic pipeline |