Dynamic Flow Aggregation in SDNs for Application-aware Routing

• Main Authors: Tsai, Tsung-Hsien, 蔡宗憲
• Other Authors: Wang, Kuo-Chen
• Format: Others
• Language: en_US
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/j23zhy

碩士 === 國立交通大學 === 網路工程研究所 === 103 === Application-aware routing is a promising approach to allocate limited network resources to meet the requirements of different classes of applications. The Software-Defined Network (SDN) presents a chance to realize application-aware routing by offering an external and programmable network control plane. Due to limited sizes of TCAMs (ternary content addressable memories) in an SDN switch, flow aggregation, which merges several flow entries into one without modifying forwarding semantics, is needed to relieve the flow table overflow problem. With flow aggregation, we can reduce the number of flow entries in the flow table of a switch and relieve the flow table overflow problem. In this thesis, we propose bit &; subset weaving, which can reduce the number of flow entries in a switch quickly, to relieve the flow table overflow problem. The proposed bit &; subset weaving utilizes the proposed instruction slicing to produce an instruction-independent partition and applies the proposed bit and subset merging to merge flow entries in each subset in a partition. The proposed instruction slicing can reduce the time of merging flow entries, and the proposed bit or subset merging can enhance the compression ratio of flow aggregation. In addition, to achieve a better compression ratio and to reduce times to trigger flow aggregation, we introduce an adaptive threshold to determine the time to trigger flow aggregation. Simulation results show that in the proposed bit &; subset weaving, the average compression ratio of applying the adaptive threshold is 8.7% better than that of applying the static threshold. The times of triggering flow aggregation of applying the adaptive threshold is 26.13% less than that of applying the static threshold. Comparing with a representative work, fast flow table aggregation (FFTA), which employs a modified-BST (binary search tree) data structure to improve the most computation-intensive part of the bit weaving and to accelerate flow aggregation in SDNs, the average compression ratio of the proposed bit &; subset weaving is 2.05% better than that of the FFTA. The average flow aggregation time of the proposed bit &; subset weaving is 12% smaller than that of the FFTA. Since the proposed bit &; subset weaving requires less flow aggregation time and achieves a better compression ratio than the FFTA, it is more feasible for flow aggregation to relieve the flow table overflow problem in SDNs.