COMPARATIVE PERFORMANCE EVALUATION OF CENTRALIZED IN-NETWORK CACHING POLICIES AND CONTENT VISIBILITY LEVELS IN INFORMATION CENTRIC NETWORK BY USING OPENFLOW TESTBED

Abstract

Performance is a key factor for designing Information Centric Network (ICN) under the support of Software Defined Networking (SDN). Moreover, the implementation of centralized in-network caching policies and content visibility levels surely affects the performance. This dissertation plans to evaluate the performance of ICN under SDN support when three different mechanisms of centralized in-network caching policy with content visibility level are applied by using OpenFlow testbed. It also aims to study the effects of network topology on the network performance. Those three mechanisms consist of non-cooperative in-network caching policy with individual content visibility mechanism, path cooperative in-network caching policy with path content visibility level mechanism and global cooperative in-network caching policy with global content visibility level mechanism. To evaluate the performance; four performance metrics including server hit ratio, average hop count, message rate between control and data plane, and bottleneck link traffic were utilized. The experiments were conducted in two experimental environments including PC-based OpenFlow testbed and emulation by Mininet. In each environment, two different network topologies were tested. The results show that each mechanism has its strengths and weaknesses. Also, the network performance corresponding to each mechanism depends on the types of network topology. While the network is simple, the global cooperative in-network caching policy with global content visibility level mechanism outperforms others. But, when the network becomes larger, this mechanism shows some disadvantages in terms of message rate between control and data plane, and the bottleneck link traffic. The non-cooperative in-network caching policy with individual content visibility mechanism performs the worst compared to others. On the other hand, emulation by Mininet is recommended to use because it utilizes less resource and gives the same results as those in PC-based OpenFlow testbed environment.