Scalability Analysis of Large-scale CCN Network with Network Virtualization

摘要

In this paper, we investigate the scalability of CCNx, an open-source CCN (Content-Centric Networking) implementation, in terms of the number of nodes. As performance metrics, we measure the total throughput of content delivery, the packet loss rate in the network, and the average content delivery time. We also examine the performance bottleneck of CCNx through system-wide profiling, which quantitatively shows that per-packet digest-based authentication is the performance bottleneck in CCNx. Our findings include that the communication performance was degraded when the number of CCN routers exceeds around 30, and that the Data-chunk digest computation consumes approximately 35 % of the total CPU time. We therefore investigate how the scalability of CCNx in terms of the number of nodes can be improved by hardware offloading of Data-chunk digest computation. We found that hardware offloading of Data-chunk digest computation reduces the average content delivery time by approximately 25%.