Analytical Investigation of On-Path Caching Performance in Information Centric Networks

Information Centric Networking (ICN) architectures are proposed as a solution to address the shift from host-centric model toward an information centric model in the Internet. In these architectures, routing nodes have caching functionality that can influence the network traffic and communication quality since the data items can be sent from nodes far closer to the requesting users. Therefore, realizing effective caching networks becomes important to grasp the cache characteristics of each node and to manage system resources, taking into account networking metrics (e.g., higher hit ratio) as well as user’s metrics (e.g. shorter delay). This thesis studies the methodologies for improving the performance of cache management in ICNs. As individual sub-problems, this thesis investigates the LRU-2 and 2-LRU algorithms, geographical locality in distribution of users’ requests and efficient caching in ICNs. As the first contribution of this thesis, a mathematical model to approximate the behaviour of the LRU-2 algorithm is proposed. Then, 2-LRU and LRU-2 cache replacement algorithms are analyzed. The 2-LRU caching strategy has been shown to outperform LRU. The main idea behind 2-LRU and LRU-2 is considering both frequency (i.e. metric used in LFU) and recency (i.e. metric used in LRU) together for cache replacement process. The simulation as well as numeric results show that the proposed LRU-2 model precisely approximates the miss rate for LRU-2 algorithm. Next, the influence of geographical locality in users’ requests on the performance of network of caches is investigated. Geographically localized and global request patterns have both been observed to possess Zipf (i.e. a power-law distribution in which few data items have high request frequencies while most of data items have low request frequencies) properties, although the local distributions are poorly correlated with the global distribution. This suggests that several independent Zipf distributions combine to form an emergent Zipf distribution in real client request scenarios. An algorithm is proposed that can generate realistic synthetic traffic to regional caches that possesses Zipf properties as well as produces a global Zipf distribution. The simulation results show that the caching performance could have different behaviour based on what distribution the users’ requests follow. Finally, the efficiency of cache replacement and replication algorithms in ICNs are studied since ICN literature still lacks an empirical and analytical deep understanding of benefits brought by in-network caching. An analytical model is proposed that optimally distributes a total cache budget among the nodes of ICN networks for LRU cache replacement and LCE cache replication algorithms. The results will show how much user-centric and system-centric benefits could be gained through the in-network caching compared to the benefits obtained through caching facilities provided only at the edge of the network