A dive into the caching performance of Content Centric Networking

Abstract

Abstract-Content Centric Networking (CCN) is a promising architecture for the diffusion of popular content over the Internet. Radically departing from content-oblivious IP networks, CCN pushes content-awareness down the networking stack. By relying on in-network caching, CCN reduces the overall network load, as requests no longer need to travel until the content originator, but are typically served by a closer CCN router along the path.While the system design of CCN is sound, gathering a reliable estimate of CCN caching performance in the current Internet scenario is challenging, due to its large scale and to the lack of agreement in some critical elements of the evaluation setup. In this work, we add a number of important pieces to the CCN puzzle. First, we pay special attention to the locality of the user request process, as it may be determined by user interest or language barrier. Second, we consider the existence of possibly multiple repositories for the same content, as in the current Internet, along with different CCN interest forwarding policies, exploiting either a single or multiple repositories in parallel.To widen the relevance of our findings, we considering multiple topologies, content popularity settings, caching replacement policies and CCN forwarding strategies. Summarizing our main result, we find that though the use of multiple content repositories can be beneficial from the user point of view, it may however counter part of the benefits in case the CCN strategy layer implements simple forwarding policies.

Performance Evaluation of Information-Centric Networking for Multimedia Services

“…The accuracy of the developed analytical model is validated via a discrete-event simulator [17], developed under the OMNeT++ framework. This open-source simulator implements the Content Store (CS), Pending Information Table (PIT) and Forwarding Information Base (FIB) data structures, and content retrieve operations of ICN.…”

“…(iii) The cache on each ICN node runs the LRU cache replacement policy. The LRU policy has low complexity and has been used in [8], [9], [16], [17]. Moreover, the caching operations of LRU can be implemented at line speed, which is one important requirement of ICN.…”

“…(v) The popularity of contents belonged to different services follows the Zipf distribution. Zipf distribution is widely used for characterising the content popularity in [8], [15], [17], [18], because it has been pointed out in [19] that the popularity real web content accesses has been observed following the Zipf distribution. As a result, contents in service k (i.e., the k-th most popualr content) are requseted with the probability…”

ccnSim: An highly scalable CCN simulator

“…Notice that while this may seem a minor implementation detail, a binary search rather becomes a requirement when the catalog size reaches N = 10 8 objects as for YouTube [12]. The popularity model possibly includes a spatial heterogeneity to account for skew in the popularity law due to geographical/cultural barriers [16].…”

“…While our previous research has focused on the performance of CCN gathered through ccnSim [14]- [16], [21], this paper focuses on the performance of the ccnSim tool itself. Specifically, two crucial aspects have been taken into account in the design of ccnSim, namely a) Memory occupancy: The simulation of very large catalogs and very large cache size is essential to gather CCN performance under realistic settings.…”