LRU-2 vs 2-LRU - An Analytical Study.

Hierarchical caching enables users to obtain content from one of (possibly) many caches between an edge router cache and an origin server, reducing latency/overall network traffic. This can be used in many network architectures, including P2P networks and Content Distribution Networks. Of particular interest are Information Centric Networks (ICNs), which decouple content identifiers from specific network hosts and explicitly consider universal caching (collaboration between network routers) as a desirable feature.Performance analysis of a large-scale hierarchy of caches requires accurate mathematical models for various cache replacement algorithms. There is no previous study that models LRU-k. This cache replacement algorithm is important since its principle is the basis of recent algorithms such as k-LRU that outperform LRU in many situations. We first model LRU-2 using Che's approximation, as a specific case of LRU-k for $k=2$. We also extend our model to a hierarchical network of LRU-2 caches. The model is validated analytically and with simulation. The experiments show that the proposed model approximates LRU-2 accurately. LRU-2 and 2-LRU are also compared analytically and with simulations. The comparison between the two algorithms illustrates that 2-LRU outperforms LRU-2.