Spatial Multi-LRU Caching for Wireless Networks with Coverage Overlaps

Abstract

This article introduces a novel family of decentralised caching policies, applicable to wireless networks with finite storage at the edge-nodes (stations). These policies are based on the Least-Recently-Used replacement principle, and are, here, referred to as spatial multi-LRU. Based on these, cache inventories are updated in a way that provides content diversity to users who are covered by, and thus have access to, more than one station. Two variations are proposed, namely the multi-LRU-One and -All, which differ in the number of replicas inserted in the involved caches. By introducing spatial approximations, we propose a Che-like method to predict the hit probability, which gives very accurate results under the Independent Reference Model (IRM). It is shown that the performance of multi-LRU increases the more the multicoverage areas increase, and it approaches the performance of other proposed centralised policies, when multi-coverage is sufficient. For IRM traffic multi-LRU-One outperforms multi-LRU-All, whereas when the traffic exhibits temporal locality the -All variation can perform better.

Online Convex Optimization for Caching Networks

“…We begin with the single cache problem which highlights some interesting features of BSCA, such as the hybrid recency/frequency criterion it uses to cache files. We then compare BSCA with state-of-the-art reactive policies in caching networks, namely multi-LRU [33] and q-LRU [34]. We find that BSCA outperforms these policies and creates, asymptotically, average utility equal to that of the best static cache configuration.…”

“…Finally, [33] proposed the multi-LRU (mLRU) heuristic strategy, and [34] the "lazy rule" extending q-LRU to provide local optimality guarantees under stationary requests. These works pioneered the extension of the seminal LFU/LRU-type policies to the case of multiple connected caches and designed efficient caching algorithms with minimal overheads.…”

“…We denote with r n,i t ∈ {0, 1} the event that a request for file n ∈ N has been submitted by a user at location i ∈ I during slot t. At each slot we assume that there is exactly one request. 4 From a different perspective, this means that the policy is applied after every request, exactly as it happens with the standard LFU/LRUtype of reactive policies, see [33], [34] and references therein. Hence, the request process can be described by a sequence of vectors {r t } T t=1 drawn from:…”

“…arises naturally. Unlike other policies that make also such proactive updates, e.g., the LRU-ALL policy [33], BSCA can take into account these costs and reconfigure when they are smaller than the expected benefits. In detail, if we denote with c n,j the cost for transferring a file unit from the origin servers to cache j, we can define the utility-cost function:…”

“…We stress that we have chose a small value for N only to facilitate the calculation of the hindsight policy. We compare BSCA to the best static configuration, and stateof-the-art reactive policies: (i) the multi-LRU policy proposed in [33] where a request is routed to a given cache (e.g., the closest) which is updated based on the LRU rule; and (ii) and the q-LRU policy with the "lazy" rule [34] for q = 1, which works as the multi-LRU but updates the cache only if the file is not in any other reachable cache. Fig.…”