Data Caching under Number Constraint

Caching can significantly improve the efficiency of information access in networks by reducing the access latency and bandwidth usage. However, excessive caching can lead to prohibitive system cost and performance degradation. In this article, we consider the problem of caching a data item in a network wherein the data item is read as well as updated by other nodes and there is a limit on the number of cache nodes allowed. More formally, given a network graph, the read/write frequencies to the data item by each node, and the cost of caching the data item at each node, the problem addressed in this article is to select a set of P nodes to cache the data item such that the sum of the reading, writing (using an optimal Steiner tree), and storage cost is minimized. For networks with a tree topology, we design an optimal dynamic programming algorithm that runs in O(|V|3P2), where |V| is the size of the network and P is the allowed number of caches. For the general graph topology, where the problem is NP-complete, we present a centralized heuristic and its distributed implementation. Through extensive simulations in general graphs, we show that the centralized heuristic performs very close to the exponential optimal algorithm for small networks, and for larger networks, the distributed implementation and the dynamic programming algorithm on an appropriately extracted tree perform quite close to the centralized heuristic.