Cooperative caching is a popular mechanism to allow an array of distributed caches to cooperate and serve each others' Web requests. Controlling duplication of documents across cooperating caches is a challenging problem faced by cache managers. In this paper, we study the economics of document duplication in strategic and nonstrategic settings. We have three primary findings. First, we find that the optimum level of duplication at a cache is nondecreasing in intercache latency, cache size, and extent of request locality. Second, in situations in which cache peering spans organizations, we find that the interaction between caches is a game of strategic substitutes wherein a cache employs lesser resources towards eliminating duplicate documents when the other caches employs more resources towards eliminating duplicate documents at that cache. Thus, a significant challenge will be to simultaneously induce multiple caches to contribute more resources towards reducing duplicate documents in the system. Finally, centralized decision making, which as expected provides improvements in average latency over a decentralized setup, can entail highly asymmetric duplication levels at the caches. This in turn can benefit one set of users at the expense of the other, and thus will be challenging to implement.
KeywordsWeb caching, cooperative caching, duplication in caching, analytical modeling, incentive-centered design, game theory
Disciplines
Experimental Analysis of Behavior | Other EconomicsThis journal article is available at ScholarlyCommons: http://repository.upenn.edu/oid_papers/220Electronic copy available at: http://ssrn.com/abstract=926927
Cooperative Cashing?: An Economic Analysis of Document Duplication in Cooperative Web
Caching
AbstractCooperative caching is a popular mechanism to allow an array of distributed caches to cooperate and serve each others' web requests. Controlling duplication of documents across cooperating caches is a challenging problem faced by cache managers. In this paper, we study the economics of document duplication in strategic and non-strategic settings. We have three primary findings. First, we find that the optimum level of duplication at a cache is non-decreasing in inter-cache latency, cache size and extent of request locality. Second, in situations in which cache peering spans organizations, we find that the interaction between caches is a game of strategic substitutes wherein a cache employs lesser resources towards eliminating duplicate documents when the other caches employs more resources towards eliminating duplicate documents at that cache. Thus, a significant challenge will be to simultaneously induce multiple caches to contribute more resources towards reducing duplicate documents in the system. Finally centralized decision-making, which as expected provides improvements in average latency over a decentralized setup, can entail highly asymmetric duplication levels at the caches. This in turn can benefit one set of users at the expense of the other and thus will be cha...