Partial Service Caching at the Edge

We consider the problem of service caching at the edge, in which an application provider can rent resources at the edge to cache its codes/libraries to serve incoming requests. A key characteristic of current edge computing platforms is that they provide pay-as-you-go flexibility via short term contracts. This implies that the client can make significant cost benefits by dynamically adjusting its caching decisions depending on the intensity of service requests. A novel feature of this work in contrast to the literature is that we allow the service to be partially cached at the edge. The probability that the edge is able to serve an incoming request is assumed to be an increasing function of the fraction of service cached. The focus of this work is on characterizing the benefits of partial service caching at the edge. The key insight from this work is that partial service caching can bring down the cost of service only if there is at least one intermediate caching level at which the probability that an incoming request can be served at the edge is strictly more than the fraction of service cached and the cost of renting edge storage resources falls within a range of values which depends on the statistics of the request arrival process. We use these insights to design near-optimal service caching policies.