|
Abstract : |
In this paper we study the problem of Transparent Edgeservice Replication in edge server architectures, which strive to push content away from a single server and toward replicas located at the edges of the network. We examine two known algorithms, FIFO invaliadate and FIFO push-all and show that they do not utilize bandwidth eectively, and that they either suer from low availability and high response times or from high bandwidth consumption and bad timeliness. We propose two new algorithms, 3-queue and FIP, that employ invalidation and prefetching to dynamically tune their behavior based on the bandwidth available to the system, while providing high availability, low response times, good bandwidth utilization, and provide timeliness guarantees. We show using a trace driven simulator that our algorithms can yield up to two orders of magnitude reduction in response times under our model while providing, on average, availability through network partitions that may last hours. 1., |
