Synergy via Redundancy: Adaptive Replication Strategies and Fundamental Limits

The maximum possible throughput (or the rate of job completion) a multi-server system is typically sum service rates individual servers. Recent work shows that launching multiple replicas and canceling them as soon one copy finishes can boost throughput, especially when time distribution has high variability. This means redundancy can, in fact, create synergy among servers such their overall greater than seeks to find fundamental limit achieved by replication optimal policy achieve it. While most previous works consider upfront policies, we expand set policies delayed launch replicas. search for adaptive be formulated Markov Decision Process, using which propose two myopic MaxRate AdaRep, adaptively replicate jobs. In order quantify optimality gap these other derive upper bounds on capacity, provide limits queueing systems with redundancy.