Latency-sensitive data allocation and workload consolidation for cloud storage

Customers often suffer from the variability of data access time in (edge) cloud storage service, caused by network congestion, load dynamics, and so on. One efficient solution to guarantee a reliable latency-sensitive service (e.g., for industrial Internet of Things application) is to issue requests with multiple download/upload sessions which access the required data (replicas) stored in one or more servers, and use the earliest response from those sessions. In order to minimize the total storage costs, how to optimally allocate data in a minimum number of servers without violating latency guarantees remains to be a crucial issue for the cloud provider to deal with. In this paper, we study the latency-sensitive data allocation problem, the latency-sensitive data reallocation problem and the latency-sensitive workload consolidation problem for cloud storage. We model the data access time as a given distribution whose cumulative density function is known, and prove that these three problems are NP-hard. To solve them, we propose an exact integer nonlinear program (INLP) and a Tabu Search-based heuristic. The simulation results reveal that the INLP can always achieve the best performance in terms of lower number of used nodes and higher storage and throughput utilization, but this comes at the expense of much higher running time. The Tabu Search-based heuristic, on the other hand, can obtain close-to-optimal performance, but in a much lower running time.