Dynamic replication and migration of data objects with hot-spot and cold-spot statuses across storage data centers

Abstract

Cloud Storage Providers (CSPs) offer geographically dispersed data stores providing several storage classes with different prices. A vital problem faced by application providers is how to exploit price differences across data stores to minimize monetary cost of applications that include hot-spot objects that are accessed frequently and cold-spot objects that are often accessed far less. This monetary cost consists of replica creation, storage, Put, Get, and potential migration costs. To optimize such costs, we first propose the optimal solution that leverages dynamic and linear programming techniques with the assumption that the workload on objects is known in advance. We also propose a lightweight heuristic solution, inspired from an approximate algorithm for the Set Covering Problem, which does not make any assumption on the object workload. This solution jointly determines object replicas location, object replicas migration times, and redirection of Get (read) requests to object replicas so that the monetary cost of data storage management is optimized while the user-perceived latency is satisfied. We evaluate the effectiveness of the proposed lightweight algorithm in terms of cost savings via extensive simulations using CloudSim simulator and traces from Twitter. In addition, we have built a prototype system running over Amazon Web Service (AWS) and Microsoft Azure to evaluate the duration of objects migration within and across regions.