Online cost optimization algorithms for tiered cloud storage services

Abstract

The new generation multi-tiered cloud storage services offer various tiers, such as hot and cool tiers, which are characterized by differentiated Quality of Service (QoS) (i.e., access latency, availability and throughput) and the corresponding storage and access costs. However, selecting among these storage tiers to efficiently manage data and improve performance at reduced cost is still a core and difficult problem. In this paper, we address this problem by developing and evaluating algorithms for automated data placement and movement between hot and cool storage tiers. We propose two practical online object placement algorithms that assume no knowledge of future data access. The first online cost optimization algorithm uses no replication (NR) and initially places the object in the hot tier. Then, based on read/write access pattern following a long tail distribution, it may decide to move the object to the cool tier to optimize the storage service cost. The second algorithm with replication (WR) initially places the object in the cool tier, and then replicates it in the hot tier upon receiving read/write requests to it. Additionally, we analytically demonstrate that the online algorithms incur less than twice the cost in comparison to the optimal offline algorithm that assumes the knowledge of exact future workload on the objects. The experimental results using a Twitter Workload and the CloudSim simulator confirm that the proposed algorithms yield significant cost savings (5%-55%) compared to the no-migration policy which permanently stores data in the hot tier. 2019