Data placement optimization through heterogeneous cloud storage

Abstract

Storage as a Service is one of the fundamental parts of the cloud computing industry. Each storage providers offer different storage plan to meet various of customers’ requirement. Data temperature is one of the well-known models in data placement. In this model, the most frequently accessed data are called hot data. On the other hand, the infrequently accessed data are called cold data. Typically, most data placement algorithms decide where to store data based on three costs: storage cost, retrieval cost, and bandwidth cost. However, there is another condition in cold data, so-called minimal storage duration. As a result, an additional cost charged if the cold data are moved or deleted before the agreed storage duration. This extra cost is called the early update cost. As far as we are aware, none of the existing work had deeply studied the effect of this retention time. We proposed a new data placement to address the penalty cost. The experiments showed that our proposed algorithm eliminated the early update cost to nearly zero in mostly cold data and a very long time storage scenario. Besides, our proposed algorithm reduces the overall cost up to 24.87% compared to the existing algorithms.