The provision of Storage as a Service (STaaS) in many geo-distributed datacenters by several Cloud Storage Providers (CSPs) has made online cloud storage a great choice for replicating and distributing objects that are accessed worldwide. Online Social Networks (OSN) such as Facebook and Twitter have billions of active users worldwide accessing shared objects. These users expect to access these objects within a tolerable time. To minimize users’ access latency time of these objects, OSN service providers must host several replicas of objects in many datacenters. However, this replication process produces a higher monetary cost. This paper addresses crucial issues, including how many replicas are required to fulfil the expected workload of the object and the optimal datacenters to host these replicas to reduce latency time for users and monetary costs for OSN service providers. Two online algorithms are proposed to determine the suitable number of replicas for each object and the optimal placement of these replicas. The DTS algorithm establishes the replication and placement of objects using deterministic time slots, while the RTS algorithm is based on randomized time slots. Experimental results show the effectiveness of the proposed algorithms for producing latency time below certain thresholds and reducing the monetary cost.

In social media, a huge number of worldwide data objects are posted every day. The contents of these data objects include text, links, images, audio, and videos which could be small, medium, or large and accessed across the world. Moving these data objects into a single cloud service provider (CSP) is risky and results in four-fold obstacles: vendor lock-in, service availability, cost-ineffective use, and increasing latency. Using multiple CSPs to replicate and distribute the data object solves such obstacles. However, replicating data objects among multiple CSPs increases the cost of creating and maintaining this replication. This study focuses on three issues of Online Social Network (OSN) which include: (1) determining the appropriate number of replicas of each data object based on its popularity on the OSN, (2) identifying the suitable datacenters that host the replicas according to latency time of different regions, and (3) deciding the suitable storage class for the data object at a specific time of its lifetime. Two algorithms are proposed to adapt the replication and placement of the data object according to its popularity in the OSN. The first algorithm is Dynamic Fixed Time (DFT) which uses fixed time periods to adapt replication and placement. The second algorithm is Dynamic Exponential Time (DET) which determines the data object replication and placement based on exponential time periods. A simulation using a synthesized workload generated based on a real Facebook statistic dataset shows that the proposed algorithms produce a monetary cost savings of more than 23% compared to the Static Replication and Local Placement (SRLP) algorithm.