Optimizing Cost for Online Social Networks on Geo-Distributed Clouds

Abstract

Geo-distributed clouds provide an intriguing platform to deploy social network (OSN) services online. Optimizing cost for online social networks on geo-distributed clouds To leverage the potential of clouds, a major concern of OSN providers is to optimize the monetary cost of using cloud resources while considering other important requirements, including providing satisfactory service quality (QoS) and data availability to OSN users.

In this Optimizing cost for online social networks on geo-distributed clouds paper, while meeting predefined QoS and data availability requirements, we study the cost optimization problem for the dynamic OSN on multiple geo-distributed clouds over consecutive time periods. We model the cost, the QoS, and the OSN’s data availability, formulate the problem, and design a cosplay algorithm. We carry out extensive experiments with a large-scale real-world Twitter trace across 10 geo-distributed clouds across the U.S.

Advantages

• Compared to existing approaches, cosplay reduces cost significantly and finds a substantially good solution of the cost optimization problem, while guaranteeing all requirements are satisfied.
• Furthermore, not only can cosplay reduce the one-time cost for a cloudbased OSN service, it can also solve a series of instances of the cost optimization problem and thus minimize the aggregated cost over time by estimating the heavy-tailed OSN activities during runtime.
• Our evaluations also demonstrate quantitatively that the tradeoff among cost, QoS, and data availability is complex; an OSN provider may have to incorporate cosplay to all three dimensions.
• For instance, according to our results, the benefits of cost reduction decline when the requirement for data availability is higher, whereas the QoS requirement does not always influence the amount of cost that can be saved.

Disadvantages

• They fail to capture the OSN features such as social relationships and user interactions, and thus their models are not applicable to OSN services.
• The cost models in all the aforementioned existing work, do not capture the monetary expense and cannot fit the cloud scenario.
• Do not explore social locality to optimize the multiOSN service.
• OSN is unique in data access patterns (i.e., social locality)

Conclusion

In this paper, we investigate the problem of optimizing the monetary cost spent on cloud resource utilization as deploying the OSN service among multiple geo-distributed IaaS clouds for consecutive time periods. We quantify the QoS by our novel vector-based approach, and model the monetary cost for the OSN data back-end, integrating social locality and exploring real-world characteristics of OSN dynamics.