Cost-Effective Geo-Distributed Storage for Low-Latency Web Services

We present two systems that we have developed to aid geo-distributed web services deployed in the cloud to serve their users with low latency. First, CosTLO [23] helps mitigate the high latency variance observed when accessing data stored in multi-tenant cloud storage services. To cost-effectively satisfy a web service’s tail latency goals, CosTLO uses measurement-driven inferences about replication and load balancing within cloud storage to combine the use of multiple forms of redundancy. Second, to support web services that enable users to share data, SPANStore [22] leverages application hints about access patterns and knowledge about cloud latencies and pricing to cost-effectively replicate data across data centers. Key to SPANStore’s design is to spread data across the data centers of multiple cloud providers, in order to present better cost vs. latency tradeoffs than possible with the use of any single cloud provider. In this paper, we summarize the key takeaways from our work in developing both systems.