Today, many web services are being deployed using cloud infrastructure services. This is usually done to minimize user-perceived latencies while deploying the service to multiple geographically distributed data centers.

The authors of this paper study the potential for improving latencies by deploying web services not just to one particular cloud infrastructure service, but also across three of the major services: Amazon EC2, Google Compute Engine (GCE), and Microsoft Azure. The authors report that, by implementing the suggested multi-cloud deployment strategy, up to half of the users can reduce their round-trip times (RTTs) by over 20 percent. Latencies are measured from 265 PlanetLab sites around the globe. Improvements are achieved either because one of the cloud infrastructure services has a data center located closer to the user, or because routing is better to the nearby data center of one cloud service than to a similarly close data center belonging to another service.

The main contribution of this paper is the increased understanding of latencies that users of web services typically perceive. Moreover, the idea of deploying web services in a multi-cloud fashion is indeed very interesting, and the latency gain study the authors performed is well done and clearly presented. Finally, the authors present technically sound arguments. They acknowledge that implementing their idea in a real-life setting would be quite challenging because latencies to a particular cloud service very commonly fluctuate at many locations. This would make it necessary to replicate data across data centers, since the data center having the lowest latency at a given location may change frequently.