While cloud providers like Amazon, Google and Microsoft offer more Storage-as-a-Service solutions than ever before, the replication necessary to deliver this data across wide-reaching physical data centers can cause serious slowdowns. That creates an issue for the thousands of service providers that depend on those data centers.
“Today the best option for most application providers is to probably deploy a distributed service on a third-party cloud provider because not everyone is Microsoft, Google or Facebook who can afford lots of data centers around the world,” said Harsha Madhyastha, Assistant Professor at the University of Michigan.
Madhyastha spoke about the challenges at the MIT Computer Science and Artificial Intelligence Laboratory, and presented his research and solution for the issue.
He showed the example of AWS cloud services, which has data centers across the world. Although Amazon is a giant service provider, it has essentially a zero replication data center with 100 milliseconds of latency outside of the United States West region.
That might not sound like a big deal, but according to Madhyastha, “as a consequence, when any particular user contacts a web server and data center, the data is not stored at that data center, you have to incur at least 100 milliseconds of latencies in many cases to fetch it from the second-closest data center.”
“One hundred milliseconds is not a trivial amount of latency,” he said. “For example, Amazon itself has observed on its own that its own website, every 100 milliseconds of additional latency results in a 1 percent drop in revenue.”
Madhyastha’s solution is SPANStore, a cost-effective geo-replicator that exports a unified view of data spread out across multiple data centers. The solution spans multiple cloud providers, from Google to AWS, and promises to lower costs by 10 times when it comes to replicating data.