Working Toward a Smarter, Faster Cloud
A proposed system would let cloud developers control the way their data travels across different machines.
Cloud computing services have taken off in recent years. They give businesses flexible access to computing hardware and resources without all the burden of having to manage infrastructure themselves. But there are times when some customers might benefit a great deal from managing the underlying infrastructure.
In a presentation this week at Usenix Annual Technical Conference ‘10 in Boston, Vytautas Valancius, a computer science researcher at Georgia Institute of Technology, described a system that would let cloud users customize the path their data takes as it travels through cloud computing platforms. Such a path could span multiple Internet service providers and networks.
Different types of cloud applications have different needs, Valancius explained. For example, a highly interactive application such as a voice chat program probably needs a high-quality connection. In contrast, a file-backup service that transfers data in bulk might benefit from the least expensive transit between machines.
Today’s cloud services, however, send data for both types of applications over the same path. To remedy this, Valancius has been working on a system called Transit Protocol that would let users set a path that matches the needs of a specific application. The work was done with Nick Feamster, an assistant professor at Georgia Tech, Jennifer Rexford, a professor at Princeton University, and Akihiro Nakao, an associate professor at the University of Tokyo.
Valancius says cloud computing companies could have a physical connection to a variety of ISPs, negotiate with each one, and then provide the Transit Protocol service to users. Transit Protocol would let a cloud provider handle these elements and then pass the benefits on to its customers through a specially designed interface.
Transit Protocol is currently deployed in sites in Atlanta, Princeton, NJ, and Madison, WI, and is being used to power a number of academic experiments. The interface requires the user to have advanced networking knowledge. However, Valancius says that future work on the system will make it easier to use.
Currently, Transit Protocol’s users have to do their own measurement and monitoring of ISPs to determine which one can provide the best service. Valancius wants to add tools that could do these measurements automatically and give users an overview. He envisions a simpler interface that might, for example, use this measurement data to let users set their preferences for how their data will travel. Transit Protocol would then translate those preferences to specific paths.
Transit Protocol could be added to a service such as Amazon’s Elastic Compute Cloud as an optional feature for customers interested in controlling this aspect of their infrastructure. Valancius notes that big companies could also use the tool internally to vary the behavior of specific applications that run on different machines within their organization.
“As cloud platforms mature to host increasingly complex and demanding applications, customers will want a greater degree of flexibility and control over these resources,” says Andrew Warfield, an adjunct professor of computer science at the University of British Columbia and technical director for storage and emerging technologies in the virtualization management division of Citrix Systems.
Warfield says the researchers have not only described a new and interesting feature that could be added to cloud-based systems, they’ve also demonstrated that there are widespread opportunities to build infrastructure services that both live in and are used by the cloud. He hopes to see approaches similar to this in areas including storage and databases.