Researchers Steal Cloud Computing Power Via Browsers

While Amazon and Google compete in a cloud computing price war, neither company can beat the price proposed by academic computer researchers: free.

Computer scientists at North Carolina State University and the University of Oregon have demonstrated that it is possible to conduct large-scale cloud computing tasks anonymously at no cost by abusing cloud-based browsers, such as Amazon Silk, Cloud Browse, Opera Mini and Puffin.

Their paper, "Abusing Cloud-Based Browsers for Fun and Profit," is scheduled to be presented on December 6 at the 2012 Annual Computer Security Applications Conference in Orlando, Fla.

Other security researchers have demonstrated ways in which cloud computing can be used to break weak security. For example, Moxie Marlinspike's CloudCracker service leverages the power of cloud-based servers to crack network passwords. But such number crunching costs money.

[ Learn how to get 10 times faster query performance at one-tenth the price. Read Amazon Debuts Low-Cost, Big Data Warehousing. ]

What William Enck, an assistant professor of computer science at N.C. State, and his five co-authors describe in their paper is a technique they call a Browser MapReduce (BMR).

MapReduce, developed by Google, is a way to handle the parallel processing of large data sets. Browser MapReduce involves the aggregation of free JavaScript processing offered by cloud-based browsers, in conjunction with a scheduling scheme to work around the computational limitations imposed by cloud-browser providers, to perform MapReduce jobs.

With careful coordination, BMR demonstrates "a new way of performing parasitic computing," the paper states. For free storage, the BMR experiments conducted relied on the bit.ly URL shortening service, through which data encoded in URLs could be preserved and accessed via the bit.ly API.

The paper's authors liken cloud browsers to insecure mail servers. "By rendering Web pages in the cloud, the providers of cloud browsers can become open computation centers, much in the same way that poorly configured mail servers become open relays," they say.

BMR outperformed Amazon's Elastic MapReduce and Hadoop running on Amazon EC2 for a distributed grep job. It performed less well when processing word count and distributed sort experiments, but the researchers note that BMR was not designed to outpace MapReduce and that the meager savings in the experiments -- the Amazon jobs cost three or four cents, while BMR cost nothing -- could be much more significant for a large job over a long period of time, like distributed password cracking.