For many online consumers, convenience and speed can make or break a purchase. A recent study by Imperva Incapsula reported that seven percent of online consumers said a webpage must load immediately or they lose interest in making a purchase. 35 percent said they’d wait between three and five seconds. A site that takes more than five seconds to load is at risk of losing both the web user’s attention and a potential sale.
A team of computer scientists at the Johns Hopkins University propose a solution for loading page wait time in their latest award-winning research, “DistCache: Provable Load Balancing for Large-Scale Storage Systems with Distributed Caching.” Computer Science faculty Xin Jin and Vladimir Braverman—along with second-year PhD student Zhihao Bai and recent alum Zaoxing Liu, Engr ’18 (PhD)—won a Best Paper Award for this work at the 17th USENIX Conference on File and Storage Technologies, held February 25–28 in Boston, Massachusetts.
Supported by a $500,000 National Science Foundation grant, DistCache is a new distributed caching mechanism that provides provable load balancing for large-scale storage systems.
“Nowadays, everyone uses online services like Google, Amazon, and Facebook for shopping. All of these e-commerce sites build upon large storage systems that are required to maintain a lot of data to serve their customers. The major challenge for these sites is how to build scalable-distributed storage that can provide both a service to billions of users and give them a satisfactory user-experience,” says Jin.
DistCache already has online providers—such as Barefoot Networks, a computer networking company that designs and produces programmable network switch silicon, systems, and software—interested in its services.
“DistCache is a general solution that can be applied to many storage systems,” says Braverman. “We demonstrate the benefits of DistCache by providing the design, implementation, and evaluation of the use case for emerging switch-based caching.”
“DistCache is solving the load imbalance issue in existing storage clusters, as well as trying to achieve ideal resource utilization for operators and the best user experience for fetching data,” says Liu. “Our goal is to help datacenter operators in the industry handle the largest and most complex workloads on the fast-growing internet.”
The team’s research has been selected to be presented in the “Best of the Rest” session at the USENIX Annual Technical Conference, held July 10–12 in Renton, Washington.