PSC to Build Early Warning Tool for Fixing Internet Traffic Jams

$300,000 National Science Foundation Grant Will Use Web10G Data to Warn Users, Administrators for Proactive Repair of Slow Data Flow

Wednesday, March 4, 2015

A new, $300,000 National Science Foundation (NSF) grant will enable software engineers at the Pittsburgh Supercomputing Center (PSC) to build the first practical tool for warning individual users and their network administrators when their connection has developed a problem that will slow or halt data flow. The one-year project will build a tool called XSight, intended to become a standard part of the toolbox for maintaining network connections for all operating systems. XSight will build on PSC’s Web10G, a set of software tools for obtaining connection data.

“XSight will represent a proactive approach toward resolving network problems,” says PSC’s Chris Rapier, principal investigator in the project. “It’s different from other approaches to measure network performance because it takes advantage of Web10G’s ability to collect data on individual data transfers, both in the network and in an application’s interaction with the network.”

Previous attempts to measure network performance, while useful for administrators trying to recognize systemic problems, were unable to provide information either timely or detailed enough to help individual users with networking problems. Some of these methods depend on average data flow—like measuring traffic flow through an intersection over the course of a month rather than recognizing a traffic jam happening today. Some of them require invasive investigation by network administrators, taking bandwidth away from the user while admins figure the issue out. Some require users to report slow flow, which isn’t realistic when users must schedule jobs for off-peak hours—and in any case, many users simply don’t recognize when their data flow has slowed to an unacceptable speed.

By making use of the data available from Web10G, XSight is intended to overcome all of these problems, providing an automated warning system that detects data slowdowns and their causes before they become acute.

“XSight’s heuristics will automatically identify flows that are under-performing, before a user would notice,” Rapier says. “It’s the first practical tool we’ve developed from what was a research project to create Web10G.”

XSight will exploit Web10G’s ability to pull data about network connections out of TCP/IP, the set of protocols for transferring data that underlie the entire Internet. Because TCP/IP was not designed originally to produce or record detailed information about data transfers, the NSF-funded Web10G was necessary to create and extract such information. PSC developed Web10G in collaboration with the National Center for Supercomputing Applications.

The XSight project will be funded through the NSF’s Early-concept Grants for Exploratory Research (EAGER) program. EAGER supports “high risk/high payoff” early stage exploratory work on “untested but potentially transformative research ideas or approaches.”