Web10G Project Will Create User-Friendly “Dashboard” for Identifying, Fixing Data Slowdowns

The Web10G Project has received a one-year, $178,000 Software Development for Cyberinfrastructure (SDCI) supplemental award from the National Science Foundation to develop a “dashboard” that will allow users of computer networks to identify when and where a networking problem is slowing or blocking their access. Web10G, funded by an earlier, three-year SDCI grant, is a collaboration between Pittsburgh Supercomputing Center (PSC) and the National Center for Supercomputing Applications.

“We’ve found that a lot of network users either have unrealistically high expectations or unrealistically low expectations for network performance,” says Chris Rapier, PSC network applications engineer. “Web10G has produced 127 different instruments that report on what’s going on with the network connection, ways in which it might be failing and ways in which it might be improved. With the supplemental grant, we’re going to automate that process to let users know what’s reasonable and then help them work with their network operations teams to actually get the performance they need.”

Web10G and the new tool target an unintended consequence of TCP/IP, the first set of rules to help make an electronically interconnected world work on a massive scale.

“The primary protocol used on the Internet, TCP/IP, actually hides everything that happens from the end user,” says Andrew K. Adams, PSC senior network engineer. That structure helped the Internet to grow and independent applications to be developed, but it carried the downside that networking glitches were essentially invisible, even to the people who run the networks.

“One of the big issues you have with network problems is that they all basically look the same,” Rapier says. “Either you can’t connect or the connection is really slow.”

Web10G and its predecessor, Web100, allowed the collaborators in essence to open up TCP/IP and acquire data about the network, making it a standard part of the Linux operating system favored by researchers. The new supplemental grant will allow the Web10G researchers to develop a simple tool to make those data useful to nontechnical users. While the final form of the tool is yet to be decided, the team envisions a kind of dashboard, including possibly the equivalent of a speedometer and warning lights.

Web100, which involved the same collaborating institutions as well as the National Center for Atmospheric Research, also created networking tools such as automatic received buffer tuning, which allows today’s network connections to automatically adjust to maximize network throughput.

“That actually became a part of every major operating system,” Rapier says. “It was a huge win, and we’d like to see the same adoption level for Web10G.”


About PSC: Pittsburgh Supercomputing Center ( http://www.psc.edu) is a joint effort of Carnegie Mellon University and the University of Pittsburgh together with Westinghouse Electric Company. Established in 1986, PSC is supported by several federal agencies, private industry and the Commonwealth of Pennsylvania, and is a major partner in the National Science Foundation XSEDE program.


Ken Chiacchia
Pittsburgh Supercomputing Center
This email address is being protected from spambots. You need JavaScript enabled to view it.
Shandra Williams
Pittsburgh Supercomputing Center
This email address is being protected from spambots. You need JavaScript enabled to view it.

FALL 2012 Internet2 Member meeting
October 02, 2012, 12:00 PM - 1:00 PM
Location: Salon 5

Web10G is the successor to the Web100 project; an initiative to instrument the TCP stack and present this data to users. While Web100 has enjoyed wide spread adoption in diagnosing networking performance, its design precludes it from inclusion in the mainstream production Linux kernels. Web10G's design addresses the issues in Web100 by leveraging an ABI based on NetLink and a streamlined API to facilitate easier creation and porting of Web10G applications. Moreover, Web10G elevates the instrumentation within Web100 to proposed standards, by implementing the RFC 4898 specification. In this meeting we will introduce Web10G, discuss the current status, and lay out our development road map. Since community input is crucial to this project we will have an open discussion following the presentation. We encourage participants to discuss their ideas regarding advanced diagnostics and applications that might benefit from an instrumented TCP stack. Lunch served on Mezzanine level.

The team just gave a very well recieved tutorial at the Summer 2012 Joint Techs conference at Stanford University. At this tutorial we demonstrated a new version of the Web10G kernel and API. This version of the kernel will be made available to everyone by Wednesday July 18th. We also believe we will have another release later this week that will include support for the CUBIC, BIC, and H-TCP congestion control algorithims.

The Web10G team is pleased to announce the availablity of Web10G patches for Linux kernels 3.2, 3.3, and 3.4 under the Software section of www.web10g.org. These patch sets include the necessary kernel patches, the loadable kernel module, sample client software, and the required libmnl library. The patches, like 3.1, no longer make use of the /proc interface but instead provide user access to kernel data through the netlinks interface. This change significantly decreases the overhead associated with instrumenting the TCP stack.

The Web10g team is pleased to announce that the alpha release of our TCP stack instrumentation kernel modules is now available as a patch and git source. You can find it under the 'Software' tab in the Kernel Patches section. It is currently available for the linux 3.1.0 kernel.

This release includes a considerable number of changes designed to increase the robustness and efficiency of the Web10G code. The most notable change is the elimination of the /proc ABI in favor of a netlinks ABI. This dramatically increases the number of connections that can be tracked by significantly reducing overhead. Second most is the move to a loadable kernel module in order to provide increased flexibility for installations.

This release does *not* include a userland API but it does include sample code based on libmnl in order to demonstrate the funtioning of the loadable kernel modules. The team expects to have a documented API availabe for release in the next 6 to 8 weeks.

You can join our discussion list by signing up at https://lists.psc.edu/mailman/listinfo/web10g-user. All of the team members are on the list and we aim to be responsive to your questions and comments. 

Page 1 of 2