Microsoft Researchers “Network” at SIGCOMM 2008

SEATTLE — Aug. 19, 2008 — This week in the Emerald City, hundreds of the world’s top researchers in the field of computer networking will converge for SIGCOMM 2008, the Association for Computing Machinery’s annual conference of the Special Interest Group on Data Communications. SIGCOMM is the field’s leading conference and a major source of ideas for academics and industry pros alike.

Some of the topics Microsoft Research (MSR) will present papers on this year include wireless networking, peer-to-peer gaming and enterprise datacenters. The common theme is how the physical design and the software that powers connected systems can be optimized to make the overall network faster, cheaper, more powerful and easier to operate.

The end results, according to Microsoft Principal Researcher Victor Bahl, co-chair of this year’s SIGCOMM, can range from reducing the energy footprint of large datacenters and driving the evolution of new online services to allowing thousands of video gamers to participate in an online battle.

“Much of this work is at the intersection of what electrical engineers and computer scientists do, as opposed to pure computer science,” Bahl says. “A lot of the breakthroughs are happening at that boundary between the two, and the really interesting part is the variety of scenarios.”

According to Bahl, conferences such as SIGCOMM have a major impact on how the technology evolves. Each year, SIGCOMM gives researchers in the field the opportunity to “network,” discuss new approaches, and decide how the latest thinking could enhance their own research and product development efforts.

It’s the most prestigious such conference in the world, and as a result, a very high bar exists to get research papers accepted for presentation to the community. This year, Microsoft Research is presenting eight of the 36 papers on the conference program.

In total, 288 papers were submitted to the committee for consideration, so the overall acceptance rate was just 12.5 percent. A complete list of the papers Microsoft Research will present at SIGCOMM 2008 can be found on the MSR Web site.

New Research Unveiled

In his day job as group manager for Microsoft Research’s networking group in Redmond, Bahl works with 14 Ph.D.s and a slew of interns on a variety of networking problems. The work is designed to enhance Microsoft’s product offerings, but also advance the state-of-the-art of networking technology in general.

With so many papers accepted for the conference this year, Bahl and his team are excited about the results of their recent work.

“The fact that it’s SIGCOMM and there’s a history, the quality of submissions is very high already,” Bahl says. “It’s very difficult to get in, and we’re very happy to have so much of our work being accepted.”

Papers at SIGCOMM 2008 cover the gamut between traditional networking subjects — routing, network traffic measurement and optimization, security — and the evolution of the field with new areas such as datacenter networking, enterprise network management, wireless, and peer-to-peer social networks. In additional to Bahl’s Redmond-based team, researchers from Microsoft research labs in Asia, Cambridge and Silicon Valley will also present papers at SIGCOMM.

Expanding Wi-Fi Range by Three Times

One example of Microsoft Research’s efforts has to do with adapting channel width for Wi-Fi networks. Twenty-year-old Wi-Fi standards dictate the width of a channel at 20 Mhz, but the paper “A Case for Adapting Channel Width in Wireless Networks being presented by Microsoft and the University of California, Santa Barbara, challenges that standard with the theory that different bandwidths are more suitable for different tasks.

“This paper asks, what if you have variable width?” Bahl says. “We show how to do it, and the properties you get, which are very good for the high-layer applications to exploit.”

The research has the potential not only to affect the evolution of networking products, but current ones as well. One scenario is with Zune’s social networking capability to discover other Zunes in the immediate area.

“The problem is the range of the Wi-Fi,” Bahl says. “With the technology Microsoft Research built, the Zune’s range can increase three times. With this simple software update, users can discover a lot more Zunes, and their social network becomes a lot more interesting.

Massive Scaling of Multiplayer Online Games

Another technology Microsoft Research is presenting this year, in conjunction with partners at Carnegie Mellon University, is called Donnybrook. The paper “Donnybrook: Enabling Large-Scale, High-Speed, Peer-to-Peer Games” describes an approach to peer-to-peer gaming networks that takes new steps toward allowing massive scalability.

Srinivasan Seshan, a professor at Carnegie Mellon University, led a team that collaborated with Microsoft Research on the Donnybrook project.

“This is a recent trend with multiplayer online games,” he says. “One weakness with previous systems is that you can’t support large open spaces or large battlefields where there are a lot of people directly interacting with each other. This paper, which has been borne out of the collaboration with Microsoft, discusses how to address a particular aspect of the game where you really want a large interacting body of people.”

The paper relies on user studies that show people can only keep track of a certain number of objects in their environment. Distilling the game down to the five or 10 objects that users really care about within a huge battlefield allows the network to handle the large overhead involved. The end result is a game that can scale potentially to thousands of players.

“Given today’s communication bandwidth available to the typical home user as well as the update for these particular games, we believe we can get to a typical, realistically sized battle,” Seshan says.

Datacenters Power Innovation

Much of the research being presented this week involves optimizing large datacenters to reduce cost and power consumption while increasing performance, a topic closely followed by researchers at Microsoft.

“Microsoft’s overall technology vision is aligned with the power of “software + services,” and datacenters are in the middle of it,” Bahl says. “Microsoft has invested heavily in creating datacenters, as have others in the industry, but to keep them running at an optimal level is a constant challenge.”

One of the more radical ideas being put forward by Microsoft Research at SIGCOMM this week is the DCell technology that fundamentally rethinks datacenter architecture. The collaborative work from Microsoft Research Asia, Tsinghua University, and UCLA is described in the paper “DCell: A Scalable and Fault-Tolerant Network Structure for Data Centers.”

“The idea is to get rid of many of the ‘switches’ inside datacenters, do the routing at the edges, the servers themselves, and let them create a mesh inside the datacenter so packets can move quickly,” says Bahl. “It’s a notion that could reduce the cost of building and operating large datacenters tremendously.”

According to Bahl, this might seem like an esoteric, IT problem, but it’s something that directly affects millions of people.

“The reality is that end users are the ones who are actually using the services,” Bahl says. “When you do a Live search, that search is executed inside a datacenter. When you look at Live Maps, or run Exchange from a Web site, all of these services are running on computers inside the datacenters. How do we ensure that when you click on something, immediately the answer comes back? How do we bring the cost down? How do we increase reliability?”

Once you do those things, he says, the end user starts to benefit. From there, more people use the service, and the ecosystem of new services now gets an incentive to evolve.

Related Posts