Hajek wins ACM SIGMETRICS Achievement Award
We rely on networks, both fiber optics running through the ground and mobile signals stretching through the sky, that allow us to communicate almost without delay across the globe.
Though our only experience of these networks is generally scrounging for Wi-Fi at a cafe, the networks we rely on are subject to immense amounts of random interference: If a thunderstorm shuts down a pathway through a network, or a football game has just ended and a torrent of fans are trying to get online at the same time, networks have to be able to deal with these random loads efficiently and fairly.
Professor Bruce Hajek has won this year’s Achievement Award from the ACM’s special interest group on performance evaluation, or SIGMETRICS, for his research into theorizing how networks should prioritize their resources to deal with random, chaotic conditions.
“Randomness is a huge part of designing networks, and in making sure that when they fail, they fail in a pattern of what’s called graceful degradation, as opposed to sudden catastrophe,” Hajek said. “Networks are exposed to all kinds of randomness: the randomness of the atmosphere affects mobile signals, the randomness of how much demand people are placing on the network at any given time. We have to find ways to design networks that operate reliably when conditions are favorable, and can gracefully degrade when they aren’t.”
The award recognizes his contributions “to stochastic analysis, optimization and control, which have provided unique mathematical insights into the performance of the Internet, wireless networks, and peer-to-peer networks such as BitTorrent,” according to the SIGMETRICS website. SIGMETRICS is the group within ACM that deals specifically with finding ways to measure how powerful computers and networks are, and how they can be improved. The group gives an achievement award yearly to recognize an individual who has made lasting contributions to the field.
Much of Hajek’s work is theoretical; he is involved in creating mathematical models to describe how networks function and coming up with control policies that other researchers could turn into workable protocols and algorithms. Hajek studies how sensitive, vital networks can retain reliability when so much of their functioning is subject to random events. For him, designing networks without knowing where outages and conflicts will arise is quite like playing a game of poker, and placing bets on where one thinks the best tricks are likely to form. In keeping with this approach, one of Hajek’s major contributions was integrating game theory research into network management.
“In a way, governing network resources is a lot like a gambling: you don’t always know where demand will pop up or where random outages can occur,” Hajek said. “It’s just like when you’re playing poker and you have to plan not just around what you know, but on what you can infer about the other players’ hands and their behaviors, and on this limited information, you have to come up with a way the network can come up with predictions and probabilities for each event and deal with them,” Hajek said.
Hajek theorized several ways networks could react to failures caused by randomness, especially the highs and lows in user demand, which is called “burstiness.” One way was developing mathematical models to figure out several efficient paths data could take through a network. He also pioneered an efficient, fair way to govern what user data is let through data conflicts in what’s called the “flow control” of a network.
“Let’s say you’re at a Starbucks, and you’re trying to access the Wi-Fi to send someone an e-mail,” Hajek said. “When you press ‘send,’ the message goes through a protocol of being converted into a series of data packets, and then gets sent to the router. If your packet gets there at the same time as another guy’s packet, how do you decide which one goes through? Our way is similar to two cars meeting at a stop sign. You both wait a random amount of time, and then try again.”
Apart from studying randomness, Hajek was also honored by the award for his research into peer- to-peer, or P2P networks. These are ad hoc connections set up throughout the Internet between clients that transfer data to each other, over services like BitTorrent. Traditional scholarship tends to ignore these types of networks because of their reputation for transmitting illicit software like pirated video games, but Hajek noted 50 percent of traffic over the Internet is over P2P networks, and simply by virtue of how prevalent they are, they merit study. Hajek is working on understanding these networks, and making predictions for how they could react to changes in the Internet ecosystem.
“In particular what interests me about P2P networks like BitTorrent are how they work, and if they could work better,” Hajek said. “I look at BitTorrent and analyze why it’s so effective, and then my research asks questions like ‘If you changed one thing here, would it work better or worse? In five or 10 years when another protocol comes along, how will that affect this service?’ So much of SIGMETRICS research is asking ‘What if?’ questions, making predictions to answer if the network changed in some drastic way, how could we react?”
As the recipient of the achievement award, Hajek will give a talk at the next SIGMETRICS conference in Portland, Ore., taking place June 15-19. Hajek was “delighted and surprised” to receive news of the award, and was especially proud that it came from his own colleagues in the SIGMETRICS field.
“It’s very nice to be appreciated, and I’m grateful to all of my colleagues and students over the years that have informed my research,” Hajek said. “The best part of the award for me is the familiarity that the community has with my research: it’s great to get an award from people that really know your work.”