When prospective students visit the MIT campus, they might see and hear about the celebrated faculty, the vibrant student life, or the modern facilities. They might not notice the hundreds of postdoctoral associates who, like the worker bees in a hive, collaborate in much of the research that has earned MIT its reputation as a premier science and technology institution. However, once you’ve talked with a post-doc, their passion and talent truly make them impossible to forget. LIDS doctoral associate Atilla Eryilmaz speaks candidly about his current research, the post-doc experience, and where the theory of networking needs to go next.
Growing up in Istanbul, Turkey, Atilla planned to become an engineer “as far back as I can remember,” a goal he accomplished when he completed both his M.S. and Ph.D. at the University of Illinois at Urbana Champaign before coming to LIDS. Cambridge may be far from Istanbul, but since his arrival in September 2005, Atilla has felt increasingly at home here. The Boston area is “more like Istanbul” than rural Illinois, and the Charles River reminds him of the Bosphorus Strait in Turkey, which separates Europe and Asia. Of his career choice, he says, “It seemed like engineering had these two sides, with theory on one side and application on the other. The boundary of the two had an interesting feel.” His current research as a LIDS postdoctoral associate, conducted with Professors Asuman Ozdaglar and Muriel Médard, continues to walk the line between theory and application, by examining control of wireless data networks from both theoretical and technological perspectives.
When seeking out the “fundamental bounds” of wireless networks, Atilla wants to discover what the absolute limits of wireless networks are, such as how much data they can handle at a time, and how long it will take to process that data. These “performance criteria” are known as “throughput,” sometimes also called “capacity,” which is the amount of data capable of being transmitted per unit of time, and “delay,” which is the time elapsed between the transmission and reception of data. In networking, throughput and delay are at odds with one another. “As you get closer to the [channel] capacity, you have to wait longer before [the data] can be served,” Atilla explains. For example, just because a sender is capable of quickly uploading and sending a large file, such as a high-resolution digital photo, doesn’t mean the data will reach its recipient in the same timely fashion. The file may have to wait in several network queues before reaching its destination. As we achieve higher throughput, more delay is usually implied. Atilla calls this the “throughput-delay tradeoff curve.” Whether a network favors throughput or delay depends on the application, or what the network will be used for. “The stringency of your delay constraints is important,” Atilla says. While you might not care if an email took two or three minutes to reach its recipient, you would certainly be upset if there was a minute delay during your cellular phone conversation. This is where Atilla’s theoretical and practical work intersect. He explains that if he is able to discover the “fundamental bounds” of a network, he can then compare those bounds to the actual performance of the network to see how good it is. For example, if he knows what the absolute minimum amount of delay is, he knows that, in theory,whether or not the performance of a specific cellular network can be improved.
While the most recent algorithms that Atilla has helped to develop have been proven “to achieve full capacity” of a given network, meaning the maximum amount of throughput, his research is far from over. He is also looking at how to better allocate resources and improve overall performance for specific wireless application scenarios. Giving the example of users of wireless laptops in a library, Atilla asks, “How do they communicate, how do they allocate their scarce resources?” Because the wireless medium is “fluctuating, not necessarily constant,” network users need to be able to transmit information using limited amounts of resources, such as energy and bandwidth. One way to accomplish this is through network coding. When data is sent over a digital channel, it is sent in small chunks, called “packets,” rather than as a continuous stream of information. Network Coding, Atilla explains, “is basically a different way of looking at data packets.” Until relatively recently, packets were sent intact over a channel to their destination. The groundbreaking idea of Network Coding was to take pieces from several different packets and to send this mix over the channel in several transmissions. The receiver then decodes the packets and rearranges them into their original order. “Before Network
Coding,” Atilla says, “if you can’t send one of the packets, that packet is lost. Now you have the choice of spreading it around in many different transmissions and paths to be able to decode at the end.” In this way, there is a better chance that the essence of the message will be retained, despite the limitations of the network. Working with Professors Ozdaglar and Médard, Atilla developed an algorithm which uses network coding to limit delay, and has applied for a patent for it. Some potential applications for this research are peer-to-peer networks, which are networks such as university networks that enable users to trade or share data, and in “cellular scenarios,” downloading large files such as songs or videos to cellular phones.
Atilla’s current work utilizes tools from both of Professors Ozdaglar and Médard’s specialties, optimization and network coding. While most MIT post-docs work closely with just one faculty member, Atilla feels lucky to work with two. “It’s much more creative,” he says, because he is able to access “a larger, more diverse pool of opinions” and this “creates a very synergistic environment.” Although combining these fields is challenging, this type of challenge is an inherent element of the postdoc experience. As a Ph.D. student, Atilla says, “you are like a second captain to a ship, where you learn how to steer it through first-hand observation…. Here with the pos-tdoc opportunity, you start to navigate your own boat in this ocean of the unknown.”
Where will Atilla steer his ship next? The fundamental questions of networking are what currently interest him most. In the future, he plans to investigate “whether there exists a unified theory of networking” and “whether one can really achieve a unification of [networking] theory and practice.” This approach fits in perfectly at LIDS, where Atilla says “people are trying to solve the most fundamental question that they can reach out for. The next step is what excites me,” Atilla says, “the boundary of my knowledge at the time.”