If the Internet Were a Plane

Professor Isaac Keslassy is the Louis and Miriam Benjamin Chair in Computer-Communication Networks in the Andrew and Erna Viterbi Faculty of Electrical and Computer Engineering. Together with his students, he focuses on how to make the internet faster and better.  

The internet relies on many server farms called data centers. For instance, when an item is ordered on Amazon, the order goes to a data center, where several servers process it. Prof. Keslassy looks at the networks that connect the servers within these data centers so orders can be processed more quickly. His team also works to improve the networks so browsing can be more efficient. 

Additionally, the data centers are interconnected in a global network called the cloud. His team of researchers looks at the cloud and tries to analyze whether its properties can help build a better internet. “In fact, when you access a server somewhere, we could say, figuratively speaking, that the internet lets you board a plane and go there relatively cheaply,” explained Prof. Keslassy.  

“However, it is sometimes a bit slow, just as planes can be delayed due to high load or bad weather. Few people have realized it, but the cloud can provide private jets that are faster but more expensive than the regular plane flights provided by the internet. I look at whether and how we can combine both, for example, by only using private jets when the usual flights are delayed.” 

To extend his airplane metaphor, Prof. Keslassy’s research considers global applications that require many network connections around the world, exactly as a large company requires many flights for its employees around the world. He then optimizes when and how to use expensive private jets through the cloud when regular flights are not good enough. 

Prof. Keslassy compares internet slowdowns and outages to elephants boarding a plane. “On the internet, there is no easy way to distinguish between users who are regular passengers and users that are elephants, and as a result, internet planes often get stuck.” 

His research introduced a fast way to tell whether a passenger is a conventional one or an elephant. This new method leverages techniques from machine learning with improved techniques for optimization.  

Finally, internet router buffers, which Prof. Keslassy says function as internet airports, are expensive to build. He is seeking a way to provide fast connections to plane travelers by using only very small airports, without getting these airports clogged due to their limited capacity. Researchers have been looking for such a way for the past 20 years.   

“It turns out these techniques are very promising and may be a way to utilize smaller airports.”