Jellyfish: Networking Data Centers, Randomly

Ankit Singla, University of Illinois at Urbana-Champaign
Host: Randal Burns

Large Internet services, “big science”, and increasingly, industrial R&D, are all powered by warehouse scale computing — tens of thousands of servers connected over a network. Increasing parallelism and big data analytics require that the network provide high throughput connectivity. In this talk, I will describe Jellyfish, our proposed design for such a network. Jellyfish uses a random graph topology, allowing construction at arbitrary sizes, easier network growth over time, and integration of newer and more powerful network equipment as it becomes available — practical problems that rigidly structured traditional networks fail to address. Surprisingly, Jellyfish also beats state-of-the-art real-world designs on throughput, by as much as 40%. In fact, we show that Jellyfish networks achieve near optimal throughput, i.e., one cannot build, using the same networking equipment, any network that provides much higher throughput.

Speaker Biography

Ankit Singla is a PhD Candidate in Computer Science at the University of Illinois at Urbana-Champaign. He received his Bachelors in Technology (Computer Science) at IIT Bombay, India, in 2008. He is a winner of the 2012 Google PhD Fellowship. These days, he is refining a plan for building a speed-of-light Internet, which he loses no opportunity to talk about.