Big data applications increasingly dominate every aspect of our lives. Healthcare, finance and transportation are only a partial list of a growing number of fields where big data is being used. These big data applications, often running in the cloud, require significant computing resources and drive the research of high-performance systems. But how do these applications look like from a networking perspective? And how does the network affect their performance? This talk presents a networking perspective on big data applications. We study the effects of networking on the performance of big data applications, and provide a breakdown of the basic latency components within a high performance system. Based on our results, and using the network profiles of different big data applications, we identify performance bottlenecks and propose new system architectures for big data applications.

Noa Zilberman is a Leverhulme Early Career Fellow at the University of Cambridge Computer Laboratory. Her research focuses on high performance systems, combining both computer networks and computer architecture. Dr Zilberman is the chief architect of the NetFPGA-SUME project, an open source platform for rapid prototyping of high-performance networking devices. She has over 15 years of industrial experience, in design, management and architecture roles. Amongst others, she led the hardware design of the world's first 100Gbps silicon switch, and the architecture of the first device enabling scalable systems up to 100Tbps. Zilberman is the PI on the CAND and PERF projects, and was a researcher on multiple EPSRC and EU projects. She is a senior member of IEEE and a member of ACM, Usenix and BCS.