Stanford Networking Seminar

12:15PM, Thursday April 12, 2012
Gates 104

Packet Transport Mechanisms for Data Center Networks

Mohammad Alizadeh
Stanford University

About the talk:
In recent years, large investments have been made in massive data centers to support services such as search, social networking, cloud computing, and Internet commerce. This has spurred a lot of interest in the industry and the research community in innovation for reducing costs and improving performance of data center networks. One of the most crucial components of a data center network (indeed, of any network) is its transport mechanism: the method by which data is transferred from one server to another. In this talk, I will present measurements from a 6000 server production cluster that reveal impairments with today's state-of-the-art Transmission Control Protocol (TCP) in data center environments. The impairments are rooted in TCP's demands on the limited buffer space available in commodity data center switches. I will then discuss the design, evaluation, and mathematical analysis of Data Center TCP (DCTCP) --- an enhancement of the TCP algorithm that will be shipping with Windows 8 Server. DCTCP enables the construction of large-scale, high-performance data center networks from commodity switches . Finally, if time permits, I will discuss HULL (for High-bandwidth Ultra-Low Latency), an architecture for delivering baseline fabric latency (only propagation + switching) and high bandwidth utilization in a data center fabric.

About the speaker:
Mohammad Alizadeh is a PhD candidate in the Electrical Engineering department at Stanford University, working with Balaji Prabhakar. His research interests are broadly in network systems and algorithms. Recently, he has focused on designing high performance packet transport mechanisms for data center networks. Prior to Stanford, Mohammad completed his Bachelors degree at Sharif University of Technology.