Logo: University of Southern California

Developing Data Center Infrastructures for Cloud Computing

Minlan Yu, a recent recipient of the NSF CAREER award, redefines the cloud computing industry
August 26, 2015 —
Minlan Yu

Behind the ephemeral "cloud" of cloud computing, which supports our everyday activities from posting on Facebook to watching NetFlix, there lies a very big network infrastructure with hundreds of thousands of computers. While such large networks are immensely beneficial to many users, they are also immensely challenging for network administrators who have to ensure myriad requirements, such as power saving, security and efficiency.

USC Viterbi’s Minlan Yu is focusing on redesigning network infrastructure to make the network management in the clouds fundamentally easier. Yu earned her degree in computer science and math at Peking University in China. She continued her graduate studies at Princeton University and her post-doctoral research at UC Berkeley before joining as an assistant professor in USC Viterbi’s Department of Computer Science.

Yu was recently awarded the prestigious National Science Foundation Career Award. This award is specifically given to junior faculty members who have exemplary research and a strong desire to teach and develop their field as a teacher-scholar. This grant will support her research on building a programmable measurement architecture, which will provide comprehensive visibility to network administrators with minimal resources. It will also support her in developing new courses in software-defined networking and cloud computing.

Currently, Yu is working closely with major cloud computing companies to redesign the management framework. She designs new abstractions for network administrators to freely express their management goals without worrying about the underlying complex implementation. She then introduces new management primitives at hosts and switches that can support a variety of management requirements and designs novel algorithms to map the high-level management goals to the low-level primitives efficiently. Her work will enable new cloud services for our everyday life, make cloud applications like Facebook and NetFlix run more efficiently, and make the cloud infrastructure more reliable.

As quoted in her SIGCOMM Doctoral Dissertation Award, “Minlan’s research goes full circle, from important practical problems, to creative data structures and algorithms, to rigorous analysis and modeling, and finally to the design and implementation of novel systems.” For example, in 2011, her scalable network-application profiler (SNAP) has been deployed in Microsoft’s production data center with over 8,000 servers and over 700 application components, which has helped network administrators and application developers uncover multiple major performance problems in application software, the network stack on the server, and the physical network. More recently, she invented OpenSketch, a new set of hash-based primitives at switches that are not only flexible in supporting different measurement queries but also easy to implement with commodity switch components. She has been collaborating with a startup Barefoot Inc. in bringing OpenSketch and other novel measurement data structures to production.

Yu hopes to majorly impact the computer science industry to “fundamentally change the management practice in today's data centers, and make network visibility as the first class citizen in cloud infrastructure.” She will also be continuing her work with major cloud computing companies to remodel the way applications interact with underlying infrastructure and how networks integrate with computing and storage resources. Within a few years, we can expect Yu’s work to revolutionize the cloud computing industry to promote automatic procedures that will regulate performance diagnostic and traffic engineering with pristine accuracy and efficiency.