Talk given by Prof. Heather Zheng: Network Design for Big-Data
Talk: Network Design for Big-Data
Speaker: Prof. Heather Zheng, U. C. Santa Barbara.
Time: 11:00AM, December 6, Thursday
Venue: Demo Room (14366, 14F)
讲座两个部分，第一部分是用flexible wireless进行data center通信建设；第二部分是scalable social network analysis 社交网络计算。
Part1.Network Design for Big-Data
- double fee
challenge #1 Link Blockage
challenge #2 Interference Footprint
challenge #3 Robustness ti Alignment Error
Part2.enable scalable graph processing for today’s massive online social networks
distance ranked social
rank search result based on social distance
Key Limition Today:Scale
A drastically different alternative
- O(1) query time
Approach for Embedding
- converge quickly
- optimize accuracy
shortest path search
Rigel 12倍于 Sketch Tree
The arrival of big-data applications has created significant challenges to network design. For example, batched data processing jobs are straining network capacity in data center networks, causing unexpected outages and service downtimes, while social network companies struggle to manage millions of users and billions of user events and queries in real time. In both cases, network architects must find new, novel ways to support traffic/service demands that display complex structure and vary significantly with time.
In this talk, I will discuss two of our recent works on big-data.
- First, I will talk about challenges in dealing with dynamic traffic hotspots in data centers, and our solution using flexible wireless interconnects to augment wired connections. We propose 3D beamforming 60GHz links that leverage ceiling reflection to overcome key challenges in deploying wireless links, creating parallel wireless links connecting any racks at wired data rates.
- Next, I will discuss our effort to enable scalable graph processing for today’s massive online social networks. We propose graph coordinate systems, a new approach that accurately approximates node distances in constant time by embedding graphs into coordinate spaces. Our design not only provides accurate results for massive graphs (43 million nodes), but also is naturally parallelizable across computer clusters. It answers node-distance queries in 10’s of microsecond, and produces shortest path results up to 18 times faster than prior solutions with similar accuracy.
- Finally, I will conclude with a brief summary of other ongoing projects.
Haitao (Heather) Zheng is currently an Associate Professor at the Computer Science department, U. C. Santa Barbara. She completed her M.S. and Ph.D. degrees in Electrical and Computer Engineering at Univ. of Maryland, College Park (1998, 1999) and her B. S. degree from Xi’an Jiaotong University (gifted class). She is a recipient of the MIT Technology Review’s TR-35 Award (Young Innovators Under 35) and the World Technology Network Fellow Award. Her work has been covered by media outlets such as New York Times, Boston Globe, MIT Technology Review, and Computer World. Her work on cognitive radios was named as the top-10 Emerging Technologies by MIT Technology Review in 2006. Her research spans areas of wireless networking, distributed systems, economics, data-intensive computing, and social networks.