Parallel Processing Institute, Fudan University

Postal Address: RM 320, Software Building, 825 Zhangheng RD

Shanghai, P.R. China 201203

Email: zwxiao at fudan dot edu dot cn

I am a third-year master student under the supervision of Prof. Binyu Zang and Dr. Haibo Chen. My research interests are mainly on distributed systems. I share some interesting stuff on my blog.

• M.Sci. in Software School, Fudan University, P.R.China (2009.9 ~ Now)
• B.Eng. in Software School, Fudan University, P.R.China (2005.9 ~ 2009.6)
• B.Eng. in Computer Science, University College Dublin. (Joint-degree) (2005.9 ~ 2009.6)

• MapReduce and Hadoop on multicore clusters
• NoSQL storage and HBase

• (07/2011 ~ 10/2011) Visiting student, Rice University
  • Project Description: Implementing ZPL programs on MapReduce and HBase cluster. ZPL is an array language for scientific computation on parallel machines.
  • Current Status: Under implementation based on our initial design.

• (07/2010 ~ Now) Distributed NoSQL storage performance
  • Project Description: Firstly we evaluate HBase with some real applications, and then study the HBase implementation to seek for optimization or better design.
  • Current Status: HBase falsely sequetializes irrelative put operations and thus causes long lock acquiring time. We’re attempting to parallelize these operations.

• (03/2008 ~ Now) MapReduce performance on cluster with multi-core nodes
  • Project Description: We studied the MapReduce performance on multi-core cluster. We tried to exploit the power of multi-core machines to maximize MapReduce efficiency.
  • Results: The poor performance of Hadoop mainly comes from object creation, large memory footprint, naive data structure and lack of good multithreading support. We have implemented a prototype system Azwraith, which outperforms the original MapReduce implementation from 1.4X to 3.5X.
  • Current Status: Identify the interrupt handling issues in Hadoop. Propose a federated schedule model for MapReduce.

• (09/2009 ~ 11/2009) Worked on understanding the characteristics of IO in Hadoop and surveyed applying SSD to Hadoop and HBase
  • Results: Mixed workloads incur a random disk access pattern in Hadoop. Using SSD as a disk cache and intermediate storage would gain a good speedup.

• (03/2009 ~ 04/2009) Evaluated parallel shared-memory applications with MapReduce cluster
  • Results: Frequent communications in this kind of applications cause significant overhead on cluster. We proposed some extensions to the MapReduce model in our APPT’09 paper.

• Zhiwei Xiao, Haibo Chen, Binyu Zang and Bo Huang. A Hierarchical Approach to Maximizing MapReduce Efficiency. In the Proceedings of the 20th International Conference on Parallel Architectures and Compilation Techniques (PACT’11, poster). 2011. [pdf]
• Chao Zhang, Chenning Xie, Zhiwei Xiao, and Haibo Chen. Evaluating the Performance and Scalability of MapReduce Applications on X10. In the Proceedings of the 9th International Symposium on Advanced Parallel Processing Technologies. 2011.
• Shengkai Zhu, Zhiwei Xiao, Haibo Chen, Rong Chen, Weihua Zhang and Binyu Zang. Evaluating SPLASH-2 Applications Using MapReduce. In the Proceedings of the 8th International Symposium on Advanced Parallel Processing Technologies(APPT'09). 2009. [pdf]

• Spring, 2011 Computer System Engineeringhttp://ppi.fudan.edu.cn/cse/2011/Overview.html
• Autumn, 2010 Artificial Intelligence
• Autumn, 2009 Compiler
• Autumn, 2008 Operating System

• First Grade Scholarship for graduate students (2010).
• Outstanding Contribution Award at PPI (2009).
• First Grade Scholarship for Excellent Freshman (2009).
• Second Grade Scholarship (2006, 2007, 2008).