In this age of data explosion, the ability to perform computation on large amounts of data within an acceptable response time is quite essential to many business companies. Hadoop answered this calling by implementing a scalable map-reduce framework, which enables people to process massive volumes of data in parallel upon a cluster of low-end commodity machines.
Although Hadoop does a great job in distributing workload and collecting results, its ability to handle node failure is quite limited, which often results in significant increase in job completion time. Hao addressed this problem by adding an adaptive failure detection mechanism to Hadoop. Experiments showed that his improvement reduced job completion time in the presence of node failure. The paper is published at the 2011 IEEE Asia-Pacific Services Computing Conference in Jeju, Korea
The above pricture was shot when he was at a keynote session. For more information about his research, check his poster at here or his paper at here. Please feel free to send your comments and suggestions to him.