Abstract: Exisiting big data systems, such as MapReduce and Spark, are designed for programmability, scalability and fault tolerance at the price of absolute performance. We discuss the fault tolerance and performance and conclude that they are not mutually exclusive. We give a high performance distributed graph processing system, as an example of a big data system designed for performance. It only requires 10% memory required by GraphX and is 50X faster than GraphX for typical graph processing workloads.

Short Bio: Wenguang Chen is a professor in Department of Computer Science and Technology, Tsinghua University. His research interest is in parallel and distributed systems and programming systems. He received the Bachelor’s and Ph.D. degrees in computer science from Tsinghua University in 1995 and 2000 respectively. Before joining Tsinghua in 2003, he was the CTO of Opportunity International Inc. He was appointed as the associate head of Department of Computer Science and Technology from 2007 to 2014. He has published over 50 papers in international conferences and journals like OSDI, PLDI, Supercomputing, EuroSys, USENIX ATC, OOPSLA and ICSE. He is a distinguished member and distinguished speaker of CCF( China Computer Foundation). He is an ACM member, Vice Chair of ACM China Council, and Editor-in-Chief of Communications of ACM( China Edition).