报告人：美国麻省大学 盛波副教授
报告时间：6月6日上午9:30
报告地点：计算机楼313会议室
Abstract: Speculation is a common mechanism that identifies‘slow’ node in a cluster and starts redundant tasks on other nodes to guarantee the reliability. This talk introduces our recent work on improving the efficiency and security of speculation particularly for the
MapReduce/Hadoop computing platform. In the first work, we target on a heterogeneous cluster, and our general goal is to accurately and quickly identify the straggler nodes during the job execution. Our approach significantly reduces unnecessary speculative executions, and detects the real stragglers at an early stage. The experimental results show that our solution is efficient and effective when handling the speculative executions. In the second work, we explore the vulnerabilities of the existing speculation scheme, and develop an effective attack launched from a compromised node. Our experiments on real systems show that this attack greatly slows down the job executions in the native Hadoop system even with some basic defense mechanisms.Then, we present a mitigation scheme that protects the system from such attacks. While causing a minor overhead in normal circumstances, our solution can keep the whole cluster running efficiently under this attack from the compromised internal node.
Bio: Bo Sheng is an Associate Professor in the Department of Computer Science at University of Massachusetts Boston. He received his B.S. from Nanjing University, and his Ph.D. from the College of William and Mary, both in Computer Science. His research interests include big data analytics, mobile computing, Internet of Things / RFID / sensor networks, wireless networks, and security. His research work has been supported by National Science Foundation (NSF) grants, Amazon Research Grant, Microsoft Research Grant, and Google IoT Technology Research Award.
