wPerf: Generic Off-CPU Analysis to Identify Bottleneck Waiting Events

Speaker Name: 
Yang Wang
Speaker Title: 
Assistant Professor
Speaker Organization: 
Ohio State University
Start Time: 
Friday, August 17, 2018 - 10:30am
End Time: 
Friday, August 17, 2018 - 11:30am
Location: 
Engineering 2, RM 399
Organizer: 
Chen Qian

Abstract:

wPerf tries to identify waiting events that are limiting the maximal throughput of a multi-threaded application. To achieve this goal, we not only need to understand an event’s impact on threads waiting for this event (i.e. local impact), but also need to understand whether its impact can reach other threads that are involved in request processing (i.e. global impact).

To address these challenges, wPerf computes a waiting event’s local impact with a technique called cascaded re-distribution; more importantly, wPerf builds a wait-for graph to compute whether such impact can indirectly reach other threads. By combining these two techniques, wPerf essentially tries to identify events with large impacts on all threads.

We have applied wPerf to a number of open-source multi-threaded applications. By following the guide of wPerf, we were able to improve their throughput by up to 4.83 times. The overhead of recording waiting events at runtime is about 10% on average.

Bio:

Dr. Yang Wang is currently an assistant professor at the Ohio State University. Previously he received his bachelor’s and master’s degrees from Tsinghua University and his Ph.D. degree from the University of Texas at Austin, under the advice of Dr. Lorenzo Alvisi and Dr. Mike Dahlin. His major research interests are in distributed systems, in particular fault tolerance, scalability, and performance analysis.