Presentation
(A06) Poster from Korea: Application Characterization by using Hardware Performance Counters with Data Mining
SessionHPC in Asia Poster Session
Event Type
HPC in Asia
Performance Analysis and Optimization
Performance Tools
Scheduling
TimeWednesday, June 27th10am - 11am
LocationAnalog 1, 2
DescriptionWith the advent of processors such as Intel Xeon Phi and NVDIA GPUs which has more number of cores than existing multicore processors and additional high-bandwidth memory, research on performance analysis of new system has received increasing attention from high performance computing community. In addition, computational scientists and engineers who are eager to obtain the best performance of their scientific applications need to understand application characteristics in the new system. Hardware Performance Counter which is included in modern processors is mainly used to monitor performance-related hardware events. However, a large number of hardware events that are provided differently by vendor make it difficult to find performance-related insights.
In this paper, we propose a method to simply and quickly classify application characteristics by using a data mining tool without understanding the correlation between hardware events. Our approach has 3 steps which are event collection, profiling, and application characteristics analysis. We applied the proposed method to NAS Parallel Benchmarks (NPB) in a system based on Intel Knights Landing processors. Results of classification of application characteristics indicate that our classification were the same as the authorized NPB categories. We show the effectiveness of the proposed scheme in a case study on analyzing the degree of interference between application characteristics.
In this paper, we propose a method to simply and quickly classify application characteristics by using a data mining tool without understanding the correlation between hardware events. Our approach has 3 steps which are event collection, profiling, and application characteristics analysis. We applied the proposed method to NAS Parallel Benchmarks (NPB) in a system based on Intel Knights Landing processors. Results of classification of application characteristics indicate that our classification were the same as the authorized NPB categories. We show the effectiveness of the proposed scheme in a case study on analyzing the degree of interference between application characteristics.
Poster PDF
