HaLock: hardware-assisted lock contention detection in multithreaded applications

  • Authors:

  • Yongbing Huang;Zehan Cui;Licheng Chen;Wenli Zhang;Yungang Bao;Mingyu Chen

  • Affiliations:

  • Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China;Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China;Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China;Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China;Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China;Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China

  • Venue:
  • Proceedings of the 21st international conference on Parallel architectures and compilation techniques
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Multithreaded programming relies on locks to ensure the consistency of shared data. Lock contention is the main reason of low parallel efficiency and poor scalability of multithreaded programs. Lock profiling is the primary approach to detect lock contention. Prior lock profiling tools are able to track lock behaviors but directly store profiling data into local memory regardless of the memory interference on targeted programs. In this paper, we find that the memory interference is non-trivial and can significantly affect programs' execution as thread number increases. To address this problem, we propose a hardware assisted lock profiling mechanism (HaLock) which leverages a specific hardware memory tracing tool (HMTT) to record large amount of profiling data with negligible overhead and impact on even large scale multithreaded programs. Experimental results show that HaLock incurs only about 14.8% additional L3 cache misses and 34.3% extra memory requests for a lock-intensive workload (bodytrack of PARSEC benchmark) with 512 threads, while the previous state of the art low-overhead approach causes 25.9% additional L3 cache misses and 73.8% additional memory requests. Compared with HaLock's profiling data, we find that the lock behaviors obtained by the state of art lock profiling tools have substantial distortions, resulting in non-negligible inaccuracy problems.