Paper Details

Communicating Process Architectures (CPA)
 Title: Building a C++CSP Channel Using C++ Atomics: a Busy Channel Performance Analysis
 Conference: Communicating Process Architectures 2017
 Authors: Kevin Chalmers
School of Computing, Edinburgh Napier University
 Abstract: Mutex-based implementations of synchronous channels are slow. This work investigates atomic operations as a technique to improve communication efficiency between two threads via a busy channel. Such a channel provides faster communication than a mutex-based one where the number of threads is identical to the hardware concurrency. To evaluate communication performance, a communication time benchmark is used alongside a selection time benchmark. The communication time benchmark is scaled to evaluate the impact of thread counts greater than the available hardware. Results show that an above ten-times improvement in communication time is possible when the hardware supports the threads fully. The improvement drops as further threads are added to the system due to operating system scheduling taking over the determination of thread activeness. Selection time similarly shows improvement when thread count matches hardware, but likewise reduces as thread count increases. We can conclude that a busy channel is useful in an environment where the thread count matches the available hardware, which is of interest to parallel application developers or control systems with similar properties. 
Files:


BibTeX Entry


Full paper


Presentation