Difference between revisions of "Micro-threading"

From Free Pascal wiki
Jump to navigationJump to search
Line 1: Line 1:
 
==Introduction==
 
==Introduction==
There are some situations where applications need to execute lots of asynchronous concurrent operations. OS-level threads could be used for this purpose but an application can create only a limited count of threads. In matter of CPU utilization there are no significant performance gain if more threads is created then physical cores CPU have. In fact threads could be used for parallel processing on a multicore CPU and for writing more readable code. OS-level threads should be used mainly used for parallelism and better CPU utilization. If a programmer needs to write code which will be called asynchronously, then OS-level threads will be rather expensive in perspective of system resources.
+
There are some situations where applications need to execute lots of asynchronous concurrent operations. OS-level threads could be used for this purpose, but an application can create only a limited count of threads. As for CPU utilization there is no significant performance gain if more threads are created than physical cores available on a CPU. In fact threads could be used for parallel processing on a multicore CPU and for writing more readable code. OS-level threads should be used mainly used for parallelism and better CPU utilization. If a programmer needs to write code which will be called asynchronously, then OS-level threads will be rather expensive in perspective of system resources.
  
 
===CPU context switching methods===
 
===CPU context switching methods===
* Cooperative (use of explicit Yield call)
+
* Cooperative (use of explicit [[Yield]] call)
 
* Preemptive (periodic timer based)
 
* Preemptive (periodic timer based)
 
* Combined
 
* Combined
Line 11: Line 11:
 
* Fast switching, creation, destruction
 
* Fast switching, creation, destruction
 
* Automatic thread pool management by physical CPU core count
 
* Automatic thread pool management by physical CPU core count
* Ability to run in main loop only (without TThread instances)
+
* Ability to run in main loop only (without [[TThread]] instances)
 
* Provide own synchronization tools (Yield, Sleep, CriticalSection, Semaphore, Mutex, WaitForMultipleObjects, Queues, Synchronize, ...)
 
* Provide own synchronization tools (Yield, Sleep, CriticalSection, Semaphore, Mutex, WaitForMultipleObjects, Queues, Synchronize, ...)
 
* Priority control
 
* Priority control

Revision as of 23:16, 31 October 2016

Introduction

There are some situations where applications need to execute lots of asynchronous concurrent operations. OS-level threads could be used for this purpose, but an application can create only a limited count of threads. As for CPU utilization there is no significant performance gain if more threads are created than physical cores available on a CPU. In fact threads could be used for parallel processing on a multicore CPU and for writing more readable code. OS-level threads should be used mainly used for parallelism and better CPU utilization. If a programmer needs to write code which will be called asynchronously, then OS-level threads will be rather expensive in perspective of system resources.

CPU context switching methods

  • Cooperative (use of explicit Yield call)
  • Preemptive (periodic timer based)
  • Combined

Objectives

  • Unlimited number of instances (limited by available memory)
  • Fast switching, creation, destruction
  • Automatic thread pool management by physical CPU core count
  • Ability to run in main loop only (without TThread instances)
  • Provide own synchronization tools (Yield, Sleep, CriticalSection, Semaphore, Mutex, WaitForMultipleObjects, Queues, Synchronize, ...)
  • Priority control
  • Support for view list of all microthreads

Implementation

  • MicroThreading - Lazarus package, functional yet not finished, not multi-platform, need patching FPC

External links