Multi-threading is how work performed by a computer can be divided into multiple concurrent streams of execution (generally referred to as threads).

Multi-threading is a common model to implement SM in multi-cored machines, where threads share memory through shared variables. In this model in parallel programming a processor can create multiple threads that will be executed in parallel. Each thread has its own stack, variables and ID, considered private state. For every thread there is a unique heap shared by all, considered shared memory.

In order to allow a volume of work to be most effectively and safely divided into multiple concurrent streams of execution, a number of critical areas need to be addressed.

  • Scheduling: ensure worker tasks are able to progress independently and effectively (e.g. deadlock/livelock avoidance).
  • Publication: ensure data altered in one thread is only visible to others as expected.
  • Synchronization: ensure critical regions are protected from multiple concurrent updates causing data loss/corruption.

The underlying tenet of concurrent processing is Amdahl's law (graph). This law governs the diminishing amount of throughput that can be achieved by greater numbers of concurrent processors/cores. Therefore the overall aim of multi-threading is to minimize the amount of serial execution (exclusive locking) within any concurrent system.

Books

More information:

history | excerpt history