Lecture 9-1

Concurrency

Lecture 9-1 Bart Massey 2015-11-22

What Is Concurrency?

  • Doing two things "at the same time". Multiple flows of control

  • Often not actually at the same time:

    • Getting the flow of control for a limited time / callbacks
    • Changing flow of control via voluntary preemption / coroutining
    • Changing flow of control via involuntary preemption / timesharing
  • Parallelism: Actually doing two things at the same time

Why Concurrency?

  • Sometimes it can be useful to think about two things as happening independently in the same program

  • For example, "generate-and-test"

    • One piece of code generates random numbers
    • The other tests them for primality
  • Each piece of the program has its own pc, stack and local state

  • Of course, you may want to take advantage of multiple processors = true parallelism

    • Shared memory
    • Shared channel

Callbacks: Extremely "Poor Person's" Concurrency

  • Give the testing code a function pointer that it can call to generate the next random number

  • Give a state argument to the callback so that it can set up its internal state

Generators: Letting Your PL Manage State

  • Another way: Let the PRNG return a random number and then set it up to "keep going" when the tester is finished.

  • Each call to the generator will run the PRNG from "where it left off"

  • Do the same thing with the tester

Threads: Splitting Off Control Completely

  • Another way: Start the PRNG as a whole separate process, except with the same memory as the generator. You can then use the shared memory to communicate answers.

  • This is pretty common in a lot of languages / OS now, and can allow pre-emptive multitasking and/or parallelism

  • It's also extremely error-prone

    • Deadlock
    • Livelock
    • Race Conditions

Thread Synchronization Primitives

  • Mutex: Blocks other threads from entry when a thread is running in the mutex

  • Semaphore: Clever blocking counter that is hard to understand

  • Mailbox: Allows shared message passing

Processes: Splitting Off Everything

  • Don't have to share memory

  • Now all communications are through channels

  • Great isolation, but can be slow and heavyweight

  • Pipes

Conclusions

  • Concurrency can be scary

  • On the other hand, it can simplify a design

  • Use good judgement; get expert help

Last modified: Wednesday, 25 November 2015, 12:50 AM