13 February 2011

Software Design Puzzle #7 - Thread Pools & Tasks

We know what is thread. The thread is an independent execution path or entity in a process. When you want to exploit multi-core (increased processing) or do some blocking operations, the thread is the default destination (increased interactivity, heavy input/output operations). In any programming language that supports multi-threading, you run the code that can be run as independent execution path as thread with each having its own context. The infrastructure that is needed to run threads in parallel like stack, program counter, registers are tightly coupled with the code that runs. Due to this tight coupling, each time you need an independent path (let us call it as task), you create a thread (the infrastructure) which has its own overheads. Is it possible to change code of the thread?

Obviously the next step is to separate out the infrastructure and the task which is achieved using thread pools. Here is the next problem.

Write a simple thread pool in Java with following requirements

  1. Have a fixed but configurable number of threads in thread pool. At any point of time, you can run up to N threads.
  2. Have a task queue (or a suitable data structure). This can be much bigger than number of threads. The tasks wait in the queue for their turn.
  3. The thread should consume tasks from the task queue and execute it. It can be any task (but you have to ensure some fairness). The role of the thread is to run task without bothering too much on what it does and how it does.