What is the difference between using a new thread and using a thread from the thread pool? What performance benefits are there and why should I consider using a thread from the pool rather than one I've explicitly created? I'm thinking specifically of .NET here, but general examples are fine.
Thread local storage is not a good idea with thread pools. It gives threads an "identity"; not all threads are equal anymore. Now thread pools are especially useful if you just need a bunch of identical threads, ready to do your work without creation overhead.
Thread pool will provide benefits for frequent and relatively short operations by
- Reusing threads that have already been created instead of creating new ones (an expensive process)
Throttling the rate of thread creation when there is a burst of requests for new work items (I believe this is only in .NET 3.5)
If you queue 100 thread pool tasks, it will only use as many threads as have already been created to service these requests (say 10 for example). The thread pool will make frequent checks (I believe every 500ms in 3.5 SP1) and if there are queued tasks, it will make one new thread. If your tasks are quick, then the number of new threads will be small and reusing the 10 or so threads for the short tasks will be faster than creating 100 threads up front.
If your workload consistently has large numbers of thread pool requests coming in, then the thread pool will tune itself to your workload by creating more threads in the pool by the above process so that there are a larger number of thread available to process requests
check Here for more in depth info on how the thread pool functions under the hood
Creating a new thread yourself would be more appropriate if the job were going to be relatively long running (probably around a second or two, but it depends on the specific situation)
@Krzysztof - Thread Pool threads are background threads that will stop when the main thread ends. Manually created threads are foreground by default (will keep running after the main thread has ended), but can be set to background before calling Start on them.
Check here for an earlier thread:
http://stackoverflow.com/questions/10274/when-should-i-not-use-the-threadpool-in-net
Summary is that Threadpool is good if you need to spawn many shortlived threads, whereas using Threads gives you a bit more control.
If you need a lot of threads, you probably want to use a ThreadPool. They re-use threads saving you the overhead of thread creation.
If you just need one thread to get something done, Thread is probably easiest.
also
new Thread().Start()
spawns Foreground thread that will not die if you close your program. ThreadPool threads are background threads that die when you close the app.
In general (I have never used .NET), a thread pool would be used for resource management purposes. It allows constraints to be configured into your software. It also may be done for performance reasons, as creation of new threads may be costly.
There may also be system specific reasons. In Java (again I don't know if this applies to .NET), the manager of the threads may apply thread specific variables as each thread is pulled from the pool, and unset them when they are returned (common way to pass something like an identity).
Example constraint: I only have 10 db connections, so I would only allow 10 worker threads for accessing the database.
This doesn't mean that you should not create your own threads, but there are conditions under which it makes sense to use a pool.
The .NET managed threadpool: -
- Sizes itself based on the current workload and available hardware
- Contains worker threads and completion port threads (which are specifically used to service IO)
- Is optimised for a large number of relatively short-lived operations
Other thread pool implementations exist that might be more appropriate for long-running operations.
Specifically, use a thread pool to prevent your app from creating too many threads. The most important feature of a threadpool is the work queue. That is, once your machine is sufficiently busy, the threadpool will queue up requests rather than immediately spawn more threads.
So, if you will create a small, bounded number of threads create them yourself. If you cannot determine up-front how many threads might be created (e.g. they're created in response to incoming IO), and their work will be short-lived, use the threadpool. If you don't know how many, but their work will be long-running, there's nothing in the platform to help you - but you might be able to find alternative threadpool implementations that fit.