Thread and Task Architecture

Complex applications may have many tasks that could be performed at the same time. Threads are an operating system feature that lets application logic be separated into several concurrent execution paths.

When an operating system executes an instance of an application, it creates a unit called a process to manage the instance. The process has a thread of execution, which is the series of programming instructions performed by the application code. In a simple application with a single set of instructions that can be performed serially, there is just one execution path, or thread, through the application. More complex applications may have several tasks that could be performed in tandem, not serially. The application could do this by starting separate processes for each task, but starting a process is a resource-intensive operation. Instead, an application can start separate threads, which are relatively resource-inexpensive. Each thread can be scheduled for execution independently from the other threads associated with a process. Each thread stores the data unique to it in an area of memory called a stack.

Threads allow complex applications to make more effective use of a CPU even on computers with a single CPU. With one CPU, only one thread can execute at a time. If one thread executes a long running operation that does not use the CPU, such as a disk read or write, then another one of the threads can execute until the first operation completes. Being able to execute threads while other threads are waiting for an operation to complete allows the application to maximize its use of the CPU. This is especially true for multiuser, disk I/O intensive applications such as a database server.

Computers with multiple CPUs can execute one thread per CPU at the same time. If a computer has eight CPUs, it can concurrently execute eight threads.

Windows NT Fibers

The Microsoft® Windows® operating system code that manages threads is in the kernel. Switching threads requires switches between the user mode of the application code and the kernel mode of the thread manager, which is a moderately expensive operation. Microsoft Windows NT® fibers are a subcomponent of threads that are managed by code running in user mode. Switching fibers does not require the user-mode to switch to the kernel-mode transition needed to switch threads. The scheduling of fibers is managed by the application, while the scheduling of threads is managed by Windows. Each thread can have multiple fibers.

  


(c) 1988-98 Microsoft Corporation. All Rights Reserved.