|
Progress in network and storage technology enables data center services to provide high-concurrency and low-latency processing responses. The user-mode thread pool is one of the technologies that can be used to improve the overall concurrent responsiveness of the service system while ensuring its stability of the system. In the system of multicore processor architecture, the scheduling management of the user-mode thread pool is much more complicated than that of the thread pool. First, the existing method cannot fully utilize the advantage of parallel execution of multicore processor architecture, and it is difficult to implement unified resource provisioning management among multiple user-mode thread pools. Second, the existing user-mode thread scheduling method cannot distinguish the performance requirements of different data processing tasks but treat all user-mode thread tasks in the same way. Last, the existing user-mode thread scheduling method controls the execution of all kernel threads, which is difficult to isolate host threads and other kernel threads in the system. It fails to reflect the performance requirements of different user-mode thread tasks, or to provide finer-grained execution timing control of the user-mode thread. This paper provides a two-level scheduling management method for user-mode thread pool, which can effectively integrate OS scheduling mechanism and user-mode thread scheduling mechanism, with a view to distinguishing the needs of different data processing tasks, enabling more targeted and fine-grained scheduling and execution control of user-mode threads, achieving isolation of host threads from non-host threads and differentiated scheduling control without affecting the scheduling mechanism of the OS itself, and enabling the global unified resource provisioning management of multi-user thread pools for multicore processor architecture. Generally, the new method can greatly improve the efficiency and performance of user thread pools on multicore processor architecture, enable the overall performance of the system to scale freely according to data processing task requirements, and provide more efficient user-mode thread pool operation management for highly concurrent and low-latency service system.
|
