Timing management for user-level real-time threads can be done with appropriate support of the kernel. When a specified time comes, the kernel makes a virtual processor to upcall a user-level scheduler for its timing management. Then, the timing management can suffer from the overhead of a user-level scheduler. The paper presents an efficient timing management mechanism for user-level real-time threads. By sharing user-level timers through the shared kernel/user structure and keeping the appropriate hints for them, redundant processing of them and events can be eliminated. The results of the performance evaluations show that the upcall performance of our user-level real-time threads is comparable to and more stable than that of kernel-provided real-time threads.
展开▼
