R. Rajkumar, "Real-time synchronization protocols for shared memory multiprocessors, " in Proceedings: The 10th International Conference on Distributed Computing Systems, May 1990, pp. 116--123.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....scheduling algorithms is that only independent tasks that do not synchronize or share resources have been considered. In contrast, tasks in real systems usually are not independent. Synchronization entails additional overhead, which must be taken into account when determining system feasibility [2, 6, 18, 19, 20, 22]. Unfortunately, prior work on real time synchronization has been directed at uniprocessor systems, or systems implemented using non fair scheduling algorithms (or both) and thus cannot be directly applied in fair scheduled multiprocessor systems. Indeed, synchronization issues in fair scheduled ....

R. Rajkumar. Real-time synchronization protocols for shared memory multiprocessors. In Proceedings of the International Conference on Distributed Computing Systems, pages 116--123, 1990.

.... cantly for the real time systems it leads to priority inversion in which a high priority task can be blocked for an unbounded time by a lower priority task [16] Several synchronization protocols have been introduced to solve the priority inversion problem for uniprocessor [16] and multiprocessor [14] systems. The solution presented in [16] solves the problem for the uniprocessor case with the cost of limiting the schedulability of task sets and also making the scheduling analysis of real time systems hard. The situation is even worse in a multiprocessor real time system, where a task may be ....

....for the uniprocessor case with the cost of limiting the schedulability of task sets and also making the scheduling analysis of real time systems hard. The situation is even worse in a multiprocessor real time system, where a task may be blocked by another task running on a di erent processor [14]. For the RTSJ, it was decided that the least intrusive speci cation for allowing real time safe synchronization is to require that implementations of the Java keyword synchronized includes one or more algorithms that prevent priority inversion among real time Java threads that share the ....

R. Rajkumar. Real-time synchronization protocols for shared memory multiprocessors. In 10th International Conference on Distributed Computing Systems, pages 116-123. IEEE, IEEE Computer Society Press, May{June 1990.

....any progress while the access to the shared resource is blocked by the lock. Using mutual exclusion can also cause deadlocks, priority inversion (which can be solved efficiently on uni processors [ 13] with the cost of more difficult analysis, although not as efficient on multiprocessor systems [12]) and even starvation. To address these problems, researchers have proposed non blocking algorithms for shared data objects. Nonblocking methods do not involve mutual exclusion, and therefore do not suffer from the problems that blocking can cause. Non blocking algorithms are either lock free or ....

R. RAJKUMAR. Real-Time Synchronization Protocols for Shared Memory Multiprocessors. loth International Conference on Distributed Computing Systems, pp. 116-123, 1990.

.... effect) KopR93] and iii) leads to priority inversion in which a high priority task can be blocked for an unbounded time by a lower priority task [ShaR90] Several synchronisation protocols have been introduced to solve the priority inversion problem for uniprocessor [ShaR90] and multiprocessor [Raj90] systems. The solution presented in [ShaR90] solves the problem for the uniprocessor case with the cost of limiting the schedulability of task sets and also making the scheduling analysis of real time systems hard. The situation is much worse in a multiprocessor real time system, where a task may ....

....for the uniprocessor case with the cost of limiting the schedulability of task sets and also making the scheduling analysis of real time systems hard. The situation is much worse in a multiprocessor real time system, where a task may be blocked by another task running on a different processor [Raj90] To address the problems that arise from blocking, researchers have proposed non blocking implementations of shared data structures. Two basic non blocking methods have been proposed in the literature, lockfree and wait free. Lock free implementations of shared data structures guarantee that at ....

R. Rajkumar. Real-Time Synchronization Protocols for Shared Mem- ory Multiprocessors. In loth International Conference on Distributed Computing Systems, pages 116-123, 1990.

.... effect) KopR93] and iii) leads to priority inversion in which a high priority task can be blocked for an unbounded time by a lower priority task [ShaR90] Several synchronisation protocols have been introduced to solve the priority inversion problem for uniprocessor [ShaR90] and multiprocessor [Raj90] systems. The solution presented in [ShaR90] solves the problem for the uniprocessor case with the cost of limiting the schedulability of task sets and also making the scheduling analysis of real time systems hard. The situation is much worse in a multiprocessor real time system, where a task may ....

....for the uniprocessor case with the cost of limiting the schedulability of task sets and also making the scheduling analysis of real time systems hard. The situation is much worse in a multiprocessor real time system, where a task may be blocked by another task running on a different processor [Raj90] To address the problems that arise from blocking, researchers have proposed non blocking implementations of shared data structures. Two basic non blocking methods have been proposed in the literature, lockfree and wait free. Lock free implementations of shared data structures guarantee that at ....

R. Rajkumar. Real-Time Synchronization Protocols for Shared Mem- ory Multiprocessors. In loth International Conference on Distributed Computing Systems, pages 116-123, 1990.

....scheduling algorithms is that only independent tasks that do not synchronize or share resources have been considered. In contrast, tasks in real systems usually are not independent. Synchronization entails additional overhead, which must be taken into account when determining system feasibility [2, 6, 18, 19, 20, 21]. Unfortunately, prior work on real time synchronization has been directed at uniprocessor systems, or systems implemented using non fair scheduling algorithms (or both) and thus cannot be directly applied in fair scheduled multiprocessor systems. Indeed, fair scheduled uniprocessor systems ....

....systems. We intend to examine these lock based techniques in more detail in future work. In real time systems in which locks are used, priority inversions must be dealt with. A priority inversion occurs when a task is blocked by a task of lower priority. Inheritance and ceiling schemes [6, 18, 19, 20, 21] limit the duration of priority inversions by temporarily boosting a lock holding task s priority when it blocks any higher priority task. In fair scheduled systems, such schemes cannot be applied directly because they disrupt allocation rates. As a result, blocking times in fair scheduled systems ....

R. Rajkumar. Real-time synchronization protocols for shared memory multiprocessors. In Proceedings of the International Conference on Distributed Computing Systems, pages 116--123, 1990.

....problem of priority inversion, and it also provides a bounded blocking time for a process. However, the implementation of the priority ceiling protocol is complicated, and adds an unacceptable overhead for small embedded systems. For multiprocessor applications, the shared memory protocol (SMP) [12] has been proposed. It is an extension of the single processor priority ceiling protocol. The protocol involves defining global semaphores for locking the global shared memory, and placing priority ceilings on accessing the semaphores to bound the waiting time of higher priority jobs. The overhead ....

R. Rajkumar, "Real-Time Synchronization Protocols for Shared Memory Multiprocessors," in Proc. of 10th International Conference on Distributed Computing Systems, Paris, France, pp. 116-123, June 1990.

....bu er operations are usually implemented using locks. When locks are used, kernel support is needed to limit the impact of priority inversions. A priority inversion occurs when a process is forced to block on a process of lower priority. Conventional mechanisms for bounding priority inversions [7, 14, 16] rely on the kernel to dynamically raise the priority of the lock holding process. This adds complexity to the kernel and complicates dynamic process creation and removal. In addition, the blocking time estimates used to account for priority inversions in the scheduling analysis of multiprocessor ....

R. Rajkumar. Real-time synchronization protocols for shared memory multiprocessors. Proceedings of the International Conference on Distributed Computing Systems, pp. 116-123, 1990.

....transition for a real time system, including the paused state, will then be defined. The dependency semaphores for multiprocessors will also be defined. Design and implementation details are considered. In Chapter 7, we will describe related work, called the ceiling protocol for multiprocessors [Rajkumar90] and discuss the comparison between it and our proposed protocols. We will give a simple conclusion, and discuss future work. In Appendix A, we will discuss the specification of the blocking time requirement for real time synchronisation protocols in this thesis. The method of specification is ....

....(4) If processor A s lock flag is set and processor B successfully stores into the locked range of physical addresses, then A s lock flag is cleared. Rajkumar suggested that a task should spin read its own cache entry [Dubois88] while it is enqueueing to avoid excessive traffic on the memory bus [Rajkumar90]. Spinlocks with dynamic backoff [Anderson90] are more useful. Guangxing Li and Jean Bacon [Li93] implemented the ceiling protocol in the WANDA Micro kernel on the Firefly shared memory multiprocessor system [Thacker87] Because a task must be non preemptive while it executes the primitives, they ....

[Article contains additional citation context not shown here]

Ragunathan Rajkumar. Real-Time Synchronization Protocols for Shared Memory Multiprocessors. In 10 th International Conference on Distributed Computing Systems, 1990. (pp 9, 114, 115)

....can be used as a hardware mutual exclusion algorithm [4] but they are not sufficient to provide real time guarantees and prevent priority inversion. A more elaborate scheme is the shared memory protocol (SMP) that has been proposed as an extension to the single processor priority ceiling protocol [5]. The overhead of implementing SMP is even greater than for the priority ceiling protocol, such that it has not been demonstrated effectively in a system with fast (over 100 Hz) processes. We previously proposed the State Variable (SVAR) mechanism as a more efficient means of providing the ....

R. Rajkumar, "Real-Time Synchronization Protocols for Shared Memory Multiprocessors," in Proc. of 10th International Conference on Distributed Computing Systems, Paris, France, pp. 116-123, June 1990.

....to satisfy application requirements, is thus necessary. In real time systems, using a lock based synchronization protocol is the traditional approach to implementing data objects shared among tasks. There have been many effective synchronization protocols developed for real time applications [6, 12, 32, 33]. One problem incurred when employing such a protocol is that of priority inversion in which a task has to wait for a lock on a shared data object to be released by another lower priority task. The waiting or blocking in addition causes problems in the scheduler as well as the schedulability of ....

Rajkumar, R. Real-Time Synchronization Protocols for Shared Memory Multiprocessors. In Proc. of the 10th International Conference on Distributed Computing Systems (1990), pp. 116--125.

....must control the access to the shared resources and bound the delay that each task can experience. In this thesis we propose an end to end scheduling based approach to the resource contention problem. Compared with the existing solution, known as the Multiprocessor Priority Ceiling Protocol (MPCP) [14], the end to end scheduling approach is more flexible and yields better performance in most cases. 1.3 Organization of the Thesis In Chapter 2, we provide an overview of the related work. The work presented in this thesis is closely related to the real time scheduling theory for uniprocessor ....

R. Rajkumar. Real-time synchronization protocols for shared memory multiprocessors. In Proceedings: The 10th International Conference on Distributed Computing Systems, pages 116--123, May 1990.

....to priority inversion is to abort lower priority transactions that conflict with higher priority ones. An alternative is to bound priority inversions using various priority inheritance protocols which temporarily boost the priority of tasks holding locks if they conflict with higher priority tasks [58, 60, 67, 76]. This temporary priority boost helps lower priority tasks complete transactions and release their locks. Since the best approach to real time concurrency control often depends upon the particular application, some researchers advocate hybrid protocols that borrow features of several earlier ....

....not been employed yet in our MDARTS prototype, but they reflect the approach we will take in the future for MDARTS transactions with semaphore queues. Prior work on real time multiprocessor synchronization minimizes the global blocking of high priority tasks at the expense of lower priority tasks [58, 60]. Global blocking in a multiprocessor system is blocking on semaphores that are shared across processor boundaries. In this chapter, we examine the relationship between global semaphore queue wait times and the schedulability of periodic tasks using rate monotonic scheduling on multiprocessors. We ....

[Article contains additional citation context not shown here]

R. Rajkumar, "Real-time synchronization protocols for shared memory multiprocessors, " in Proc. Int'l Conf. on Distributed Computing Systems, pp. 116--123, 1990.

....we believe that our uniprocessor schemes are important in their own right, our main interest in them derives from the fact that they are stepping stones towards our multiprocessor implementations. The differences between our multiprocessor schemes and previously proposed multiprocessor protocols [10, 11, 12, 14] are very striking. Previous multiprocessor schemes give rise to worst case blocking overheads that are so high as to often preclude practical implementations. This is because of difficulties associated with priority inversions involving tasks executing on different processors. In contrast, ....

....q i execute on processor q, each lower priority task can be helped at most once. This is why G q i is used in the third term. 4. Experimental Results In this section, we present results from experiments conducted to compare the performance of the multiprocessor priority ceiling protocol (MPCP) [10, 11, 12] with that of the four wait free multiprocessor schemes presented in this paper. The comparison is based on RM scheduling conditions given in this paper and in [10, 11, 12] We have also conducted similar experiments to compare our schemes with the distributed priority ceiling protocol (DPCP) 10, ....

[Article contains additional citation context not shown here]

R. Rajkumar, "Real-Time Synchronization Protocols for Shared Memory Multiprocessors", Proc. of the InternationalConference on DistributedComputingSystems, 1990, pp. 116-123.

....mechanisms are needed to bound the effects of priority inversion. Rajkumar et al. proposed two such mechanisms by extending the uniprocessor priority ceiling protocol (PCP) 8, 10] The resulting protocols are called the distributed PCP (DPCP) and the multiprocessor PCP (MPCP) respectively [7, 8, 9]. Both protocols assume a model in which a task can perform one or more transactions. In the DPCP, global resources are guarded by synchronization processors. A task accesses a global resource by making an RPC like call to a global critical section (GCS) server, which executes on its behalf on the ....

R. Rajkumar, "Real-Time Synchronization Protocols for Shared Memory Multiprocessors", Proc. Int'l Conf. on Distr. Comp. Sys., 1990, 116-123.

.... case, a pair of the native priority of a processor and the time stamp it obtains should be 2 Though induced from the same policy, the behavior of priority ceiling spin lock is quite different from those of the priority ceiling protocol [6] or its extension for shared memory multiprocessors [7]. This is because the processor which cannot acquire a lock is blocked with those protocols, while it spins with our situation. used as the priorities for inner locks. 2) Priority inheritance scheme is indispensable for nested priorityordered spin locks. Note that uncontrolled priority ....

R. Rajkumar, "Real-time synchronization protocols for shared memory multiprocessors," in Proc. Int'l Conf. Distributed Computing Systems, pp. 116--123, May 1990.

....to interact with a communication system. The global table can be accessed by processes executing on different CPUs, and therefore a multiprocessor solution for lock the table is required. The type of multiprocessor synchronization we require for the global SVAR table has been addressed in [23]. The shared memory protocol (SMP) is presented as an extension of the single processor priority ceiling protocol [29] The protocol involves defining global semaphores for locking the global shared memory, and placing priority ceilings on accessing the semaphores to bound the waiting time of ....

R. Rajkumar, "Real-Time Synchronization Protocols for Shared Memory Multiprocessors," in Proc. of 10th International Conference on Distributed Computing Systems, Paris, France, pp. 116-123, June 1990.

....priority inversion. The most well known solution to this problem in uniprocessors is the priority ceiling protocol (PCP) 15, 16] To ensure predictability in multiprocessor systems, Rajkumar et al. proposed the distributed PCP (DPCP) and the multiprocessor PCP (MPCP) both of which extend the PCP [12, 13, 14]. Both approaches assume a model in which a task can perform one or more transactions. In the DPCP, global resources are guarded by synchronization processors. A task accesses a global resource by making an RPC like call to a global critical section (GCS) server, which executes on its behalf on ....

R. Rajkumar, "Real-Time Synchronization Protocols for Shared Memory Multiprocessors", Proceedings of the International Conference on Distributed Computing Systems, May 1990, pp. 116-123.

....real time systems using shared memory multiprocessors, a predictable interprocessor synchronization mechanism is of primary importance. In addition to adopting a real time scheduling algorithm with resource constraints (e.g. the algorithm in [3] or a realtime synchronization protocol (e.g. [4]) the execution time of the underlying mutual exclusion 1 ITRON MP is a shared memory multiprocessor extension of ITRON, a real time kernel specification for embedded systems [1] mechanism must be bounded. In this paper, we focus on bounded spin lock algorithms, with which the time to ....

R. Rajkumar, "Real-time synchronization protocols for shared memory multiprocessors," in Proc. Int'l Conf. Distributed Computing Systems, pp. 116--123, May 1990.

....to priority inversion is to abort lower priority transactions that conflict with higher priority ones. An alternative is to bound priority inversions using various priority inheritance protocols which temporarily boost the priority of tasks holding locks if they conflict with higher priority tasks [34, 33, 38, 44]. This temporary priority boost helps lower priority tasks complete transactions and release their locks. Since the best approach to real time concurrency control often depends upon the particular application, some researchers advocate hybrid protocols that borrow features of several earlier ....

R. Rajkumar. Real-time synchronization protocols for shared memory multiprocessors. In Proc. Int'l Conf. on Distributed Computing Systems, pages 116--123, 1990.

....table. The key is then to ensure that the local and global tables are updated to always contain the most recent data, and that the local table is never updated while a task is using the table. Multiprocessor synchronization based on this type of shared memory architecture has been addressed in [30]. The shared memory protocol (SMP) is presented as an extension of the single processor priority ceiling protocol [35] The protocol involves defining global semaphores for locking the global shared memory, and placing priority ceilings on accessing the semaphores to bound the waiting time of ....

R. Rajkumar, "Real-time synchronization protocols for shared memory multiprocessors," in Proc. of 10th Int'l Conf. on Distributed Computing Systems, Paris, France, pp. 116-123, June 1990.

....promising approach to satisfying this requirement (Figure 1) In order to realize a real time system on a multiprocessor environment, predictable and scalable inter processor synchronization is among the significant issues. A real time synchronization protocol for multiprocessors such as those in [1, 2, 3] is necessary to be adopted in the application level. In addition, the underlying synchronization mechanism used in the run time system must also be predictable and scalable. In implementing the priority ceiling protocol for shared memory multiprocessors in [1] for example, a spin lock is used as ....

....such as those in [1, 2, 3] is necessary to be adopted in the application level. In addition, the underlying synchronization mechanism used in the run time system must also be predictable and scalable. In implementing the priority ceiling protocol for shared memory multiprocessors in [1], for example, a spin lock is used as an underlying synchronization primitive to guard the queue of tasks that are waiting to enter an application level critical section. Another queue (called a ready queue) of the tasks that are ready to execute on a processor should also be guarded with a spin ....

[Article contains additional citation context not shown here]

R. Rajkumar, "Real-time synchronization protocols for shared memory multiprocessors," in Proc. Int'l Conf. Distributed Computing Systems, pp. 116--123, May 1990.

No context found.

R. Rajkumar, "Real-time synchronization protocols for shared memory multiprocessors, " in Proceedings: The 10th International Conference on Distributed Computing Systems, May 1990, pp. 116--123.

No context found.

R. Rajkumar, Real-time synchronization protocols for shared memory multiprocessors, in Proceedings of the 10th International Conference on Distributed Computing Systems, 1990, pp. 116123.

No context found.

Ragunathan Rajkumar. Real-time synchronization protocols for shared memory multiprocessors. In Proc. 10th Conference on Distributed Computing, pages 116-125, 1990.

No context found.

Rajkumar, R. "Real-Time Synchronization Protocols for Shared Memory Multi-Processors". Proceedings of The 10th International Conference on Distributed Computing (1990).

No context found.

R. Rajkumar, "Real-time synchronization protocols for shared memory multiprocessors, " in Proceedings: The 10th International Conference on Distributed Computing Systems, May 1990, pp. 116--123.

No context found.

R. Rajkumar, "Real-time synchronization protocols for shared memory multiprocessors," Proc. 10th IEEE International Conference on Distributed Computing Systems, Paris, May 1990, pp. 116 -- 123.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC