J. H. Anderson, S. Ramamurthy, and R. Jain. Implementing wait-free objects on priority-based systems. Proceedings of the Sixteenth Annual ACM Symposium on Principles of Distributed Computing, pp. 229--238, August 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... queues to enable communication between the regular Java threads and the real time NHRT follows research in recent years, in which several researchers have investigated the use of wait free shared object algorithms as an alternative to lock based mechanisms in object based real time systems [1, 2, 3, 7, 15, 17]. Moreover, research in real time operating systems [6, 11, 9, 19] has also shown how to incorporate wait free techniques in real time kernels. Wait free implementation of shared data objects is an alternative approach for the problem of inter task communication and synchronization. Wait free ....

....then object calls by high priority processes automatically appear to be atomic to lower priority processes executing on the same processor. Consequently they show an implementation of the Compare and Swap from reads and writes in a priority based uniprocessor system [15] In a consequent paper [2], a wait free implementation of a linked list from compare and swap for priority based systems is presented. These results combined can o er an ecient implementation, with respect to time complexity, that satis es the speci cations of the wait free queue classes in RTSJ. The space complexity of ....

[Article contains additional citation context not shown here]

J. Anderson, S. Ramamurthy, and R. Jain. Implementing wait-free objects on priority-based systems. In Proceedings of the 16th Annual ACM Symposium on Principles of Distributed Computing (PODC '97), pages 229-238. ACM, Aug. 1997.

....ed values and, if they all match, updates the locations with a further set of values. All this is performed atomically with respect to other CASN operations and specialized reads. The implementation of a non blocking multi word compare and swap operation has been the focus of many research papers [7, 10, 2, 3, 15, 5]. As we will show none of these provides a solution that is practicable in terms of the operations it requires from the processor, its storage costs and the features it supplies. This paper presents a new design that solves these problems. The solution is valuable because it nally allows many ....

.... p processes, a machine word size of w bits, a maximum CASN width of n locations from a addresses, showing which algorithms are disjoint access parallel (D A parallel) and which require support from the operating system kernel (OS) D A parallel Requires Bits per word [7] No CAS1 0 [5] No CAS2 0 [3] No CAS1 OS 0 [2] Yes Strong LL SC p(w l) l where l = lg 2 p lg 2 a [15] Yes Strong LL SC lg 2 p lg 2 n [10] Yes Strong LL SC lg 2 p [3] Yes CAS1 OS 1 lg 2 n lg 2 p [10] Yes CAS1 p lg 2 p New Yes CAS1 0 or 2 that is not intrinsically wait free but which, at key points ....

[Article contains additional citation context not shown here]

J. H. Anderson, S. Ramamurthy, and R. Jain. Implementing wait-free objects on priority-based systems. In Proc. 16th PODC, pp 229-238, Aug. 1997.

....primitive like CAS2. These data structures include general singly linked and doubly linked lists [Gre99] A number of techniques exist for implementing lock free and wait free MWCAS on top of CAS and CAS2, enabling nonblocking synchronization for arbitrarily complex data structures [Her93, Moi97, ARJ97a, Gre99] These techniques have considerable overhead in both space and runtime complexity, especially when compared to common lock based operations, making them less interesting for kernel design. MORE COMPLEX OBJECTS. For data structures more complex than those mentioned in preceding ....

....In a kernel, the simplest way to prevent descheduling is to additionally disable interrupts in the local CPU during critical sections. 2. Use Greenwald s and Cheriton s globally locked software CAS2 or MWCAS with roll back [GC96] 3. Use Anderson s and colleagues wait free MWCAS implementation [ARJ97a] The first two methods use a lock to protect critical regions. In both cases, read operations as well as write operations have to be protected using a lock to prevent reads from seeing intermediate states. For both methods it is possible to reduce the number of data objects per lock, thereby ....

James H. Anderson, Srikanth Ramamurthy, and Rohit Jain. Implementing wait-free objects on priority-based systems. In Proceedings of the Sixteenth Annual ACM Symposium on Principles of Distributed Computing, pages 229--238, Santa Barbara, California, 21--24 August 1997.

....a straightforward mechanism for deploying many of the interesting non blocking data structures presented in the literature that have previously required CAS2. 1 Introduction The implementation of a non blocking multi word compare and swap operation has been the focus of many research papers [8, 10, 2, 3, 15, 6]. As we will show in Section 2, none of these has yet proposed a solution that is practicable in terms of the operations it requires from the processor, its storage costs and the features it provides. This paper presents a new CASN design that solves these problems, a proof of its correctness and ....

....conditionally wait free design for CASN [15] meaning one that is not intrinsically wait free but which, at key points after contention is detected, evaluates a user supplied function to determine whether to retry. Anderson et al. provide two further algorithms for priority based schedulers [3]. One is only suitable for uniprocessors. The other is not disjoint accessparallel. In their designs Anderson et al. use a restricted form of CAS2 which is much the same as the RDCSS operation we define in Section 3. However, it requires priority based scheduling and nonpreemption guarantees for ....

[Article contains additional citation context not shown here]

James H. Anderson, Srikanth Ramamurthy, and Rohit Jain. Implementing wait-free objects on priority-based systems. In Proceedings of the 16th Annual ACM Symposium on Principles of Distributed Computing, pages 229--238, August 1997.

....These data structures include general single linked and double linked lists. 6] A number of techniques exist for implementing lock free and wait free general multi word compare and swap (MWCAS) on top of CAS and CAS2, enabling nonblocking synchronization for arbitrarily complex data structures [11, 19, 2, 6]. These techniques have considerable overhead in both space and runtime complexity, especially when compared to common lock based operations, making them less interesting for kernel design. The most common technique to implement atomic multi word updates on uniprocessors is to prevent preemption ....

James H. Anderson, Srikanth Ramamurthy, and Rohit Jain. Implementing wait-free objects on priority-based systems. In Proceedings of the Sixteenth Annual ACM Symposium on Principles of Distributed Computing, pages 229--238, Santa Barbara, California, 21--24 August 1997.

....algorithm [15] and implementation [14] is fairly recent, however. The idea of non blocking and wait free synchronization was developed by Herlihy [29] There has been a long line of work attempting to make the idea more practical via universal constructions [11, 28] useful primitives [2, 3, 39], and specific data objects [3, 36, 45] In fact, our non blocking implementation of work stealing uses the bounded tags technique of [39] Nevertheless, to this day, few applications or systems have been built with non blocking synchronization. Of notable exception is a study of non blocking ....

....[14] is fairly recent, however. The idea of non blocking and wait free synchronization was developed by Herlihy [29] There has been a long line of work attempting to make the idea more practical via universal constructions [11, 28] useful primitives [2, 3, 39] and specific data objects [3, 36, 45]. In fact, our non blocking implementation of work stealing uses the bounded tags technique of [39] Nevertheless, to this day, few applications or systems have been built with non blocking synchronization. Of notable exception is a study of non blocking applications [37] and two non blocking ....

James H. Anderson, Srikanth Ramamurthy, and Rohit Jain. Implementing wait-free objects on priority-based systems. In Proceedings of the Sixteenth ACM Symposium on Principles of Distributed Computing (PODC), Santa Barbara, California, August 1997.

....algorithm [15] and implementation [14] is fairly recent, however. The idea of non blocking and wait free synchronization was developed by Herlihy [30] There has been a long line of work attempting to make the idea more practical via universal constructions [11, 29] useful primitives [2, 3, 41], and specific data objects [3, 38, 49] In fact, our non blocking implementation of work stealing uses the bounded tags technique of [41] Nevertheless, to this day, few applications or systems have been built with non blocking synchronization. Of notable exception is a study of non blocking ....

....[14] is fairly recent, however. The idea of non blocking and wait free synchronization was developed by Herlihy [30] There has been a long line of work attempting to make the idea more practical via universal constructions [11, 29] useful primitives [2, 3, 41] and specific data objects [3, 38, 49]. In fact, our non blocking implementation of work stealing uses the bounded tags technique of [41] Nevertheless, to this day, few applications or systems have been built with non blocking synchronization. Of notable exception is a study of non blocking applications [39] and two non blocking ....

James H. Anderson, Srikanth Ramamurthy, and Rohit Jain. Implementing waitfree objects on priority-based systems. In Proceedings of the Sixteenth ACM Symposium on Principles of Distributed Computing (PODC), Santa Barbara, California, August 1997.

....hidden from the programmer by means of the Tr Read and Tr Write routines, which perform all necessary address translation and bookkeeping. These routines are called within the programmer s transaction code in order to read or write a word of the MEM array. Thus, instead of writing MEM [1] MEM [10] , the programmer would write Tr Write(1# Tr Read(10) Figure 5.1 shows a simple example transaction, which enqueues an item onto a shared queue. This transaction is executed by a task by calling Tr Exec(enqueue) 164 Block 1 Current Blocks BANK of Pointers Replacement Blocks Writes block ....

....may not apply if wait free objects are implemented using techniques other than a Herlihy like helping scheme. In fact, Anderson, Ramamurthy, and Jain have shown recently that it is possible to dramatically reduce helping overhead in wait free implementations for priority based real time systems [10]. For such implementations, it may indeed be the case that wait free implementations of some objects are superior to their lock free counterparts, although the extent to which this is true is currently unknown. 6.2 Simulation Results In this section, we present results from simulation ....

[Article contains additional citation context not shown here]

J. H. Anderson, S. Ramamurthy, and R. Jain. Implementing wait-free objects in priority-based systems. In Proceedings of the 16th Annual ACM Symposium on Principles of Distributed Computing, pages 229--238. ACM, August 1997. 223

....can be prohibitively large. In recent years, several researchers have investigated the use of wait free shared object algorithms as an alternative to lock based mechanisms in object 1 Work supported by NSF grants CCR 9732916, CCR 9972211, CCR 9988327, and ITR 0082866. based real time systems [3, 4, 5, 6, 9, 10, 15]. In a waitfree object implementation, operations must be implemented using bounded, sequential code fragments, with no blocking synchronization constructs. Thus, a process never blocks while accessing a wait free object, and hence priority inversions cannot arise due to object accesses. In this ....

J. Anderson, S. Ramamurthy, and R. Jain. Implementing wait-free objects in priority-based systems. Proceedings of the 16th Annual ACM Symposium on Principles of Distributed Computing, pp. 229-238. Aug. 1997.

No context found.

J. Anderson, S. Ramamurthy, and R. Jain. Implementing wait-free objects in priority-based systems. In Proceedings of the 16th Annual ACM Symposium on Principles of Distributed Computing, pages 229-238. August 1997.

....loops, such as that depicted in Figure 1. Finally, recentwork by the #rst two authors and colleagues has shown that, in real time systems, the priority structure that exists often can be exploited to simplify object implementations, eliminating copying overhead entirely #Ramamurthy et al. 1996; Anderson et al. 1997#. The lock free approach to real time object sharing that we espouse is actually rooted in work done by Sorenson and Hamacher in the real time systems community some twentyyears ago #Sorensen 1974; Sorensen and Hemachar 1975#. Sorenson and Hamacher s work involved a real time communication ....

....apply if wait free objects are implemented using techniques other than a Herlihy like helping scheme. In fact, Anderson, Ramamurthy, and Jain have shown in a recent paper that it is possible to dramatically reduce helping overhead in wait free implementations for priority based real time systems #Anderson et al. 1997#. For such implementations, it may indeed be the case that wait free implementations are superior to lock free ones for certain objects, although the exact extent #if any# to which this is true is currently unknown. 6. EXPERIMENTAL COMPARISON In this section, we provide empirical evidence that ....

[Article contains additional citation context not shown here]

Anderson, J. H., Ramamurthy, S., and Jain, R. 1997. Implementing wait-free objects in priority-based systems. Manuscript.

....schemes considered in this paper are summarized in Figure 1. In total, six new object sharing schemes are proposed herein: two for uniprocessors and four for multiprocessors. The uniprocessor schemes listed in Figure 1 are based on a novel helping technique called incremental helping [5]. Two such schemes are presented, IHI (incremental helping with inheritance) and IHC (incre mental helping with ceilings) Incremental helping exploits the priority structure that exists on real time systems. The general idea of incremental helping, as applied within the IHI scheme, is ....

....cycle during any single wait free operation. CCAS is useful because the compare only value can be used to ensure that a late CCAS operation by a task that has been preempted and then resumed has no effect. Fortunately, CCAS is easy to implement even if CAS2 is not available. In a recent paper [5], we showed that CCAS can be implemented in only three high level language statements if CAS (a commonly available instruction) is available. 2.1. A Framework for Implementing Objects The general framework just mentioned is presented in this subsection. In presenting this framework, we focus on ....

[Article contains additional citation context not shown here]

J. Anderson, S. Ramamurthy, and R. Jain "Implementing Wait-Free Objects on Priority-Based Systems", Proc. of the 16th ACM Symp. on Principles of Distributed Computing, 1997, pp. 229-238.

....of short to medium duration allowed (the case we most expect and optimize for) but also calls that are quite long, approaching the length of an entire quantum. Our work builds upon recent research by us and others on using lock free and wait free shared object algorithms in real time systems [3, 4, 5, 6, 15, 19]. Operations on lockfree objects are optimistically performed using a user level retry loop. Such an operation is atomically validated and committed by invoking a synchronization primitive such as compare and swap (CAS) The retry loop is executed repeatedly until this validation step succeeds. ....

J. Anderson, S. Ramamurthy, and R. Jain. Implementing wait-free objects in priority-based systems. In Proceedings of the 16th ACM Symp. on Principles of Distributed Computing, pp. 229--238. 1997.

....other. In a sense, a helping based wait free scheme is a pessimistic notion of nonblocking user level synchronization that is similar to a lock based scheme, with helping taking the place of blocking. This is illustrated quite well by Figure 1, which is considered below. 2 In a recent paper [3], we showed that the cost of helping can be greatly reduced on priority based real time uniprocessor systems by using a technique called incremental helping . The general idea of incremental helping is illustrated in Figure 1. Before beginning a transaction, a task must first announce its ....

....variable. The fact that object access requirements do not have to be predeclared makes it easier to support mode changes and dynamically generated transactions. The notion of incremental helping can be extended for application on multiprocessors. We call the resulting scheme cyclic helping [1, 3]. With cyclic helping, the processors are thought of as if they were part of a logical ring. Tasks are helped through the use of a help counter , which cycles around the ring. To advance the help counter past a processor, a task must first help the currentlyannounced transaction on that ....

[Article contains additional citation context not shown here]

J. Anderson, S. Ramamurthy, R. Jain "Implementing Wait-Free Objects on Priority-Based Systems", Proc. 16th ACM Symp. on Prin. of Distr. Comp., to appear.

....must be taken to ensure that each operation is executed at most once, and that a helped task can retrieve its return values from memory. With most helping implementations that have been proposed [1, 2, 9] performance is at least linear in the number of tasks sharing an object. In a recent paper [4], we showed that the cost of helping can be greatly reduced on priority based realtime uniprocessor systems by using a technique called incremental helping . The general idea of incremental helping is illustrated in Figure 2. Before beginning a transaction, a task must first announce its ....

....an operational testbed. Despite our current limitations in conducting definitive performance studies, we have conducted a preliminary study to evaluate the effectiveness of cyclic helping. In this study, a wait free linked list implementation based on cyclic helping presented by us elsewhere [4] was compared with a lock free list implementation presented recently by Greenwald and Cheriton [8] These experiments were performed on a five processor SGI R10000 machine. The priority based preemption model was simulated at the user level. The wait free list implementation that 200 400 600 ....

J. Anderson, S. Ramamurthy, and R. Jain "Implementing Wait-Free Objects on Priority-Based Systems", manuscript, January 1997.

....duration allowed (the case we most expect and optimize for) but also calls that are quite long, approaching the length of an entire quantum. Our work has been heavily influenced by recent research by us and others on using lock free and wait free shared object algorithms in real time systems [2, 3, 4, 5, 6, 15, 17, 22]. Operations on lock free objects are optimistically performed using a user level retry loop. Such an operation is atomically validated and committed by invoking a synchronization primitive such as compare and swap (CAS) Figure 1 depicts a lock free enqueue operation that is implemented in this ....

J. Anderson, S. Ramamurthy, and R. Jain. Implementing wait-free objects in priority-based systems. In Proceedings of the 16th Annual ACM Symposium on Principles of Distributed Computing, pages 229--238. ACM, August 1997.

....loops, such as that depicted in Figure 1. Finally, recent work by the first two authors and colleagues has shown that, in real time systems, the priority structure that exists often can be exploited to simplify object implementations, eliminating copying overhead entirely [Ramamurthy et al. 1996; Anderson et al. 1997]. The lock free approach to real time object sharing that we espouse is actually rooted in work done by Sorenson and Hamacher in the real time systems community some twenty years ago [Sorensen 1974; Sorensen and Hemachar 1975] Sorenson and Hamacher s work involved a real time communication ....

....apply if wait free objects are implemented using techniques other than a Herlihy like helping scheme. In fact, Anderson, Ramamurthy, and Jain have shown in a recent paper that it is possible to dramatically reduce helping overhead in wait free implementations for priority based real time systems [Anderson et al. 1997]. For such implementations, it may indeed be the case that wait free implementations are superior to lock free ones for certain objects, although the exact extent (if any) to which this is true is currently unknown. 6. EXPERIMENTAL COMPARISON In this section, we provide empirical evidence that ....

[Article contains additional citation context not shown here]

Anderson, J. H., Ramamurthy, S., and Jain, R. 1997. Implementing wait-free objects in priority-based systems. Manuscript.

No context found.

J. H. Anderson, S. Ramamurthy, and R. Jain. Implementing wait-free objects on priority-based systems. Proceedings of the Sixteenth Annual ACM Symposium on Principles of Distributed Computing, pp. 229--238, August 1997.

No context found.

J. H. Anderson, S. Ramamurthy, and R. Jain, "Implementing wait-free objects on priority-based systems", Proceedings of the Sixteenth Annual ACM Symposium on Principles of Distributed Computing, pp. 229--238, August 1997. 23

No context found.

J. H. Anderson, S. Ramamurthy, and R. Jain. Implementing wait-free objects on priority-based systems. Proceedings of the Sixteenth Annual ACM Symposium on Principles of Distributed Computing, pp. 229238, August 1997.

No context found.

J. H. Anderson, S. Ramamurthy, and R. Jain. Implementing wait-free objects on priority-based systems. Department of Computer Science, University of North Carolina at Chapel Hill.

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