Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In Proc. of the Fifth International Workshop on DistributedAlgorithms, pages 9--30, Oct. 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... a major thrust of Maya has been the examination of different protocols for shared memories, especially the implementations of memories that are not sequentially consistent [Lam79] These weaker memories have a lower latency and when used correctly, can enhance the performance of parallel programs [ABHN91, AH90, DSB86, GLL 90, LS88] Maya is capable of simulating the execution of a number of such memories including causal memory [ABHN91] and pipelined random access memory [LS88] Preliminary evaluation results for a number of user applications appear in [Maya93] This paper is organized as ....

.... are not sequentially consistent [Lam79] These weaker memories have a lower latency and when used correctly, can enhance the performance of parallel programs [ABHN91, AH90, DSB86, GLL 90, LS88] Maya is capable of simulating the execution of a number of such memories including causal memory [ABHN91] and pipelined random access memory [LS88] Preliminary evaluation results for a number of user applications appear in [Maya93] This paper is organized as follows. An overview of the design of Maya is presented in Section 2. In Section 3, the simulation environment and the modeling of parallel ....

[Article contains additional citation context not shown here]

Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In Proceedings of the 5th International Workshop on Distributed Algorithms, pages 9-- 30. LNCS, October 1991.

....section E. C. Weak consistency model As described above, for efficiency purpose, several copies of the same object can be distributed in the system. It is clear that these replicas have to be kept coherent. A lot of consistency models have been studied from a theoretical point of view[Lam79, LH89, ABWN92, RM93] but few of them have been implemented. Furthermore, most of actually implemented models are strong consistency oriented. For efficiency and scalability purpose, DOSMOS proposes a very weak protocol: the release consistency. This model [GLL 90] provides two synchronization operators: ....

Mustaque Ahamad, James E. Burns, Hutto Phillip W., and Gil Neiger. Causal memory. IEEE Computer, pages 9--29, 1992.

....the degree of coupling is closely related to the latency of memory access and to the cost of maintaining consistency. Most definitions of shared memories abandon atomicity of access, the usual correctness condition in sequential systems. Examples of such non atomic memories are causal memory [2] and pipelined random access memory [19] and memories based on the notions of release consistency [11] entry consistency [5] and hybrid consistency [4] In addition, some of these definitions are specific to a particular architecture. Though these non atomic memories improve the execution ....

....extension of C that enables processes to share variables and presents an abstraction of shared memory to the programmers. This paper describes the results of our experiments with distributed shared memory based on three different notions of consistency: sequential consistency [16] causal memory [2, 3], and pipelined random access memory (PRAM) 19] Since weak memories reduce the amount of implicit synchronization, explicit primitives are needed in order to synchronize user processes. Maya supports locks and barriers as explicit synchronization primitives. We consider two different user ....

[Article contains additional citation context not shown here]

Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In Proceedings of the 5th International Workshop on Distributed Algorithms, pages 9--30. LNCS, October 1991.

....among read and write events is derived from Lamport s definition [21] of the term for message passing systems. In [20] a write and read events are considered similar to message send and receive events, respectively A more precise definition of causal memory is presented by Ahamad et al. in [3]. To help define a causal computation and other interesting non coherent computations, we introduce two relations R w=w and R r=w over Delta. R w=w (figure 2) relates two distinct writes ff and fi to the same location if and only if ff follows fi in R but the value written by fi is read by a ....

Mustaque Ahamad, Phillip W. Hutto, James E. Burns, and Gil Neiger. Causal memory. In Proc. of the Fifth International Workshop on Distributed Algorithms, pages 9--30, Oct. 1991.

....the consistency guarantees provided by a memory system allows simpler and more efficient implementations of the memory system but on the other hand may also increase the complexity of programming applications. This thesis explores one such weakly consistent memory model called causal memory [5] and shows that it can be efficiently implemented and easily programmed. The goal of this thesis is to demonstrate that: ffl Causal memory can be efficiently implemented and is a viable architecture for programming distributed applications. ffl The execution of the applications on causal memory ....

....that the execution of memory operations of all processes is equivalent to some sequential execution of these operations as in a sequentially consistent system. Examples of weakly consistent memory systems include Pipelined RAM (PRAM) 41] processor consistency [4, 24, 26] and causal memory [5]. PRAM Lipton and Sandberg proposed a memory system, Pipelined RAM [41] which offered weaker consistency guarantees than sequential consistency. In a PRAM memory system, a read operation on a processor would just return the value in the local memory of the processor. A write operation on a ....

[Article contains additional citation context not shown here]

Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In Proceedings of the 5th International Workshop on Distributed Algorithms, October 1991.

....are read by all processes in the order they were written by the writer. Writes to different locations, even by the same process, can be interleaved in different orders at different processes. A design for a Causal memory system is described in [AHJ91] and a formal specification is given in [ABHN91] We do not know of an implementation of any of these memories. In Chapter 3 we argue for the combination of different behaviors in one system. MultiVersion memory [WW90] combines the behaviors of Weak memory and Dynamic Atomic memory. An implementation of it was used to speed up a parallel ....

Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In Proc. of the Fifth International Workshop on Distributed Algorithms, pages 9--30, Oct. 1991.

....imposes very strict requirements and inhibits many optimizations such as pipelined writes and out of order reads. Consequently, a number of weaker consistency conditions have been proposed recently. These include weak ordering [10] release consistency [14] hybrid consistency [6] causal memory [3], and pipelined random access memory (PRAM 1 ) 23] This paper proposes a new programming model called mixed consistency for DSM. The architecture that we target consists of a set of processors, each equipped with its own memory, interconnected by a message passing network. Our aim is to ....

....fax 1 (212) 689 0481, or permissions acm.org . tify a limited kind of synchronization by labeling memory operations. Software implementations of these consistency models replace labels by lock operations [7, 18] Some of the other memory models do not provide any synchronization operations [3, 6, 23]. The mixed consistency model combines two different abstractions of DSMs: PRAM and causal memory. PRAM admits efficient implementation as suggested by Lipton and Sandberg [23] Causal memory on the other hand expresses the causality constraints of a user program very naturally and may simplify ....

[Article contains additional citation context not shown here]

Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In Proceedings of the 5th International Workshop on Distributed Algorithms, pages 9--30. LNCS, October 1991.

....operations because it can allow both read and write operations to complete with cached copies of shared objects. Both reads and writes cannot be completed at a disconnected site when serializability has to be ensured. CC is closely related to causal memory which was developed by us earlier [1], but there are differences in the way the two are formulated. For example, causal memory defines a value for a shared location at acerrain point in a process s execution based on all preceding memory operations whereas CC defines mutual consistency requirements among the copies of a set of ....

....follows an update type scheme to propagate changes to servers. This is appropriate when servers maintain copies of all shared objects but its benefits are not clear when clients cache objects. Causality, from which CC is derived, has been investigated extensively. In our work on causal memory [1], we have developed implementations that are ap propriate when data is shared between parallel appli cations [7] These implementations either suffer from unnecessary invalidations or require that the size of the timestamp be linear in the number of shared objects which could lead to excessive ....

Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In Proceedings of the 5th International Workshop on Distributed Algorithms, October 1991.

....memories such as these in large systems; a simple argument [23] can be used to show that these memories cannot retain low access latency in large systems. This fact represents a significant efficiency problem in distributed shared memory systems. Recent research in high performance shared memories [2,3,17,23] has suggested the weakening of memory consistency to reduce the cost of maintaining the consistency of shared memory. Such memories scale well, but they sacrifice some of the consistency guarantees of sequential consistency. Other research [1,5,7,12] has proposed memories that are hybrids of ....

....here a non operational definition. As noted above, Goodman s definition of processor consistency was a combination of PRAM and coherence. The DASH definition (which we refer to as PCD) combines coherence with semi causality (this is similar to the causality relation used to define causal memory [2]) Since all PCD histories are coherent, there is a legal serialization S x of operations on each location x. The semi causality relation is defined as follows. First, a weaker notion of program order is used; this is because reads are allowed to bypass earlier writes by the same processor. We ....

[Article contains additional citation context not shown here]

Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In S. Toueg, P. G. Spirakis, and L. Kirousis, editors, Proceedings of the Fifth International Workshop on Distributed Algorithms, volume 579 of Lecture Notes on Computer Science, pages 9--30. Springer-Verlag, October 1991.

....the execution of memory operations of all processors is equivalent to some sequential execution of these operations, as in a sequentially consistent system. Examples of weakly consistent memory systems included pipelined RAM (PRAM) 34] processor consistency 1 [4, 21, 22] and causal memory [5]. To execute applications in such weakly consistent memory systems, either the applications must have data sharing patterns that are not effected by the weaker consistency (e.g. conservative programs for PRAM [34] or the program must explicitly deal with the lack of strong consistency. In this ....

....are properly synchronized (e.g. are data race free) Other memory models have been proposed that make the weaker consistency of a memory system visible to the programmers. Examples of these include pipelined RAM (PRAM) 34] processor consistency (PC) 22, 21] and causal and slow memories [5, 24]. Programmers must either show that programs are not affected by the weaker consistency (e.g. conservative programs with respect to PRAM [34] or must include code in their applications that deals with such weak consistency. It is argued that better performance can be achieved in these systems ....

Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In Proceedings of the 5th International Workshop on Distributed Algorithms, October 1991.

....memory systems. These problems have been recognized both in the study of multiprocessors, where the hardware provides support for shared memory, and in distributed systems, where the shared memory abstraction is implemented in software. As a result, a number of new memory models have been proposed [1,3,5,6,14,15,17] that seek to enhance performance and scalability by weakening the consistency guarantees provided by shared memory. In the first approach, taken in the systems described in [1,5,6] strong consistency is only provided for a subset of the operations (e.g. synchronization operations) and other ....

....due to their weakened consistency. In these systems, programs that meet certain requirements (properly labeled or data race free) do not need to be aware of the weak consistency and can be programmed as if the system provides strong consistency. The second approach is taken in distributed systems [3,15] where the application programmer must directly program with the weakly consistent memory. It is argued that, for many applications, programming is simpler than when message passing is used and the weaker memory consistency leads to improved performance. Many systems advocate high performance ....

[Article contains additional citation context not shown here]

Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In S. Toueg, P. G. Spirakis, and L. Kirousis, editors, Proceedings of the Fifth International Workshop on Distributed Algorithms, volume 579 of Lecture Notes on Computer Science, pages 9--30. Springer-Verlag, October 1991.

....operations because it can allow both read and write operations to complete with cached copies of shared objects. Both reads and writes cannot be completed at a disconnected site when serializability has to be ensured. CC is closely related to causal memory which was developed by us earlier [1], but there are differences in the way the two are formulated. For example, causal memory defines a value for a shared location at a certain point in a process s execution based on all preceding memory operations whereas CC defines mutual consistency requirements among the copies of a set of ....

....follows an update type scheme to propagate changes to servers. This is appropriate when servers maintain copies of all shared objects but its benefits are not clear when clients cache objects. Causality, from which CC is derived, has been investigated extensively. In our work on causal memory [1], we have developed implementations that are appropriate when data is shared between parallel applications [7] These implementations either suffer from unnecessary invalidations or require that the size of the timestamp be linear in the number of shared objects which could lead to excessive ....

Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In Proceedings of the 5th International Workshop on Distributed Algorithms, October 1991.

....described in Section 3. Section 4 discusses the programming support required to run applications efficiently on Causal memory. In Section 5 we describe how to program applications and give performance results. We talk about related work in Section 6 and conclude in Section 7. 2 Causal Memory In [4] we define correct execution on Causal memory by the possible set of values a read to a shared location can return. We use a more generalized framework here, since this can be used to classify a range of different memory models [21] 2.1 Formal Definition This section formally describes the ....

Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In Proceedings of the 5th International Workshop on Distributed Algorithms, October 1991.

No context found.

Mustaque Ahamad, James E. Burns, Phillip W. Hutto, and Gil Neiger. Causal memory. In Proc. of the Fifth International Workshop on DistributedAlgorithms, pages 9--30, Oct. 1991.

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