J. Aspnes and M. Herlihy, Wait-free data structures in the asynchronous pram model, in ACM Symposium on Parallel Algorithms and Architectures, 2000, pp. 340349.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....variables to be read atomically# [1, 3, 17] algorithms for maintaining timestamps [25, 28]# and mechanisms for implementing any object whose oper The MWSC primitivemust be implemented in software because it is not supported byany real machine. 50 ations satisfy certain algebraic requirements [5, 16]. For example, a construction is given in [5] that implements any object suchthat,for each pair of operations on the object, either the two operations commute with each other, or one overwrites the other (i.e. the effects of executing both operations is the same as executing just one of them) ....

J. Aspnes and M. Herlihy. Wait-free data structures in the asynchronous pram model. In Proceedings of the Second Annual ACM Symposium on Parallel Architectures and Algorithms, pages 340--349, June 1990.

....can think of the shared memory as a snapshot object, rather than as a collection of individual registers. For example, snapshot objects have been used to solve randomized consensus [7, 8] and approximate agreement [11] and to implement bounded concurrent timestamps [18] and other types of objects [9, 19]. Because snapshot objects are much easier for programmers to use, researchers have spent a great deal of e#ort on finding e#cient implementations of snapshots from registers, which, unlike snapshot objects, are provided in real systems. We consider the problem of wait free implementation of a ....

....these results to construct an execution in which the troublesome SCAN takes many steps. We conclude with a brief discussion in Section 6. 2. RELATED WORK Snapshot objects were introduced independently by Afek, Attiya, Dolev, Gafni, Merritt, and Shavit [1] Anderson [3] and Aspnes and Herlihy [9]. Two types of snapshot objects have commonly been considered. In a multi writer snapshot object, any process can UPDATE any component, while in a single writer snapshot, each process owns one component of the snapshot object, and only the owner of a component can UPDATE it. Anderson [4] showed ....

James Aspnes and Maurice Herlihy. Wait-free data structures in the asynchronous PRAM model. In Proc. 2nd ACM Symposium on Parallel Algorithms and Architectures, pages 340--349, 1990.

....P after history H, P H Q, if hH P Qi = hH Qi Q (H P ) Q (H) P commutes with Q, P Q, if for all histories H, P H Q. Similarly, Q overwrites P , P Q, if for all histories H, P H Q. We take the following two de nitions from Anderson and Moir [AM93] and Aspnes and Herlihy [AH90] De nition 2 An object X is statically resilient i for any two operations P and Q in O at least one of the following hold: P Q, P Q or Q P . De nition 3 An object X is dynamically resilient i for every history H and any two operations P and Q in O at least one of the following hold: ....

....history H and any two operations P and Q in O at least one of the following hold: P H Q, P H Q or Q H P . The second de nition allows operations to have a di erent ordering depending on the history. Clearly, a statically resilient object is also dynamically resilient. Aspnes and Herlihy [AH90] showed that any dynamically resilient object has an (unbounded) wait free implementation in the asynchronous PRAM model. 3 Theorem 4 ( AH90] An object has a wait free implementation using only atomic read write registers if it is dynamically resilient. Matching this, Anderson and Moir [AM93] ....

[Article contains additional citation context not shown here]

Aspnes, J., and Herlihy, M. P. Wait-free data structures in the asynchronous PRAM model. In 2nd Ann. Symp. on Parallel Algorithms and Architectures (Crete, Greece, 1990), ACM Press, pp. 340-349.

....the processors is malfunctioning. When constructing shared memory objects like atomic registers, this issue is addressed by considering wait free constructions which guarantee that any operation executed by a single processor is able to complete even if all other processors crash in the meantime [AH90, Her91] In the second model a distributed system is required to overcome arbitrary changes to its state within a bounded amount of time. If the system is able to do so, it is called self stabilizing [Dij74, Sch93] To develop truly reliable systems both failure models must be considered ....

....nWnR atomic register using 0 stabilizing 1W1R regular registers. Fourth, the rami cations of so far having only 1 stabilizing atomic registers available for communication in self stabilizing protocols should be investigated further. Finally, following the work of Aspnes and Herlihy [AH90] it is an interesting venture to classify, based on their sequential speci cation, all k stabilizing shared memory objects that can be constructed from k stabilizing atomic registers, and to provide a general method to do so. In particular, we would be interested to know whether there are ....

Aspnes, J., and Herlihy, M. P. Wait-free data structures in the asynchronous PRAM model. In 2nd Ann. Symp. on Parallel Algorithms and Architectures (Crete, Greece, 1990), ACM Press, pp. 340-349.

....the processors is malfunctioning. When constructing shared memory objects like atomic registers, this issue is addressed by considering wait free constructions which guarantee that any operation executed by a single processor is able to complete even if all other processors crash in the meantime [5, 10]. In the second model a distributed system is required to overcome arbitrary changes to its state within a bounded amount of time. If the system is able to do so, it is called self stabilizing [7, 20] To develop truly reliable systems both failure models must be considered together. Research in ....

....nWnR atomic register using 0 stabilizing 1W2R regular registers. Fourth, the ramifications of so far having only 1 stabilizing atomic registers available for communication in self stabilizing protocols should be investigated further. Finally, following the work of Aspnes and Herlihy [5], it is an interesting venture to classify, based on their sequential specification, all k stabilizing shared memory objects that can be constructed from k # stabilizing atomic registers, and to provide a general method to do so. In particular, we would be interested to know whether there are ....

J. Aspnes and M. P. Herlihy, Wait-free data structures in the asynchronous PRAM model, in 2nd Ann. Symp. on Parallel Algorithms and Architectures, Crete, Greece, July 1990, ACM Press, pp. 340--349.

....level of subdivision of an input simplex may differ from one set of inputs to the next. Considering just the complexity of the worst case execution over all input sets would in many cases make a complexity theorem useless, since for example, for the approximate agreement problem Aspnes and Herlihy [3] have shown that for any k one can find a set of inputs that will require tinhe k in the worst case. The power of our theorem lies in its ability to allow one to reason about the complexity of problems in a purely topo logical setting. As we show, the subdivisions of a complex are a clean and ....

....will prove to be an invaluable tool for designing and analyzing concurrent algorithms. We provide two example applications of our theorem. In Section 6 we show tight upper and lower bounds on the tinhe to achieve N process approximate agreement. The best known results, due to Aspnes and Herlihy [3], imply an O(log 2 N) upper bound and an f2(log 3 N)lower bound computation, which will be specified in Section 2. We close this gap, proving matching upper and lower bounds of log d mput razg J where d = 3 for two processes and d = 2 for 3 or more. Our second result, in Section 7, is the ....

[Article contains additional citation context not shown here]

J. Aspnes, M.P. Herlihy, Wait-Free Data Structures in the Asynchronous PRAM Model. Proceedings of the 3rd Annual ACM Symposium an Principles af Distributed Computing, pages 377 408, July 1991. Also appeared as technicalreport.

....by a single processor is able to complete even if all other processors crash in the meantime. Originally, research in this area focussed on the construction of atomic registers from weaker (i.e. safe or regular) ones [VA86, Lam86, PB87, LTV89, IS92] Later attention shifted to stronger objects [AH90, Her91] See [KK89] for a brief survey. In the second model a distributed system is required to overcome arbitrary changes to its state within a bounded amount of time. If the system is able to do so, it is called selfstabilizing . Self stabilizing protocols have been extensively studied in the ....

....nWnR atomic register uses unbounded timestamps to invalidate old values. We would like to know whether this necessarily so, or if the space requirements of a k stabilizing atomic register can be bounded as in the non self stabilizing case [IS92] Second, following the work of Aspnes and Herlihy [AH90] it is an interesting venture to classify, based on their sequential specification, all k stabilizing shared memory objects that can be constructed from k 0 stabilizing atomic registers, and to provide a general method to do so. Acknowledgements It s a pleasure to thank Moti Yung; he ....

Aspnes, J., and Herlihy, M. P. Wait-free data structures in the asynchronous PRAM model. In 2nd SPAA (1990), pp. 340--349.

....the processors is malfunctioning. When constructing shared memory objects like atomic registers, this issue is addressed by considering wait free constructions which guarantee that any operation executed by a single processor is able to complete even if all other processors crash in the meantime [AH90, Her91] In the second model a distributed system is required to overcome arbitrary changes to its state within a bounded amount of time. If the system is able to do so, it is called self stabilizing [Dij74, Sch93] To develop truly reliable systems both failure models must be considered ....

....nWnR atomic register using 0 stabilizing 1W1R regular registers. Fourth, the rami cations of so far having only 1 stabilizing atomic registers available for communication in self stabilizing protocols should be investigated further. Finally, following the work of Aspnes and Herlihy [AH90] it is an interesting venture to classify, based on their sequential speci cation, all k stabilizing shared memory objects that can be constructed from k 0 stabilizing atomic registers, and to provide a general method to do so. In particular, we would be interested to know whether there are ....

Aspnes, J., and Herlihy, M. P. Wait-free data structures in the asynchronous PRAM model. In 2nd Ann. Symp. on Parallel Algorithms and Architectures (Crete, Greece, 1990), ACM Press, pp. 340-349. 23

.... of the register or, alternatively, an integer not exceeding a constant multiple of the number of writers per component) We believe that the tool of using uboundedly many memory locations is stronger than the method of unbounded time stamps (for constructions with unbounded time stamps see [5] and [8] Moreover, introducing randomness in the choice of the memory location recycled by a read, we obtain a conceptually very simple probabilistic protocol. If m is the number of writers per component, c is the number of components, and l and q are constants that can be chosen by the ....

.... the reader, as explained above, chooses the address ma[j] to be recycled, erases the value of the corresponding memory location and cyclically rotates the array ma[j] ma[dim] Then, it complements the value of its local variable f lag and updates the values of vb f lag ( fptr[f lag] ma[5]g) and ptr[f lag] ma[5] The alternation of the values of the boolean variables, and the consequent alternation between the two entries of PTR:ptrfield, where the writer gets the address it uses in order to write its component value, guarantees that the reader has the correct knowledge about ....

[Article contains additional citation context not shown here]

J. Aspnes and M. Herlihy, "Wait-free data structures in the asynchronous PRAM model," Proceedings of the 2nd Annual ACM Symposium on Parallel Architectures and Algorithms (ACM Press, New York), pp. 340--349, 1990.

....the processors is malfunctioning. When constructing shared memory objects like atomic registers, this issue is addressed by considering wait free constructions which guarantee that any operation executed by a single processor is able to complete even if all other processors crash in the meantime [AH90, Her91] In the second model a distributed system is required to overcome arbitrary changes to its state within a bounded amount of time. If the system is able to do so, it is called self stabilizing [Dij74, Sch93] To develop truly reliable systems both failure models must be considered ....

....nWnR atomic register using 0 stabilizing 1W1R regular registers. Third, the ramifications of so far having only 1 stabilizing atomic registers available for communication in self stabilizing protocols should be investigated further. Finally, following the work of Aspnes and Herlihy [AH90] it is an interesting venture to classify, based on their sequential specification, all k stabilizing shared memory objects that can be constructed from k 0 stabilizing atomic registers, and to provide a general method to do so. In particular, we would be interested to know whether there are ....

Aspnes, J., and Herlihy, M. P. Wait-free data structures in the asynchronous PRAM model. In 2nd Ann. Symp. on Parallel Algorithms and Architectures (Crete, Greece, 1990), ACM Press, pp. 340--349.

....of shared register accesses per operation as a function of the number of processes. 2. The Model 2 The atomic snapshot memory object is a powerful tool to construct other atomic wait free objects, for instance counters, logical clocks, or bounded concurrent time stamp schemes. Aspnes and Herlihy [AH90] give a general method to convert a sequential specification of a shared memory object that satisfies certain constraints to a wait free implementation of that object using an atomic snapshot memory object as a primitive. They also give a a polynomial time implementation of a wait free atomic ....

J. Aspnes and M. P. Herlihy. Wait-free data structures in the asynchronous pram model. In 2nd PAAA, pages 340--349, July 1990.

....may have parameters; here it is assumed that for each operation and each of its possible parameters we have a separate entry in O. Can an Operation Both Update the State and Return a Meaningful Value 3 We take the following two definitions from Anderson et al. AM93] and Aspnes et al. [AH90]. Definition 2.2. An object X is statically resilient i# for any two operations P and Q in O at least one of the following hold: P # Q, P # Q or Q # P . Definition 2.3. An object X is dynamically resilient i# for every history H and any two operations P and Q in O at least one of ....

....history H and any two operations P and Q in O at least one of the following hold: P #H Q, P #H Q or Q #H P . The second definition allows operations to have a di#erent ordering depending on the history. Clearly, a statically resilient object is also dynamically resilient. Aspnes et al. [AH90] showed that any dynamically resilient object has an (unbounded) wait free implementation in the asynchronous PRAM model. Matching this, Anderson et al. AM93] showed that this condition is necessary: for an object to have such a wait free implementation, it must satisfy Definition 2.3. This is ....

[Article contains additional citation context not shown here]

ASPNES, J., AND HERLIHY, M. P. Wait-free data structures in the asynchronous PRAM model. In 2nd SPAA (Crete, Greece, 1990), ACM Press, pp. 340--349.

No context found.

J. Aspnes and M. Herlihy,"Wait-Free Data Structures in the Asynchronous PRAM Model", Proceedings of the Second Annual ACM Symposium on Parallel Architectures and Algorithms , 1990, pp. 340-349.

....bus congestion and contention. Interestingly, this is the (trivial) solution that is used in current literature on wait free shared memory applications, including nearly all algorithms known to us for consensus, snapshots, coin ipping, bounded round numbers, timestamps, and multi writer registers [1, 2, 5, 7, 8, 9, 10, 11, 13, 20, 21, 24, 25, 26, 28, 30, 32, 38, 34, 35, 39, 40, 46, 59] . Indeed, the cost of this na ve implementation is easily shown to be a lower bound on the worst case cost of any implementation. Here, the worst case is taken over the set of adversarially chosen schedules of events (we give more details below) In this paper we show that in the interesting ....

J. Aspnes and M. P. Herlihy. Wait-Free Data Structures in the Asynchronous PRAM Model. In Proceedings of the 2nd Annual Symposium on Parallel Algorithms and Architectures, July 1990, pp. 340-349, Crete, Greece.

....an operation after taking a nite number of steps, despite failures of other processes. It is k bounded wait free, for some xed k 0, if every process completes an operation after taking k steps. The wait free ACM Symposium on Parallel Architectures and Algorithms, Crete, Greece, July 1990 [7], and in the Proceedings of the Third Annual ACM Symposium on Parallel Architectures and Algorithms, Hilton Head, North Carolina, July 1991 [25] 2 Some of these models also include primitives for barrier synchronization. 2 property excludes starvation: any process that continues to take steps ....

....asymptotic results to the independent work of Attiya, Lynch, and Shavit [9] Hoest and Shavit [28] have recently shown that, when translated to an iterated snapshot model, the constant factors in our results are the best possible. Since the rst appearance of the preliminary versions of this paper [7, 25], there have been many advances in the study of wait free objects built from atomic registers. In particular, there has been considerable improvement in algorithms for atomic snapshots. The lattice agreement technique [8] where processes agree on a chain in a lattice, is closely related to the ....

J. Aspnes and M.P. Herlihy. Wait-free data structures in the asynchronous PRAM model. In Proceedings of the 2nd Annual Symposium on Parallel Algorithms and Architectures, pages 340-349, July 1990.

....required to be linearizable [24] i.e. each operation appears to take effect instantaneously at some point between the operation s invocation and response. Atomic snapshot scan algorithms have been constructed by Anderson [4] bounded registers and exponential running time) Aspnes and Herlihy [7] (unbounded registers and O(n 2 ) running time) and by Afek, Attiya, Dolev, Gafni, Merritt, and Shavit [3] bounded registers and O(n 2 ) running time) Here running time is measured by the number of accesses to shared memory. Chandy and Lamport [15] considered a closely related problem in ....

.... precedence order on operations can be extended to a total order = such that the value returned by each observer is the result of applying all the mutator operations ordered before it by = This kind of object has a straightforward wait free linearizable implementation using atomic snapshot scan ([7]) A weakly linearizable implementation is one that permits = to be a partial order instead of a total order. This paper s contribution is to observe that (1) weakly linearizable objects can be implemented more efficiently than any algorithm known for their fully linearizable counterparts, and ....

J. Aspnes and M. P. Herlihy. Wait-Free Data Structures in the Asynchronous PRAM Model. In Proceedings of the 2nd Annual Symposium on Parallel Algorithms and Architectures, July 1990, pages 340--349, Crete, Greece.

....started and the collect nished. Curiously, the trivial implementation is the one used in almost all of the many asynchronous shared memory algorithms based on collects, including algorithms for consensus, snapshots, coin ipping, bounded round numbers, timestamps, and multi writer registers [1, 2, 5, 6, 8, 9, 12, 13, 15, 24 26, 29 31, 33, 35, 36, 38, 40 42, 52]. Noteworthy exceptions are [49, 50] which present interesting collect algorithms that do not follow the pattern of the trivial algorithm, but which depend on making strong assumptions about the schedule. Part of the reason for the popularity of this approach may be that the trivial algorithm ....

.... [11] to prove low competitive throughput for an algorithm that improves on the algorithm of [3] We show in Section 6 that relative competitiveness, combined with a throughput competitive collect algorithm, does in fact give throughput competitive solutions to problems such as atomic snapshot [2, 5, 9, 13, 15]and bounded round numbers [31] and argue that most algorithms that use collects can be shown to be throughputcompetitive using similar techniques. Finally, in Section 7 we discuss some related approaches to analyzing distributed algorithms and consider what questions remain open. 8 2 Model We ....

[Article contains additional citation context not shown here]

James Aspnes and Maurice P. Herlihy. Wait-free data structures in the asynchronous PRAM model. In Proceedings of the 2nd Annual Symposium on Parallel Algorithms and Architectures, pages 340-349, July 1990.

....be read atomically. Each process executes a loop. During the ith iteration, it writes its preferred value to a fresh area of shared memory and then takes an instantaneous snapshot of the values written by all of the processes. Such a snapshot operation can be implemented from ordinary registers [1, 2, 3]. Processes that see fewer than b values written submit their own preferred values to a (b 1) consensus object, write the response r to shared memory and take a snapshot again. If, during the second snapshot, a process sees that the number of processes that have started iteration i is still less ....

J. Aspnes and M. Herlihy. Wait-free data structures in the asynchronous PRAM model. In Proceedings of the 2nd ACM Symposium on Parallel Algorithms and Architectures, pages 340--349, 1990.

No context found.

J. Aspnes and M. Herlihy, Wait-free data structures in the asynchronous pram model, in ACM Symposium on Parallel Algorithms and Architectures, 2000, pp. 340349.

No context found.

J. Aspnes, and M. Herlihy, Wait-free data structures in the asynchronous PRAM model, in Proc., 2nd ACM Symp. on Parallel Alg. and Arch., 1990, pp. 340--349.

No context found.

James Aspnes and Maurice Herlihy. Wait-free data structures in the asynchronous PRAM model. In Proc. 2nd ACM Symposium on Parallel Algorithms and Architectures, pages 340--349, 1990.

No context found.

J. Aspnes and M.P. Herlihy. Wait-free data structures in the asynchronous pram model. In preparation, 1996.

No context found.

J. Aspnes and M. Herlihy. Wait-free data structures in the asynchronous PRAM model. In Proc. 2nd ACM Symposium on Parallel Algorithms and Architectures, pages 340--349, 1990.

No context found.

James Aspnes and Maurice Herlihy. Wait-free data structures in the asynchronous PRAM model. In Proc. 2nd ACM Symposium on Parallel Algorithms and Architectures, pages 340--349, 1990.

No context found.

J. Aspnes and M. Herlihy, "Wait-Free Data Structures in the Asynchronous PRAM Model", Proc. of the 2nd ACM Symposium on Parallel Algorithms and Architectures, pp. 340-349, 1990.

First 50 documents

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