Ramesh, S. (1987). A new and efficient implementation of multiprocess synchronization. In Proceedings Conference on PARLE, Lecture Notes in Computer Science 259, pages 387--401, Berlin.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of communication is usually asynchronous and bipartied. The multiparty interaction scheduling problem then is concerned with synchronizing asynchronous processes so that a ready process knows when and which interaction to execute. Numerous algorithms have been proposed for the problem, e.g. (Ramesh, 1987; Bagrodia, 1989; Park and Kim, 1990; Kumar, 1990; Tsay and Bagrodia, 1994; Joung and Smolka, 1994) Note that in general it is impossible to determine a priori when a process will become ready and what interactions it can potentially execute. 5 The scheduling of multiparty interaction is ....

Ramesh, S. (1987). A new and efficient implementation of multiprocess synchronization. In Proceedings Conference on PARLE, Lecture Notes in Computer Science 259, pages 387--401, Berlin.

....them will attend some meeting. The crux of this problem is that two or more committees might share a professor. When that professor becomes available, she can only choose one of the meetings, while the others continue to wait. Coordinating this is non trivial, and various solutions have appeared [9, 17, 22, 23]. Several proposals for concurrency primitives [5, 11, 19, 20, 24] further extend the CSP model in a direction briefly mentioned in Hoare s original paper. These introduce communication entities known as channels. A process no longer explicitly names other processes in its send and receive ....

S. Ramesh. A new and efficient implementation of multiprocess synchronization. In Proceedings of PARLE '87 (Parallel Architectures and Languages Europe) , number 259 in Lecture Notes in Computer Science, pages 387--401 (vol.2). Springer-Verlag, June 1987.

....fairness notions, while strong interaction fairness is much stronger. In general, stronger fairness notions induce more liveness properties, but are also more difficult to implement. Therefore, it is not surprising to see that only weak interaction fairness has been widely implemented (e.g. [16, 14, 4, 13, 11, 17, 9]) It is also not surprising to see that all of these algorithms are asymmetric and deterministic, as weak interaction fairness (and thus strong interaction fairness) has been proven impossible by any symmetric, deterministic, and distributed algorithm [7, 12] Given that a process decides ....

S. Ramesh. A new and efficient implementation of multiprocess synchronization. In Proc. Conf. on PARLE, Lecture Notes in Computer Science 259, pages 387--401, 1987.

....corresponds to a multiparty 4 interaction. The horizontal nondeterminism is exactly what is needed to avoid duplicate elements (see [EFK89] for further details) A number of distributed algorithms have been proposed for various instances of the multiparty interaction scheduling problem (e.g. Ram87, CM88, Bag89, GB89, PK90, Kum90, JS90a, JS90b, JS94] but, to our knowledge, we are the first to present a comprehensive analysis of the complexity of multiparty interaction. The algorithms of [Ram87, CM88, Bag89, GB89, PK90, Kum90] concern fixed interactions in the presence of disjunctive ....

....have been proposed for various instances of the multiparty interaction scheduling problem (e.g. Ram87, CM88, Bag89, GB89, PK90, Kum90, JS90a, JS90b, JS94] but, to our knowledge, we are the first to present a comprehensive analysis of the complexity of multiparty interaction. The algorithms of [Ram87, CM88, Bag89, GB89, PK90, Kum90] concern fixed interactions in the presence of disjunctive parallelism, such as those found in Action Systems, IP, LOTOS, PPSAs, and Raddle. The algorithm of [JS90a] is for fixed interactions in a multipartied extension of CSP [Hoa78] where conjunctive and ....

S. Ramesh. A new and efficient implementation of multiprocess synchronization. In Proceedings Conference on PARLE, Lecture Notes in Computer Science 259, pages 387-- 401, Berlin, 1987. Springer-Verlag.

....of semantic constraint weak interaction fairness (WIF) requiring that an enabled interaction be eventually disabled. To allow comparison with existing algorithms for scheduling interactions, we first discuss the message complexity of Ramesh s efficient algorithm for multiparty interactions [Ram87a]. Ramesh s algorithm also allows each process p i , when ready for an interaction x, to behave as a coordinator of x, which p i does by attempting to capture all of the participants of x. Unlike our algorithm, 37 however, p i must capture the participants in the strict ordering of the process ....

....that can be distributedly implemented. Although our distributed algorithm is general, it is still efficient. In average, a process generates at most 3m 1 messages in an attempt to establish an 38 interaction, which is nearly the same as Ramesh s efficient algorithm for multiparty interactions [Ram87a]. However, Ramesh s algorithm can only guarantee a very weak form of semantic constraint weak interaction fairness. To our knowledge, no semantic constraint stronger than weak interaction fairness has ever been claimed for multiparty interactions. Acknowledgments. The work of the paper is ....

S. Ramesh. A new and efficient implementation of multiprocess synchronization. In Proceedings Conference on PARLE, Lecture Notes in Computer Science 259, pages 387--401, Berlin, 1987. Springer-Verlag.

....same design method; the present paper is different in that the distributed protocol uses less communication overhead and in that the exposition is less technical. Many published verifications of multiway synchronization algorithms assert that an algorithm satisfies certain correctness properties [2, 4, 6, 8, 9, 18]; our verification methods yields a more general result: all properties of the ideal protocol, formulated in terms of communication events with the clients, are also properties of the distributed protocol. The following section explains in more detail what we mean by multiway synchronization. ....

....for multiway synchronization of processes. The algorithm is designed for large scale environments with many processes, where it is not suitable to implement process synchronization in a centralized manner. Many different algorithms have been proposed for implementing multiway synchronization [1, 2, 4, 6, 8, 13, 18]; the protocol presented here is an improved version, in terms of the number of messages needed to establish synchronization, of our previously published algorithm [16] We have outlined a way of demonstrating the correctness of the protocol, based on the csequivalence relation between transition ....

S. Ramesh. A new and efficient implementation of multiprocess synchronization. In Parallel Architectures and Languages Europe, volume 259 of Lecture Notes in Computer Science, pages 387--401. Springer-Verlag, 1987. 13

....committed themselves to the performance of a certain interaction, the body of the interaction has to be executed. This can be done serially by one of the processes, or in parallel by some or all of them. A methodology for choosing a single process to execute the common code was presented by Ramesh [32]. Using only one process might seem wasteful at first glance. After all, we know that all the participating processes are available when the interaction commences. However, some or even most of them may not be executing. It is certainly plausible that there would be more processes than processors ....

S. Ramesh, "A new and efficient implementation of multiprocess synchronization". In Parallel Arch. & Lang. Europe, vol. II, pp. 387--401, Springer-Verlag, 1987. Lecture Notes in Computer Science Vol. 259.

....synchronously, the underlying model of communication is usually asynchronous and bipartied. The multiparty interaction scheduling problem then is concerned with synchronizing asynchronous processes so that a ready process knows when and which interaction to execute (e.g. see algorithms of [Ram87, Bag89, PK90, Kum90, TB92, JS94] Note that in general it is impossible to determine a priori when a process will become ready and what interactions it can potentially execute. The scheduling of multiparty interaction is typically associated with some notion of fairness to prevent unfair ....

S. Ramesh. A new and efficient implementation of multiprocess synchronization. In Proc. Conf. on PARLE, LNCS 259, pages 387-- 401, 1987.

....design a synchronous language that uses SCCS like interaction, then we may allow conjunctive and disjunctive parallelism to be used in isolation. Related Work A number of distributed algorithms have been proposed for various instances of the multiparty interaction implementation problem (e.g. [26, 5, 2, 11, 24, 19, 15, 16, 17]) but, to our knowledge, we are the first to present a comprehensive analysis of the complexity of multiparty interaction. The algorithms of [26, 5, 2, 11, 24, 19] concern primitive interactions in the presence of disjunctive parallelism, such as those found in Action Systems, IP, LOTOS, PPSAs, ....

....distributed algorithms have been proposed for various instances of the multiparty interaction implementation problem (e.g. 26, 5, 2, 11, 24, 19, 15, 16, 17] but, to our knowledge, we are the first to present a comprehensive analysis of the complexity of multiparty interaction. The algorithms of [26, 5, 2, 11, 24, 19] concern primitive interactions in the presence of disjunctive parallelism, such as those found in Action Systems, IP, LOTOS, PPSAs, and Raddle. The algorithm of [15] concerns primitive interactions in Hybrid CSP where conjunctive and disjunctive parallelism are used in isolation. An improved ....

Ramesh, S. "A new and efficient implementation of multiprocess synchronization," in: Proceedings Conference on PARLE, Lecture Notes in Computer Science 259, Springer-Verlag, Berlin (1987), pp. 387-401.

.... Response time Message Complexity Buckley and Silberschatz [5] O(N) O(k) m = 2 Sistla [21] O(k 3 ( c ) O(k) Bagrodia and Tsay [4] O(k 2 ) O(k) Tsay and Bagrodia [23] O(k 2 ) O(k) Our algorithm O(k 2 ) O(k) Kumar [15] O(N) O(km) Arbitrary Park and Kim [18] O(N) O(km) m Ramesh [19] O(N) O(km) Our algorithm O(k 2 m 2 ) O(km) Figure 3: Comparison with existing algorithms to the zeroth order interaction problem. Here, m is the number of processes required to execute an interaction, k is the number of interactions in which a process may participate, N is the total number ....

S. Ramesh. A new and efficient implementation of multiprocess synchronization. In PARLE: Parallel Architectures and Languages Europe. Lecture Notes in Computer Sciences 259, pages 387--401, Springer-Verlag, Berlin, 1987.

....known fairness notions, while strong interaction fairness is much stronger. In general, stronger fairness notions induce more liveness properties, but are also more difficult to implement. Therefore, it is not surprising to see that only weak interaction fairness has been widely implemented (e.g. [18, 15, 4, 14, 12, 20, 10]) It is also not surprising to see that all of these algorithms are asymmetric and deterministic, as weak interaction fairness (and thus strong interaction fairness) has been proven impossible by any symmetric, deterministic, and distributed algorithm [8, 13] Given that a process decides ....

.... system (which, in turn, could be dependent on the total number of professors) in practice, it is generally known that both parameters must be kept small and independent of the total number of professors in the system [7] 4 In contrast, deterministic algorithms for Committee Coordination such as [15, 12, 10] have time complexity c 0 Delta C max Delta S max Delta N where c 0 is a constant and N is the total number of professors in the system. 5 The time complexity of these algorithms depends explicitly on N because they use priority to beak the symmetry among professors. As such, a lower ....

S. Ramesh. A new and efficient implementation of multiprocess synchronization. In Proceedings Conference on PARLE, Lecture Notes in Computer Science 259, pages 387--401, Berlin, 1987. Springer-Verlag.

....Saks 1990] and the committee coordination problem [Chandy and Misra 1988] These problems can be solved by using an underlying solution to the dining philosophers problem. We focus our attention on the committee coordination problem and develop solutions that outperform the existing algorithms [Ramesh 1987; Kumar 1990; Park and Kim 1990] under the measures of response time and failure locality. The rest of this article is organized as follows. In Section 2, the system model is discussed, and the dining philosophers problem is specified. In Section 3, we present the idea of doorways. In Section 4, ....

....and A. K. Singh Authors Response Message Failure Algorithms Time Complexity Locality Buckley and Silberschatz [1982] O(n) O(D) O(n) K = 2 Sistla [1984] O(D 3 ) O(D) O(D) Tsay and Bagrodia [1994] O(D 2 ) O(D) O(D) Kumar [1990] O(n) O(KD) O(n) Park and Kim [1990] O(n) O(KD) O(n) Arbitrary Ramesh [1987] O(n) O(KD) O(n) K Algorithm 1 O(K 2 D 2 ) O(KD) O(KD) Algorithm 2 O(KD KD 2 ) O(KD KD 3 ) 3 Algorithm 3 O(K 2 D 2 ) O(K 2 D 2 ) 4 Fig. 11: Comparison with existing algorithms to the committee coordination problem. Here, K is the number of processes required to execute ....

[Article contains additional citation context not shown here]

Ramesh, S. 1987. A new and efficient implementation of multiprocess synchronization. In Proceedings of the 1st Conference on Parallel Architectures and Languages Europe. Lecture Notes in Computer Science, vol. 259. Springer-Verlag, Berlin, 387--401.

....initially, and then the following scenario is repeated forever: ffl C 1 and M 1 establish access 1 ; ffl C 1 and M 1 exit access 1 and then respectively become ready again. WIF has been widely implemented in CSP like biparty interactions [8,31,29,5,33] as well as in the multiparty case [28,4,27,20,17]. Although WIF can ensure some form of liveness, it is sometimes too weak to be useful. For example, consider another execution of the replica program: All four processes are ready for interaction initially, and then the following scenario is repeated forever: ffl C 1 and M 1 establish access 1 ....

.... is dominated by 4c Delta m Delta k m In the above, since m messages are sent in parallel in each interval c, the expected number of messages needed to establish an interaction per process is no greater than 4m 2 Delta k m For comparison, the efficient deterministic algorithm by Ramesh [28] has a worst case time complexity in the order of 3c Delta n Delta k and a message complexity 3m Delta k. Note that, unlike TB (and other randomized algorithms [12,30,19] the time complexity of deterministic algorithms typically depends on n the total number of processes in the system. ....

S. Ramesh. A new and efficient implementation of multiprocess synchronization. In Proceedings Conference on PARLE, Lecture Notes in Computer Science 259, pages 387--401, Berlin, 1987. Springer-Verlag.

....of a first order interaction can be concurrently activated. As more than one process can enrole into the same role, the failure of any one of them should not preclude the others from filling the role. Thus, first order interactions are inherently fault tolerant. The distributed algorithms of [Ramesh 1987; Chandy and Misra 1988; Bagrodia 1989a; Kumar 1990; Park and Kim 1990] for multiparty interaction guard scheduling are zeroth order in nature. The algorithms of [Chandy and Misra 1988; Bagrodia 1989a; Park and Kim 1990] employ a fixed coordinator for each interaction and thus cannot realize ....

....zeroth order in nature. The algorithms of [Chandy and Misra 1988; Bagrodia 1989a; Park and Kim 1990] employ a fixed coordinator for each interaction and thus cannot realize concurrent activations of a first order interaction, nor can they tolerate the failure of the coordinator. The algorithms of [Ramesh 1987; Kumar 1990] employ mutual coordination among the participants, but are highly dependent on the fact that the participants of each interaction are fixed. Therefore, there is no straightforward way to adapt these algorithms to a first order setting. Summary of Technical Results The main ....

[Article contains additional citation context not shown here]

Ramesh, S. 1987. A new and efficient implementation of multiprocess synchronization. In Proceedings Conference on PARLE, Lecture Notes in Computer Science 259, Berlin, pp. 387--401. Springer-Verlag.

....of a first order interaction can be concurrently activated. ffl As more than one process can enrole into the same role, the failure of any one of them should not preclude the others from filling the role. Thus, first order interactions are inherently fault tolerant. The distributed algorithms of [17, 3, 2, 13, 15] for multiparty interaction guard scheduling are zerothorder in nature. The algorithms of [3, 2, 15] employ a fixed coordinator for each interaction and 2 In IP the situation is somewhat different. IP s team is an abstraction mechanism for multiparty interactions, the zeroth order construct of ....

....construct of IP. Enrolement into a team does not delay a process, but rather the multiparty interactions to be executed first within the role [6] thus cannot realize concurrent activations of a firstorder interaction, nor can they tolerate the failure of the coordinator. The algorithms of [17, 13] employ mutual coordination among the participants, but are highly dependent on the fact that the participants of each interaction are fixed. Therefore, there is no straightforward way to adapt these algorithms to a first order setting. Summary of Technical Results The main contribution of this ....

Ramesh, S. "A new and efficient implementation of multiprocess synchronization," in: Proc. Conference on PARLE, Lecture Notes in Computer Science 259, Springer-Verlag, Berlin, 1987, pp. 387401.

....by mapping each interaction to a node and placing an edge between interactions that share a common process. We concentrate on the binary version of the problem in which every interaction has two members. Both the binary and the general versions of the problem have been solved by many researchers [4, 5, 6, 12, 18, 19]. Sistla [19] calls a solution to the problem to be real time if its complexity is independent of the size of the communication network. He also investigates the formation of waiting chains and then uses this to derive a lower bound on the response time. We define the failure locality of solutions ....

S. Ramesh. A new and efficient implementation of multiprocess synchronization. In PARLE Parallel Architectures and Languages Europe. Lecture Notes in Computer Sciences 259, pages 387--401, Springer-Verlag, Berlin, 1987.

..... P4 external actions Fig. 2. A central implementation. Traditionally, correctness proofs for multiway synchronization algorithms [2, 6, 10, 14, 15] amount to proving particular safety properties such as deadlock freedom and particular liveness properties such as an enabled synchronization must eventually be scheduled for execution . But our implementation strategy admits a more general correctness assertion, namely that the implementation ....

....strategy does not exclude any synchronization possibilities in the specification. Thus the implementation will inherit all global properties such as fairness or deadlock freedom from the specification. Other proposals for distributed algorithms for multiway synchronization can be found in [2, 6, 10, 11, 14, 15]. We have given priority to verification and thus chosen a simple algorithm, so the algorithm is not optimal, in terms of the number of messages required to establish a synchronization, and there are several possible improvements of the algorithm. For a discussion on this topic we refer to [17] ....

S. Ramesh. A new and efficient implementation of multiprocess synchronization. In Parallel Architectures and Languages Europe, volume 259 of Lecture Notes in Computer Science, pages 387--401. Springer-Verlag, 1987.

....Let M = PM ; I M ) Assume (1) jI M j 2, or (2) 8x; y; z 2 I M , if P x P y 6= and P x P z 6= then either P x P y P z or P x P z P y . Then SIF can be implemented for M . 6 Concluding Remarks Numerous algorithms have been proposed for scheduling multiparty interactions, e.g. [16, 5, 15, 12, 17, 10]. From our results, it is not surprising to see that only weak interaction fairness 4 has been widely implemented in these algorithms. Our impossibility results for SIF and SPF improve upon previous results by Tsay and Bagrodia [18] and by Joung [9] in three ways: First, our results do not 4 ....

S. Ramesh. A new and efficient implementation of multiprocess synchronization. In Proc. Conf. on PARLE, LNCS 259, pp. 387--401, 1987.

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