D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. In Journal of the ACM 34, 1987.  Home/Search   Document Not in Database   Summary   ACM   TOC   Related Articles   Check

....result has been misunderstood by a large community of system implementors [16] but has challenged other researchers to find a set of minimal assumptions that, when satisfied by an asynchronous distributed system, makes the Consensus problem solvable in this system. Minimal synchronism [5], partial synchrony [7] and unreliable failure detectors [3] constitute answers to this challenge. In this chapter, we consider the unreliable failure detectors formalism. 3.3 Unreliable Failure Detectors The unreliable failure detectors formalism, introduced by Chandra and Toueg in [3] is a ....

Dolev D., Dwork C. and Stockmeyer L. On the Minimal Synchronism Needed for Distributed Consensus. Journal of the ACM, 34(1):77--97, January 1987.

....and an adversary scheduler, which tries to prevent them from doing so. At each step, the scheduler decides which processor operates next. How much information should be made available to the adversary for making this decision Most previous papers assume a dynamic adversary scheduler (e.g. [12, 10, 11, 1, 3, 14, 4]) having full knowledge of the entire state of the system at all previous steps. In the present work, we make a distinction between two separate components of the system s history: the control history and the content history. The control history includes all information regarding the dynamics of ....

....Related Work. Fischer, Lynch, and Paterson [12] prove the impossibility of deterministic asynchronous consensus in the message passing model, even if only one processor fails. Chor, Israeli, and Li [10] and Loui and AbuAmara [15] show that the same holds in the shared memory model (see also [11]) Randomized solutions for the standard, full knowledge adversary scheduler, are presented in [10, 1, 3, 8, 4] All of these solutions require O(n) operations per processor. Our model and results differ in several ways. On the one hand, we consider a weaker adversary scheduler. In addition, we ....

D. Dolev, S. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--97, January 1987.

....membership (see [41, 11] for survey) virtual synchrony [5] atomic broadcast [10] total ordering of messages [29, 19] etc. Consensus is known to be unsolvable in most realistic models such as asynchronous message passing systems [18] and asynchronous shared memory with read write registers [36, 27, 16] if even a single process can fail by crashing. While it is usually straightforward to guarantee the consistency of a consensus decision alone (safety) the difficulty is in guaranteeing progress in face of uncertainty regarding process failures. The usual approaches to circumventing Consensus ....

....of a consensus decision alone (safety) the difficulty is in guaranteeing progress in face of uncertainty regarding process failures. The usual approaches to circumventing Consensus impossibility include strengthening the basic model by assuming different degrees of synchrony (see e.g. [16, 17, 14]) augmenting the system with unreliable failure detectors [10] and employing randomization (see a survey in [13] Specifically, our solution uses one of the most widely deployed implementations of the state machine replication [30, 45] the Paxos algorithm [31, 32, 15, 34] At the core of Paxos ....

D. Dolev, C. Dwork and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM 34(1):77--97, January 1987.

.... the theoretical impossibility of solving these problems in time free asynchronous systems, several authors have investigated mechanisms that add a sufficient amount of synchrony to the time free asynchronous system model to allow deterministic implementations for these basic distributed services [9, 10, 1]. This paper investigates under which conditions leader election and consensus services can be implemented in the timed asynchronous system model, a model that has been implicitly in use for some time (e.g. 3] was named in [8] and is formally defined in [7] We call conditions which restrict ....

....detector [1] for implementing consensus. Failure detectors do not seem appropriate for describing timed systems since they are time free. The always eventually majority stable progress assumption is strictly weaker than the process and communication assumptions of the global stabilization model [9], the model that resembles ours the most. However, the timed asynchronous system model assumes the existence of local hardware clocks. Unlike the other models, the timed asynchronous system model allows crashed processes to recover. This reflects the often used technique of automatically ....

[Article contains additional citation context not shown here]

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--97, Jan 1987.

....of an application. 2 Related Work Many problems such as consensus [17] and weak membership [3] cannot be solved in asynchronous systems. Several approaches to overcome this problem have been proposed: a) the usage of randomization [2] b) the introduction of partially synchronous models [12, 14], failure detectors [4] and progress assumptions [15] and (c) the investigation of weaker problems [13, 1] Fail awareness is a method for transforming problems into weaker problems such that they become implementable in timed asynchronous systems. However, fail awareness can be combined with ....

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--97, Jan 1987.

....relatively rare event of unexpected situations happening. Various definitions of normal circumstances are made, in terms of synchrony or reliability assumptions about the network. There are, for example, partially synchronous systems where delays and relative processor speeds are bounded [12, 14], failure detectors [9, 23, 28] that abstract away from the details of how a processor suspects failures without referring to particular synchrony assumptions, or group communication systems [11] that assume the network eventually stabilizes. In another approach the problem requirements are ....

Dolev D., Dwork C. and Stockmeyer L., On the Minimal Synchronism Needed for Distributed Consensus. Journal of the ACM, 34(1):77--97, 1987.

....case where the problem definition is weakened to allow aontermination if certain nice timing conditions (i.e. upper bounds on message delivery time) fail. An in teresting technical open question remains about the time requirements for consensus in the model of [46] Dolev, Dwork and Stockmeyer [41] noticed that there were several different kinds of asynchrony in the execution model of [55] e.g. asynchrony of messages and of processes. They classified systems, based on the various combinations of these factors, and ob tained impossibility results for many of these cases (and algorithms ....

....deciding process. In some cases, they obtained impossibility of 2 resilient consensus, rather than 1 resilient consen sus (because analysis of the decision point showed that the rest of the system might have to proceed without 2 processes, in order to produce a contradiction) The models used in [41] are quite detailed and spe cialized, for example, in their assumption that time has a minimum granularity; it makes me think that a more abstract or general model could have been used here. For example, some recent work on impossibility results for atomic register problems seems quite similar to ....

[Article contains additional citation context not shown here]

D.' Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77-97, 1987.

.... clients, to be deployed within a partially synchronous distributed system to overcome the FLP impossibility result [6] A partially synchronous distributed system is an asynchronous distributed system with some additional timing assumptions on message transfer delays, process execution steps [9], etc. Partial synchrony can be easily guaranteed on small distributed systems deployed over either a LAN or a CAN (controlled area network) 20 Therefore, enforcing strong replica consistency among a set of replicas of a service deployed across an asynchronous distributed system, such as the ....

Dolev D, Dwork C, Stockmeyer L. On the minimal synchronism needed for distributed consensus. Journal of the ACM 40 1987; 34(1):77--97.

....synchronous distributed system as they rely on distributed agreement protocols to maintain replica consis tency. A partially synchronous system is an asynchronous distributed system with some additional timing assumptions e.g. bounds on message transfer delays, process execution speeds etc. [12]. Partial synchrony can be guaranteed (most of the time) only on small size distributed systems deployed over either a LAN or a CAN (Controlled Area Network) This makes then impossible in practice the deployment of server replicas implementing a stateful service over an asynchronous distributed ....

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of theACM, 34(1):77 97, January 1987.

....unreliable failure detectors, but there are few papers about implementing of these detectors [12] In this paper we investigate how to implement and dynamically adapt failure detectors. We propose a new implementation of an Eventually Perfect ( P ) failure detector in models of partial synchrony [6]. This implementation is a variant of the heartbeat detector which is adaptable and can support scalable applications. Our algorithm is based on all to all communications where each process periodically sends an I am alive message to all processes using IP Multicast capabilities. To provide a ....

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--97, 1987.

....are discussed starting in Section 3. An excellent survey on work using randomization up to 1989 can be found in [29] 2. Timing assumptions. Consensus can be achieved despite the FLP result by adding timing assumptions to the model that exclude bad executions. Dolev, Dwork, and Stockmeyer [31] characterize the effects of adding limited synchrony assumptions to a message passing system, and show in particular that consensus becomes possible with up to n faulty processes under a variety of restrictions on the order in which processes take steps or messages are delivered. Dwork, Lynch, ....

Danny Dolev, Cynthia Dwork, and Larry Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77-97, January 1987. 27

....distributed system, as they rely on distributed agreement protocols . A partially It is well known that these primitives are not implementable synchronous system is an asynchronous distributed system with some additional timing assumptions on message transfer delays, process speeds etc. [14, 15]. As examples, the timed asynchronous model [17] or the asynchronous system model with unreliable failure detectors [10] are partially synchronous distributed systems models. Partial synchrony can be guaranteed (most of the time) only on small size distributed systems deployed over either a LAN or ....

D. Dolev, C. Dwork, and L. Stockmeyer. On the Minimal Synchronism Needed for Distributed Consensus. Journal of the ACM, 34(1):77--97, January 1987.

....synchronous distributed system as they rely on distributed agreement protocols to maintain replica consistency. A partially synchronous system is an asynchronous distributed system with some additional timing assumptions e.g. bounds on message transfer delays, process execution speeds etc. [12]. Partial synchrony can be guaranteed (most of the time) only on small size distributed systems deployed over either a LAN or a CAN (Controlled Area Network) This makes then impossible in practice the deployment of server replicas implementing a stateful service over an asynchronous distributed ....

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--97, January 1987.

....be solved deterministically in an asynchronous system in which processes can fail. This result generated a series of works that tried to identify the amount of synchrony needed to solve Consensus in the presence of failures, and showed how to solve Consensus in these partially synchronous systems [3, 4]. An alternative and elegant approach to circumvent the unsolvability of Consensus in asynchronous systems was proposed by Chandra and Toueg [2] They augmented the asynchronous model of computation with unreliable failure detectors. Informally, an unreliable failure detector is a distributed ....

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--98, January 1987.

....be solved deterministically in an asynchronous system in which processes can fail. This result generated a series of works that tried to identify the amount of synchrony needed to solve Consensus in the presence of failures, and showed how to solve Consensus in these partially synchronous systems [4, 5]. An alternative and elegant approach to circumvent the unsolvability of Consensus in asynchronous systems was proposed by Chandra and Toueg [3] They augmented the asynchronous model of computation with unreliable failure detectors. Informally, an unreliable failure detector is a distributed ....

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--98, January 1987.

....of consensus protocols, Chandra and Toueg [4] proposed extending the network with failure detectors. However, the leader election problem can be solved if and only if a perfect failure detector is available one that suspects no alive processes, and eventually suspects every faulty one [20] [6] discusses several weakened system models and what types of consensus are possible in these models, while [7] presents a weakened asynchronous model which assumes that message deliveries are always time bounded. Since real systems lack such guarantees, these results have been valuable mostly in ....

D. Dolev, C. Dwork, L. Stockmeyer, "On the minimal synchronism needed for distributed consensus", JACM, vol. 34, no. 1, Jan 1987, pp. 77-97.

....correctness. Even in faultless runs, Total may further delay admittance of a message after any number of acknowledgments had been received from all other machines. 1.1.2 Total Ordering and Fault Tolerance The total ordering of messages requires a consensus decision. Dolev, Dwork and Stockmeyer [11] show that with the presence of a virtually synchronous broadcast protocol, such that all the machines see the messages in the same (total) order, it is possible to achieve consensus with any number of faulty machines. This is easily done by having all the machines send their initial value and ....

....Proof: Assume in contradiction the existence of a block free total ordering protocol P s.t. any schedule starting at a non quiescent configuration and involves all the machines delivers some message. Using P we can implement a protocol contradicting Theorem 6. 4 in the following simple way (see [11]) every machine sends an agreed multicast message with its initial value to all other machines. The register p:O 1 (holding the first message to be delivered) is regarded as p s decision value. It is not difficult to prove that this is a blockfree weak consensus protocol s.t. every schedule ....

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. In Proc. of the 24th IEEE Annual Symp. on Foundation of Computer Science, number 24, pages 393--402, 1983.

....a fee and or specific permission. 1988 ACM 0004 5411 88 0400 0288 01.50 Journal of the Association for Computing Machinery, Vol. 35, No. 2, April 1988, pp. 288 323. 1. Introduction 1.1 BACKGROUND. The role of synchronism in distributed computing has recently received considerable attention [1, 4, 10]. One method of comparing two models with differing amounts or types of synchronism is to examine a specific problem in both models. Because of its fundamental role in distributed computing, the problem chosen is often that of reaching agreement. See [8] for a survey; see also [6] 11] 12] ....

....any number of faults can be tolerated. For Byzantine faults without authentication, t resilient consensus is possible iff N 3t I [14, 15] Recent work has shown that the existence of both bounds x and is necessary to achieve any resiliency, even under the weakest type of faults. Dolev et al. [4], building on earlier work of Fischer et al. 10] prove that if either a fixed upper bound x on message delivery time does not exist (communication is asynchronous) or a fixed upper bound on relative processor speeds does not exist (processors are asynchronous) then there is no consensus ....

[Article contains additional citation context not shown here]

DOLEV, D., DWORK, C., AND STOCKMEYER, L. On the minimal synchronism needed for distributed consensus. J. ACM 34, I (Jan. 1987), 77-97.

....is divided into synchronous unison steps separated by rounds of message ex change (see, e.g. DS, LSP] A message sent at step s from a correct processor p to a correct processor q is received by q at step s l. This assumption is justified by the impossibility results of [FLP] and [DDS] which show that if the system is asynchronous then there is no protocol for distributed agree ment tolerant to even one benign processor failure. Another common assumption is that all processors begin the algorithm simultaneously, i.e. at the same step. In an actual distributed system in ....

Dolev, D., Dwork, C., and Stockmeyer, L., On the minimal synchronism needed for distributed consensus, Proc. 24th Syrup. on Foundations of Computer Science, 1983, pp. 393-402.

....ensures that all nonfaulty processes halt, but different processes are allowed to terminate at different times. For asynchronous systems, in which a very slow process cannot be distinguished from a dead process, exact agreement cannot be reached by any algorithm that is guaranteed to terminate [5, 9]. Exact agreement can, however, be attained by algorithms that only terminate with probability 1 [1, 3] An interesting contrast to the results in [5] and [9] is our second algorithm, which enables processes in an asynchronous system to get as close to agreement as one chooses. Our algorithm for ....

.... slow process cannot be distinguished from a dead process, exact agreement cannot be reached by any algorithm that is guaranteed to terminate [5, 9] Exact agreement can, however, be attained by algorithms that only terminate with probability 1 [1, 3] An interesting contrast to the results in [5] and [9] is our second algorithm, which enables processes in an asynchronous system to get as close to agreement as one chooses. Our algorithm for the asynchronous case also works by successive approximation. In this case, however, the total number of processes required by the algorithm is more ....

[Article contains additional citation context not shown here]

DOLEV, D., DWORK, C., AND STOCKMEYER, L. On the minimal synchronism needed for distributed consensus. In Proceedings of 24th Annual Symposium on Foundations of Computer Science (Nov.). IEEE, New York, 1983, pp. 393-402.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. In Journal of the ACM 34, 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--97, January 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. In Journal of the ACM 34, 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. In Journal of the ACM 34, 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM (JACM), 34(1):77--97, 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--98, January 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--97, January 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--97, January 1987.

No context found.

Danny Dolev, Cynthia Dwork, and Larry Stockmeyer, "On the minimal synchronism needed for distributed consensus," Journal of the ACM (JACM), vol. 34, no. 1, pp. 77--97, 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer, "On the minimal synchronism needed for distributed consensus," 24th Annual Symposium on Foundations of Computer Science, Nov. 1983.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. "On the Minimal Synchronism Needed for Distributed Consensus". Journal of ACM, 34(1):77--97, 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--98, January 1987.

No context found.

Dolev D., Dwork C. and Stockmeyer L. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1), pp. 77--97, January 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. "On the Minimal Synchronism Needed for Distributed Consensus". Journal of ACM, 34(1):77--97, 1987.

No context found.

D. Dolev, S. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77-97, January 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer, \On the Minimal Synchronism Needed for Distributed Consensus," Journal of the ACM, vol. 1, pp. 77-97, January 1987. 24

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. J. ACM, 34(1):77--97, 1987. Sridhar, Hallstrom, and Sivilotti

No context found.

Dolev D., Dwork C. and Stockmeyer L., On the Minimal Synchronism Needed for Distributed Consensus. Journal of the ACM, 34(1):77-97, 1987.

No context found.

D. Dolev, C. Dwork and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM 34(1):77--97, January 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--97, 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer, \On the Minimal Synchronism Needed for Distributed Consensus," Journal of the ACM, vol. 1, pp. 77-97, January 1987. 12

No context found.

Danny Dolev, Cynthia Dwork, and Larry Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--97, 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer, \On the Minimal Synchronism Needed for Distributed Consensus," Journal of the ACM, vol. 1, pp. 77-97, January 1987.

No context found.

Dolev, D., C. Dwork, and L. Stockmeyer. On the Minimal Synchronism Needed for Distributed Consensus. Journal of the ACM, 34(1):77--97, January 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. In Journal of the ACM 34, 1987.

No context found.

Dolev D., Dwork C. and Stockmeyer L., On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1), pp:77--97, (1987).

No context found.

D. Dolev, C. Dwork and L. Stockmeyer, "On the Minimal Synchronism Needed for Distributed Consensus," Journal of the ACM, vol. 34, no. 1, Jan. 1987, pp. 77--97.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. J. ACM, 34(1):77-- 97, Jan. 1987.

No context found.

D. Dolev, C. Dwork, and L. Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1):77--97, 1987.

No context found.

Danny Dolev, Cynthia Dwork, and Larry Stockmeyer. On the minimal synchronism needed for distributed consensus. Journal of the ACM, 34(1), pages 77--97, January 1987.

First 50 documents  Next 50

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