L. Lamport. Using Time Instead of Timeout for FaultTolerant Distributed Systems. ACM TOPLAS, 6:254--280, 1984.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....a reduction of our set of allowable algorithms to a more restrictive set of uniform algorithms. Section 5 contains a restatement of the relevant results of [PSL] Section 6 co.tains our main lower bound result. Section 7 contains an important open question. The reader is urged to read [LSP] and [L] for discussion of the practical importance of assuring interactive consistency, and [IdSL] for additional results not immediately relevant to this paper. Other related papers are [DW] and [D] 2. Motivation for the Definitions A general model for solving the interactive consistency problem ....

Lamport, L. Using Time Instead of Timeout For Fault-Tolerant Distributed Systems. June 1981.

....is possible and is described in [20] 6. DISCUSSION 6.1 Related Work Few general techniques have been developed to aid in the design of programs that must cope with operational failures in hardware or support software. One paradigm, based on the use of state machines, was pioneered by Lainport [12, 22]. A program is viewed as a state machine that receives input, generates actions (output) and has an internal state. A reliable system is constructed by replicating these state machines and running them in parallel. By using a solution to the Byzantine Generals Problem, each machine is guaranteed ....

LAMPORT, L. Using time instead of timeout for fault-tolerant distributed systems. Tech. Rep. 59, SRI Int., June 1981.

....is not assumed; instead, state information is replicated at other processors. However, this turns out to be just an approximation of the Stable Storage Property. The only work we know of that does not involve fail stop or stronger assumptions about processor failures is described in [6] [7], and [9] ACM Transactions on Computer Systems, Vol. 2, No. 2, May 1984. Real processors do not satisfy the Halt on Failure, Failure Status, or Stable Storage properties. In fact, most real processors are not even good approximations of fail stop processors. This is disappointing in light of ....

....of a k fail stop processor is an application of the state machine approach, a general approach for constructing distributed programs first ACM Transactions on Computer Systems, Vol. 2, No. 2, May 1984. described in [5] and later extended for environments in which failures could occur in [6, 7] and [13] Given any program, a distributed version that can tolerate up to k failures can be constructed by running that program on 2k I processors connected by a communications network in which message origins can be au thenticated. 2 Byzantine agreement is used to ensure that each instance ....

LAMPORT, L. Using time instead of timeout for fault-tolerant distributed systems. Op. 59, Computer Science Laboratory, SRI International, Menlo Park, California, June

....general techniques have been developed to aid in the design of programs that must cope with operational failures in hardware. One approach, based on the use of state machines, was pioneered by Lamport [Lamport 78a] and later extended for environments in which failures could occur in [Lamport 78b] Lamport 81] Schneider 82] The implementation of a k failstop processor described in section 5 is an application of this technique. In the state machine approach, a program is viewed as a state machine that receives input, generates actions (output) and has an internal state. Given any program, a ....

Lamport, L. Using Time Instead of Timeout for Fault-Tolerant Distributed Systems. Technical Report 59, SRI International, June 1981.

....should regard m as a late message if m arrives after the expiry of a 2d timeout that was set soon after its logical clock has exceeded T. Thus, maintaining logical clocks and setting of timeouts can provide processors with a means to detect late and absent messages. Lamport s subsequent work [Lampo84] presents a generic, Byzantine fault tolerant solution which is based on synchronised clocks and is useful for solving not just the message ordering problem but also other fundamental problems of distributed computing such as distributed semaphore, transaction commit, etc. Following the ....

....a generic, Byzantine fault tolerant solution which is based on synchronised clocks and is useful for solving not just the message ordering problem but also other fundamental problems of distributed computing such as distributed semaphore, transaction commit, etc. Following the terminology of [Lampo84], we will call the synchronised clock approach the time based approach, and the logical clock approach the timeout based one. Lampo84] also demonstrates that the time based solutions work faster compared to the timeout based ones, if clocks can be synchronised using low level processes with fast ....

[Article contains additional citation context not shown here]

L. Lamport, "Using Time Instead of Timeout for Fault-Tolerant Distributed Systems", ACM Transactions on Programming Languages and Systems, 6 (2) pp. 254-280, April 1984.

....broadcast problem is deterministically solvable in synchronous systems) enables all events that affect the state of a system (failures, joins, updates) to be seen by all correct processes in the same order. Such a total order on events can greatly facilitate fault tolerant distributed programming [Lamport 1984]. On the other hand, in time free asynchronous systems the impossibility of deterministically solving consensus implies that processes cannot reach a consensus in bounded time on a common group membership. So, the liveness properties can only require that the probabilities that a starting process ....

L. Lamport, "Using Time Instead of Timeout for Fault-Tolerant Distributed Systems", ACM Transactions on Programming Languages and Systems, 6 (2), pp.254-80, April 1984.

....to maintain a consistent global time. A fast and accurate time keeping mechanism can also be of help in implementing robust real time distributed computing systems. Solutions to several important problems in distributed computing can be simplified if a global time base is available in the system [8, 9]. Maintaining a consistent global notion of time in a distributed computing environment involves synchronization of all the local clocks. Clock synchronization problem has been studied extensively in the past and several solutions have been proposed [10, 11, 12, 13, 14, 15] However, most of these ....

L. Lamport, "Using time instead of timeout for fault-tolerant distributed systems," ACM Transactions on Programming Languages and Systems, vol. 6, pp. 254--280, April 1984.

....(xAMp) is presented. To our knowledge this is the first real time atomic protocol to support totally ordered preemption of state machine commands. Keywords: Distributed Systems, Fault Tolerance, Real Time, Atomic Broadcast, Replicated State Machine. 1 Introduction The state machine approach[10,16] is a general paradigm to implement fault tolerant distributed applications. In particular, it provides a clean and elegant abstraction to design systems based on active replication [11] a technique well suited for real time since it relies on error masking[6] avoiding the error ....

Leslie Lamport. Using Time Instead of Timeout for Fault-Tolerant Distributed Systems. ACM Transactions on Prog. Lang. and Systems, 6(2), April 1984.

....to maintain a consistent global time. A fast and accurate time keeping mechanism can also be of help in implementing robust real time distributed computing systems. Solutions to several design problems in distributed computing can be simplified if a global time base is available in the system [12]. Maintaining a consistent global notion of time in a distributed computing environment involves synchronizing all the local clocks. Clock synchronization problem has been studied extensively in the past and several solutions have been proposed [9] 16] 5] 13] 14] 1] But, most of these ....

L. Lamport. Using time instead of timeout for fault-tolerant distributed systems. ACM Transactions on Programming Languages Syst., 6(2):254--280, April 1984. An Accurate Time-management Unit for Real-time Processors 11

....for the problem in synchronous and asynchronous systems [4] The fault tolerant community has also benefited significantly from the traditional work on distributed systems to maintain consistency across cooperating server replicas. These works include examination of the concepts of logical time [5], causal ordering in message delivery [6] consensus protocols [7] What seems to be lacking is a systematic deployment of the theoretical techniques in realworld scenarios. It was not very clear how the various mechanisms would fit together to give a holistic view of a system capable of ....

L. Lamport, "Using Time instead of Timeout for Fault-Tolerant Distributed Systems," ACM Transactions on Programming Languages and Systems, pp. 254--280, April 1984.

....processors to convey systemspecific state updates to the other correct processors) enable all events that affect the state of a system to be seen by all correct processors in the same order. This total event ordering can substantially simplify the programming of replicated fault tolerant services [Lam84]. 4 The periodic broadcast membership protocol We begin our presentation of membership protocols by introducing a simple periodic broadcast protocol. In sketching it, we first make the simplifying assumption that j, the bound on task scheduling delays, is 0. We relax this assumption later ....

L. Lamport. Using time instead of timeout for fault-tolerant distributed systems. ACM Transactions on Programming Languages and Systems, 6(2):254--280, 1984.

....difficult. When the system is reliable it is however relatively easy to get a consistent global state through which the system has passed or could have passed [5] It is also possible to take consistent decisions based on past global states when the system, albeit unreliable, is synchronous [6, 16]. However, when the system is asynchronous, i.e. when there is no timing assumption neither on message transfer delay nor on process scheduling delay, the capture of consistent global states, or the taking of consistent decisions, become hard problems or even impossible problems, if failures can ....

Lamport L. Using Time instead of Timeouts for Fault-Tolerant Distributed Systems. ACM Transactions on Programming Languages and Systems, 6(2):254-280, 1984.

.... ISO hierarchy: Tan88, Com91, CS91, CS93, ANSA91a, ANSA91b, ANSA89, CD90, CDK94, XTP95] Pros and Cons of layered architectures: CT87, RST88, RST89, Ous90, AP93, KP93, KC94, BD95] Reliable stream communication: Rit84, Jac88, Tan88, Com91, CS91, CS93, CDK94] Failure Models and Classification: [Lam78b, Lam84, Ske82b, FLP85, ST87, CD90, Mar90, Cri91a, CT91, CHT92, GR93, SM94]. Kenneth P. Birman Building Secure and Reliable Network Applications 44 44 2. Communication Technologies Historically, it has rarely been necessary to understand details of the hardware components from which a computing system was constructed if one merely wishes to develop software for it. ....

.... say that one event occurs and then another does so, or that two events are concurrent, if no means is available by which a program could label events and compare their times of occurrence Looking at this question in 1978, Leslie Lamport proposed a model of logical time that answers this question [Lam78b, Lam84]. Lamport considered sets of processes (they could be static or dynamic) that interact by message passing. In his approach, the execution of a process is modeled as a series of atomic events, each of which requires a single unit of logical time to perform. More precisely, his model represents a ....

[Article contains additional citation context not shown here]

Leslie Lamport. Using Time Instead of Timeout For Fault-Tolerant Distributed Systems. ACM Transactions on Programming Languages and Systems 6:2 (April 1984), 254280.

....or not the failure of a process to send an expected message can be detected. If so, then the expectant receiver gains the valuable knowledge that the sender is faulty. In a model with accurate clocks and bounds on message transit times, such detection is possible through the use of timeouts. cf. [21]. Also, detection is automatic in a synchronous model in which the processes run in lock step and messages sent at one step are received at the next. However, detection is impossible in a fully asynchronous model in which no assumptions are made about relative step times or message delays, for ....

....systems which actually have the reliability and synchronization properties that were assumed in the models. Real distributed systems are quasi asynchronous, and to avoid the difficulties of Theorem 4 one must make reasonable timing assumptions and make effective use of clocks and timeouts. Lamport [21] gives some insights as to how this can be done. Finally, we should mention the papers by Dolev and Strong [13] and Mohan, Strong, and Finkelstein [28] that describe serious attempts to apply agreement protocols to real problems of distributed databases. 7 Acknowledgement The author is grateful ....

L. Lamport. Using time instead of timeout for fault-tolerant distributed systems. ACM Transactions on Programming Languages and Systems, to appear. See also technical report, Computer Science Laboratory, SRI International (June 1981).

No context found.

Leslie Lamport. Using time instead of timeout for fault-tolerant distributed systems. ACM Transactions on Programming Languages and Systems, 6(2):254--280, April 1984.

....already been proved correct. Functions are much easier to design, and to get right, than distributed algorithms. The first algorithm for implementing an arbitrary state machine appeared in [Lamport 1978] Later, algorithms were devised to tolerate up to any fixed number f of arbitrary failures [Lamport 1984]. These algorithms guarantee that, if fewer 24 Leslie Lamport than f processes fail, then state machine commands are executed within a fixed length of time. The algorithms are thus suitable for applications requiring realtime response. 14 But if more than f failures occur, then di#erent ....

Lamport, L. 1984. Using time instead of timeout for fault-tolerant distributed systems. ACM Trans. on Programm. Lang. Syst. 6, 2 (April), 254--280.

....that a process know the values # # min , # # max , and # # for messages it receives over the path #, which usually requires knowledge of the values of # c min and # c max for each channel c in the path. If these values can change, then new values can be broadcast 8 using the method of [3], which assures that the same values are used by all processes. For simplicity, I assume that the # # min , # # max , and # # are constants for each fixed path #. If R is an interval and # a path, then R # is defined to be the interval R [# # min , # # max ] Suppose that # is a ....

....is true for any # # #, which implies the desired result # # #. Thus, p and q must agree upon a set of Universal Time providers whose values they will use in computing C (i) p and C (i) q . This set may change for di#erent values of i (di#erent resynchronizations) The method described in [3] can be employed to obtain agreement on the current set of Universal Time providers that are to be used. Here, let us assume that the values UT (1) UT (m) from m providers are used. To perform the i th resynchronization, each process p obtains a set of intervals U (1) p , ....

[Article contains additional citation context not shown here]

Leslie Lamport. Using time instead of timeout for fault-tolerant distributed systems. ACM Transactions on Programming Languages and Systems, 6(2):254--280, April 1984.

....executed concurrently. Those instances that are not actually carrying out any interesting computation can be implemented in a trivial way by letting all of their messages be special null messages that are not actually sent. This trick of sending a message by not sending a message is also used in [Lam84] to give fault tolerant distributed simulations of centralized algorithms. Another consensus problem is establishing and maintaining synchronized local clocks in a distributed system. It is closely related to both of the preceding problems (reaching approximate agreement and achieving ....

Leslie Lamport. Using time instead of timeout for fault-tolerant distributed systems. ACM Transactions on Programming Languages and Systems, 6(2):254--280, April 1984.

No context found.

L. Lamport. Using Time Instead of Timeout for FaultTolerant Distributed Systems. ACM TOPLAS, 6:254--280, 1984.

No context found.

L. Lamport. Using Time Instead of Timeout for FaultTolerant Distributed Systems. ACM TOPLAS, 6:254--280, 1984.

No context found.

L.Lamport, "Using Time Instead of Timeout for Fault-Tolerant Distributed Systems," ACM Transactions on Programming Languages and Systems, Vol.6, No.2, Apr.1984.

No context found.

L. Lamport. Using time instead of timeout for fault-tolerant distributed systems. ACM Transactions on Programming Languages and Systems, 6(2):254--280, April 1984.

No context found.

L. Lamport. Using Time Instead of Timeout for Fault-Tolerant Distributed Systems. ACM TOPLAS, 6:254--280, 1984.

No context found.

31 L. Lamport. Using Time Instead of Timeout for Fault-Tolerant Distributed Systems. ACM Trans. on Progr. Languages and Syst., 6(2):254--280, April 1984.

No context found.

Lamport, L. Using Time Instead of Timeout for Fault-Tolerant Distributed Systems. Technical Report 59, SRI International, June 1981.

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