Ajei Gopal. Fault-Tolerant Broadcasts and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, January 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....solving URB in systems with fair links and f n 2. As mentioned above, this cannot be done without failure detectors, and [HR99] determined that failure detector G is Intuitively, a problem P is correct restricted if its specification does not refer to the behavior of faulty processes [BN92, Gop92]. Note that URB is not correct restricted. necessary and sufficient to solve this problem in . A discussion of the differences between [HR99] and this paper was given in Section 1. 11 Concluding Remarks In some environments, URB can be solved without failure detectors at all, and this seems ....

Ajei Gopal. Fault-Tolerant Broadcasts and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, January 1992.

....that require reliable send receive or uniform reliable broadcast by plugging in the implementations given in Section 9, provided a majority of processes are correct. Intuitively, a problem P is correct restricted if its specification does not refer to the behavior of faulty processes [BN92, Gop92] quiescent and terminating) instance #1 of atomic commitment Reliable Communication (quiescent, non terminating) Failure Detection (non quiescent, non terminating) instance #1 of consensus Applications Instance #2 of consensus Figure 6: Layering that separates applications, reliable ....

Ajei Gopal. Fault-Tolerant Broadcasts and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, January 1992.

....solving URB in systems with fair links and f # n 2. As mentioned above, this cannot be done without failure detectors, and [HR99] determined that failure detector G f is 8 Intuitively, a problem P is correct restricted if its specification does not refer to the behavior of faulty processes [BN92, Gop92]. Note that URB is not correct restricted. 14 necessary and sufficient to solve this problem in E f . A discussion of the differences between [HR99] and this paper was given in Section 1. 11 Concluding Remarks In some environments, URB can be solved without failure detectors at all, and ....

Ajei Gopal. Fault-Tolerant Broadcasts and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, January 1992.

....that require reliable send receive or uniform reliable broadcast by plugging in the implementations given in Section 9, provided a majority of processes are correct. 17 Intuitively, a problem P is correct restricted if its specification does not refer to the behavior of faulty processes [BN92, Gop92] 24 (quiescent and terminating) instance #1 of atomic commitment Reliable Communication (quiescent, non terminating) Failure Detection (non quiescent, non terminating) instance #1 of consensus Applications Instance #2 of consensus Figure 6: Layering that separates applications, ....

Ajei Gopal. Fault-Tolerant Broadcasts and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, January 1992.

....losses. To avoid this piggybacking, in this paper we adopted the model of fair links: message losses can now be overcome by separately sending each message repeatedly. 12 Intuitively, a problem P is correct restricted if its specification does not refer to the behavior of faulty processes [7, 19]. consider the problem of consensus for networks that may partition, and we use HB to solve this problem with a quiescent protocol (we also use a generalization of the Eventually Strong failure detector [12] 12 Quiescence versus Termination In this paper we considered communication ....

A. Gopal. Fault-Tolerant Broadcasts and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, Jan. 1992.

.... In [BCBT96] Basu et al. pose the following question: given a problem that can be solved in asynchronous systems with process crashes only, can this problem still be solved if links can also fail by losing messages They show that the answer is yes if the problem is correct restricted [BN92, Gop92] 18 or if more than half of the processes do not crash. However, the communication algorithms that they give are not quiescent (and do not use failure detectors) ACT97a] was the first paper to study the problem of achieving quiescent reliable communication by using failure detectors in a ....

Ajei Gopal. Fault-Tolerant Broadcasts and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, January 1992.

.... In [BCBT96] Basu et al. pose the following question: given a problem that can be solved in asynchronous systems with process crashes only, can this problem still be solved if links can also fail by losing messages They show that the answer is yes if the problem is correct restricted [BN92, Gop92] 12 or if more than half of the processes do not crash. However, the communication algorithms that they give are not quiescent (and do not use failure detectors) ACT97a] was the first paper to study the problem of achieving quiescent reliable communication by using failure detectors in a ....

Ajei Gopal. Fault-Tolerant Broadcasts and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, January 1992.

....stop. A precise definition of inconsistency and contamination with respect to broadcasts is beyond the scope of this paper. We also omit the description of algorithms that prevent inconsistency and or contamination. For a more complete treatment of this subject the reader is referred to [GT91,Gop92] 3.12 Amplification of Failures A fault tolerant broadcast is usually implemented by a broadcast algorithm that uses lowerlevel communication primitives, such as send and receive (Figure 3) With such a broadcast algorithm, the broadcasting or delivery of a message requires the execution of ....

....this is not always so. In general a broadcast algorithm is likely to amplify the severity of failures that occur at the low level. For example, there are Atomic Broadcast algorithms where the omission to receive messages causes a faulty process to deliver messages in the wrong order [Gop92] But what if processes are only subject to crash failures Can we assume that the message deliveries that a process makes before crashing are always correct (i.e. consistent with those of correct processes) Intuitively, this seems very reasonable, since by definition a process that crashes ....

[Article contains additional citation context not shown here]

Ajei Gopal. Fault-Tolerant Broadcasts and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, January 1992.

....result characterizes a large class of problems that remain solvable even with fair lossy links. Informally, this class consists of all the problems whose specifications refer only to the behavior of correct processes (i.e. processes that do not crash) these are called correct restricted [Gop92] or failure insensitive [BN92] problems. 2 This class of problems includes Reliable, FIFO, and Causal Broadcast, and correct restricted versions of Approximate Agreement, Renaming, and k set Agreement. For such problems, we show how to automatically transform any algorithm that works in a ....

....S 0 is a non trivial reduction of S implies S 0 2 S. Henceforth, we consider only such specifications. 2. 10 Correct Restricted Problem Specifications Intuitively, a problem specification is correct restricted if it refers only to the states of correct processes (those with infinite traces) Gop92, BN92] Formally, let H and H 0 be any two traces in Vec(S) with the same dimension, say k. Traces H and H 0 are correct equivalent, denoted H c H 0 , if for all i, 1 i k, if H[i] or H 0 [i] is infinite then H 0 [i] H[i] For any S and S 0 in S , we say that S 0 is a ....

A. Gopal. Fault-Tolerant Broadcasts and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, January 1992.

....losses. To avoid this piggybacking, in this paper we adopted the model of fair links: message losses can now be overcome by separately sending each message repeatedly. 14 Intuitively, a problem P is correct restricted if its specification does not refer to the behavior of faulty processes [Gop92, BN92] 12 Generalization to Networks that Partition In this paper, we assumed that every pair of correct processes are reachable from each other through fair paths. In [ACT97b] we drop this assumption and consider the more general problem of quiescent reliable communication in networks that ....

Ajei Gopal. Fault-Tolerant Broadcasts and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, January 1992.

....range of real world distributed systems. We believe that Dynamic Uniformity (D Uniformity) is the fundamental property required for many applications in asynchronous settings. Recently, focus has been drawn to uniform, or consistent, coordination, a strengthening of traditional agreement problems [15, 2, 12]. In non uniform agreement, faulty processes are never required to behave consistently with their cohorts, even before they commit a fault. Uniform agreement addresses this by requiring that decisions made by any process in the system, even one that is later declared faulty, must eventually be ....

Ajei Gopal. Fault-Tolerant Broadcasts and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, 1992.

.... algorithm all processes perform the same code [6, 25, 10, 4] whereas in an asymmetric algorithm one process plays a special role, i.e. defines the ordering of messages [23, 7] Asymmetric algorithms require less phases and are thus more efficient, but are subject to the contamination problem [16, 35]. Token based algorithms [11, 3] can be classified somewhere in between symmetric and asymmetric algorithms. Moreover, some of these algorithms ( 6, 7, 35] assume the dynamic group model, and an underlying layer implementing view synchronous multicast. Total order multicast is however related to ....

A. S. Gopal. Fault-Tolerant Broadcast and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, Ithaca, NY, March 1992.

.... algorithm all processes perform the same code [2, 18, 5] whereas in an asymmetric algorithm one process plays a special role, i.e. defines the ordering of messages [16, 3] Asymmetric algorithms require less phases and are thus more efficient, but are subject to the contamination problem [12, 24]. Token based algorithms [7, 1] can be classified somewhere in between symmetric and asymmetric algorithms. We consider in the paper total order algorithms from a different perspective, i.e. we consider the following issues: open vs closed group, global vs local total order, and genuine vs ....

A. S. Gopal. Fault-Tolerant Broadcast and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, Ithaca, NY, March 1992.

No context found.

A. S. Gopal. Fault-Tolerant Broadcast and Multicasts: The Problem of Inconsistency and Contamination. PhD thesis, Cornell University, Ithaca, NY, March 1992.

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