John Turek and Dennis Shasha. The many faces of consensus in distributed systems. IEEE Computer, 25(6):8--17, June 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of messages cannot realistically be assumed to be independent of the algorithm that generates them. A detailed quantitative performance analysis of agreement protocols represents a huge work. Where should such a work start As most agreement problems are related to the abstract consensus problem [11, 12, 13] it seems natural to start by a performance analysis of a consensus algorithm, and to extend the work of [10] This is the goal of this paper, which analyzes a consensus algorithm. The consensus problem is defined over a set of processes. Informally, each process in this set proposes a value ....

J. Turek and D. Shasha, "The many faces of consensus in distributed systems," IEEE Computer, vol. 25, pp. 8--17, June 1992.

....free runs. This only gives a partial and incomplete understanding of their quantitative behavior. A detailed quantitative performance analysis of agreement protocols represents a huge work. Where should such a work start Most agreement problems are related to the abstract consensus problem [6, 22, 15], defined over a set of processes: each process in this set proposes a value initially, and the processes must decide on the same value, chosen among the proposed values. For this reason, it seems natural to start by a performance analysis of consensus algorithms, and to extend the work of [9] ....

J. Turek and D. Shasha. The many faces of consensus in distributed systems. IEEE Computer, 25(6):8-- 17, June 1992.

....among the correct processes on the history of groups that exist over time as well as on the membership of each group. Such a service simplifies the programming of many higher level applications. The membership problem can be viewed as a particular form of the well known consensus problem [Turek Shasha 1992], that allows processes to reach a common decision (for example, about a numerical value) which depends on their initial inputs, despite failures. The importance of the consensus problem derives from its omnipresence in the area of distributed systems. Indeed, consensus is at the basis of ....

....synchronous message order; 4) synchronous communication, broadcast transmission, and atomic receive and send. Figure 2 shows the maximum resiliency (where crash resilient protocols exist) for each setting of the five system parameters. Figure 2a presents the results for separate receive and send [Turek Shasha 1992] and Figure 2b for atomic receive and send. Empty table entries mean that, for that setting, consensus is unsolvable even in the presence of one crash failure. Circled table entries show where it is sufficient to devise a protocol in order to solve the consensus problem in all of the entries of ....

J. Turek and D. Shasha, "The Many Faces of Consensus in Distributed Systems", IEEE Computer, 25 (6), pp.8-17, June 1992.

....mainly provide transparent process execution, and communication and other distributed services. In many distributed services and user applications a group of processes running in di erent processors need to arrive at a common decision in spite of processor and transient communication failures [18]. To solve this problem a consensus protocol (also called uniform broadcast protocol) is needed. For simplicity, in what follows we will assume that each process executes on a di erent processor. The distributed consensus protocol has been studied in depth. Distributed consensus protocols dealing ....

Turek, J. and Shasha, D. 1992. The Many Faces of Consensus in Distributed Systems. IEEE Computer, V25, N6 (June).

....mainly provide transparent process execution, and communication and other distributed services. In many distributed services and user applications a group of processes running in di erent processors need to arrive at a common decision in spite of processor and transient communication failures [13]. To solve this problem a consensus protocol (also called reliable broadcast protocol) is needed. For simplicity, in what follows we will assume that each process executes on a di erent processor. The distributed consensus protocol has been studied in depth. Distributed consensus protocols ....

Turek, J. and Shasha, D. 1992. The Many Faces of Consensus in Distributed Systems. IEEE Computer, V25, N6 (June).

....destination nodes, of CMDSs originated from multiple source nodes. Thus, the principal communication models required by a DMMA are those for many to one, one to many and many to many real time communications. These models can be adequately supported by the real time group communication paradigm [6, 13, 32, 57, 59]. Hence, the algorithms we propose are designed to support reliable group communications over the k AMT abstraction introduced earlier. Those group communications over the k AMT can be based on the implementation of a message diffusion algorithm that allows its users to transmit messages to ....

J. Turek, D. Shasha, The Many Faces of Consensus in Distributed Systems, IEEE Computer, Vol. 25, N. 6, June 1992, pp. 8 - 17.

....the following property holds: before being notified of the failure of q by the timeout mechanism, p will receive all the messages q sent to it before crashing. 3. 3 The Consensus Problem This section defines the Consensus Problem ( 21] which is a fundamental problem is distributed systems (see [24] for a survey) This problem is far from being trivial in presence of failures. Solutions to and use of this problem will be addressed in Section 3.4 and in Section 7, respectively. Let us consider a set of processes that can fail by crashing. A non failed process is said to be correct. Each ....

J. Turek and D. Shasha. The Many Faces of Consensus in Distributed Systems. Computer, Vol. 25(6), (June 1992), pp. 8-17.

....distributed systems, not only from a theoretical point of view, but also from a practical point of view. Six frequent misunderstandings are discussed. Misunderstanding 1: Consensus is for theoreticians only Consensus can be viewed as a general form of agreement in distributed systems [17]. The problem is defined over a set of processes fp 1 ; p 2 ; pn g: each process p i has an initial value v i , and the correct processes (those that do not crash) have to decide on a common value v that is the initial value of one of the processes [3] This problem has attracted ....

J. Turek and D. Shasha. The Many Faces of Consensus in Distributed Systems. IEEE Computer, 25(6):8-- 17, June 1992.

....execute multiple TS operations atomically is important for using Linda to program faulttolerant applications. For example, distributed consensus, in which multiple processes in a distributed system reach agreement on some common value, is an important building block for many fault tolerant systems [40]. However, Linda with single op atomicity has been shown to be insufficient to reach distributed consensus with more than two processes in the presence of failures or with arbitrarily slow (or busy) processors [38] The key is lack of sufficient atomicity. Even typical Linda programs cannot be ....

John Turek and Dennis Shasha. The many faces of consensus in distributed systems. Computer, 25(6):8--17, June 1992.

....that any principal can eventually pass a message to any other principal. In pathological cases, as when two mobile computers are never connected at the same time, it may be necessary to communicate through other principals. These conditions are sufficient to make distributed consensus possible [Turek92]. 1.2 Frameworks A framework is an object oriented description of the components that make up a system and how they are connected. It generalizes concepts such as layered design, often used in specifying network protocols, and structured design. It is related to the Object Oriented Design ....

John Turek and Dennis Shasha. The many faces of consensus in distributed systems. IEEE Computer, 25(6):8--17, June 1992.

....multiple TS operations atomically, is important for using Linda to program fault tolerant applications. For example, distributed consensus, in which multiple processes in a distributed system reach agreement on some common value, is an important building block for many fault tolerant systems [TS92] 33 Initialization out( count ; value) Inspection rd( count ; value) Updating in( count ; oldvalue) out( count ; newvalue) Figure 2.1: Distributed Variables with Linda However, Linda with single op atomicity has been shown to be insufficient to reach distributed consensus with more ....

John Turek and Dennis Shasha. The many faces of consensus in distributed systems. Computer, 25(6):8--17, June 1992.

....client cache altogether. If volatile data stored in the server is overwritten by another client, the former client agent is notified of this event. There are several failure situations in the system. When the client has sent data to the server, the client can crash (we assume fail stop processors [15]) leaving a single copy of the data in the server machine. If the server crashes, the only data copy remains in the client cache. To protect these single copies against failures, both client and server periodically poll the other party. If there is no response within a bounded time, either the ....

John Turek and Dennis Shasha. The many faces of consensus in distributed systems. IEEE Computer, 25(6):8--17, June 1992.

....from multiple (yet again, possibly distributed) source nodes. Thus, the principal communication models required by a DMMA are those for many to one, one to many and many to many real time communication. These models can be adequately supported by the real time group communication paradigm [7, 19, 35, 37]. Hence, the DCCS communicationsoftware architecture that weproposeembodiesa collection of protocols whichsupport reliable group communications, over the k AMT abstraction introduced earlier. Those group communications over the k AMT are implemented by means of a message diffusion protocol that ....

J. Turek, D. Shasha, The Many Faces of Consensus in Distributed Systems, IEEE Computer, Vol. 25, N. 6, June 1992, pp. 8 - 17.

....(swapping data values held by a register and a memory location) instruction. 4 Consensus between Reader and Writer In the consensus problem, a set of processes must come to an agreement on a common output value; trivial solutions in which processes agree on a predetermined value are not allowed [20]. The decision made on this output value is based on initial states of the processes, not their relative speeds. Reaching consensus among asynchronous processes has been extensively studied from various aspects [1, 7, 6, 20] In particular, Herlihy showed that achieving consensus in shared memory ....

....trivial solutions in which processes agree on a predetermined value are not allowed [20] The decision made on this output value is based on initial states of the processes, not their relative speeds. Reaching consensus among asynchronous processes has been extensively studied from various aspects [1, 7, 6, 20]. In particular, Herlihy showed that achieving consensus in shared memory systems requires synchronization primitives and defined the consensus number associated with a data object shared by concurrent processes [7] The consensus number is the maximum number of processes for which the associated ....

Turek, J., and Shasha, D. The Many Faces of Consensus in Distributed Systems. IEEE Computer 26, 6 (June 1992), 8--17.

....can deliver messages synchronously, within a bounded time, or eventually in a finite but unbounded time. Strong consistency requirements are impossible to meet in the most general cases. For example, if there are no bounds on message delivery time it is not possible to guarantee consistency [12]. Further, if processes can fail in arbitrary ways, providing reliable delivery is equivalent to Byzantine Agreement. For most applications the Internet can be treated as an unreliable, bounded, broadcast (as opposed to strict point to point) network. The weak consistency protocols we have ....

J. Turek and D. Shasha, "The many faces of consensus in distributed systems," IEEE Computer, vol. 25, pp. 8--17, June 1992.

....distributed systems, not only from a theoretical point of view, but also from a practical point of view. Six frequent misunderstandings are discussed. Misunderstanding 1: Consensus is for theoreticians only Consensus can be viewed as a general form of agreement in distributed systems [17]. The problem is defined over a set of processes fp 1 ; p 2 ; pn g: each process p i has an initial value v i , and the correct processes (those that do not crash) have to decide on a common value v that is the initial value of one of the processes [3] This problem has attracted ....

J. Turek and D. Shasha. The Many Faces of Consensus in Distributed Systems. IEEE Computer, 25(6):8-- 17, June 1992.

No context found.

John Turek and Dennis Shasha. The many faces of consensus in distributed systems. IEEE Computer, 25(6):8--17, June 1992.

No context found.

J. Turek and D. Shasha, "The many faces of consensus in distributed systems," IEEE Computer, vol. 25, pp. 8--17, June 1992.

No context found.

J. Turek and D. Shasha. The many faces of consensus in distributed systems. IEEE Computer, 25(6):8--17, June 1992.

No context found.

John Turek and Dennis Shasha. The many faces of consensus in distributed systems. IEEE Computer, 25(6):8--17, June 1992.

No context found.

Turek, J. and Shasha, D.: The Many Faces of Consensus in Distributed Systems, Distributed Computing Systems, IEEE Comp. Soc. Press, pp.83-91, 1994.

No context found.

Turek, J. and Shasha, D., "The Many Faces of Consensus in Distributed Systems," Distributed Computing Systems, IEEE Computer Society Press, 1994, pp.83-91. 6

No context found.

John Turek and Dennis Shasha. The Many Faces of Consensus in Distributed Systems. IEEE Computer 25:6 (1992), 817.

No context found.

Turek, J. and Shasha, D., "The Many Faces of Consensus in Distributed Systems," Distributed Computing Systems, IEEE Computer Society Press, 1994, pp.83-91.

No context found.

Turek, J. and Shasha, D., (1992). The Many Faces of Consensus in Distributed Systems, in: IEEE Computer Magazine, pp. 8-17.

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