F. Pedone and A. Schiper, "Optimistic atomic broadcast," in Proc. 12th International Symposium on Distributed Computing (DISC), 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....needed. The broadcast used to send update transactions to all sites needs total order semantics. This order determines the serialization order of the transactions. We use an aggressive version, optimistic delivery total ordered broadcast [13] of the optimistic total ordered broadcast presented in [18]. Optimistic delivery in total ordered broadcast takes advantage of the fact that in a local area network, messages are often spontaneously totally ordered. This optimistic delivery broadcast is defined by three primitives. To broadcast(m) broadcasts the message # to all the sites in the system. ....

F. Pedone and A. Schiper. Optimistic Atomic Broadcast. In Proc. of DISC, 1998.

....and depends on timing assumptions. More recently, a combination of randomization and failure detectors has been proposed [AT96] but they still rely on the correct working of the failure detectors and work only in the crash failure model. Pedone and Schiper apply optimism to the consensus problem [PS98], but their approach also works only in the crash failure model and requires a reliable broadcast as well as a failure detector in the pessimistic case. To the best of my knowledge, no protocol has combined the synchronous and the asynchronous model, maintaining the strength of the asynchronous ....

F. Pedone and A. Schiper. Optimistic atomic broadcast. In Proceedings of the 12th International Symposium on Distributed Computing (DISC'98, formerly WDAG), September 1998.

....this work is that, unlike previous works, they consider partial replication, that is, each node within the system possesses a partial copy of the database. The authors base their algorithm on two communication primitives: a fast atomic broadcast protocol, that provides optimistic message delivery [22], and a Resilient Atomic Commit (RAC) protocol. This protocol provides the same guarantees as an atomic commit protocol [11] but allows for a) nodes having only partial copies of the database, and b) allows that participants in the protocol decide commit even if some nodes are suspected to have ....

....each site has a queue CQ x associated to each conflict class C x . Upon delivery of a transaction T that accesses a compound conflict class C T , each site adds T to the queues of the basic conflict classes contained in C T . The replication scheme makes use of an advanced total order algorithm [22] that provides optimistic message delivery. This means that the communication protocol first makes a tentative deliver of the message, which can be considered a good estimate on its final, definitive order. The protocol later delivers the same message in its final order. This tentative delivery is ....

[Article contains additional citation context not shown here]

F. Pedone and A. Schiper. Optimistic atomic broadcast. In Proceedings of the 12th International Symposium on Distributed Computing (DISC'98), Andros, Greece, September 1998.

....of di erent techniques to capture and express obsolescence relations is given in [30] 3.2 Optimistic Total Order We now give another example of an adaptation of a basic abstraction to create a more powerful, and ecient, new abstraction. This new abstraction, called optimistic total order [27] has been proposed to address the performance limitations of totally ordered multicast protocols. It can be shown that the problem of ordering multicast messages in a total order is, in asynchronous systems, equivalent to the consensus problem [8] On the other hand, there are also known lower ....

....the termination of the algorithm; Ensure that in most runs the estimated order is accurate with regard to the nal order. The rst proposed algorithms supporting optimistic delivery were based on characteristics speci c to concrete classes of systems. For instance, the algorithm described in [27] assumes the availability of a local area network, as the estimated order is based on the network spontaneous order. On the other hand, the algorithm proposed in [36] can operate over large scale networks but the estimate order is based on an evaluation of the network delay, and there is a ....

F. Pedone and A. Schiper. Optimistic atomic broadcast. In Proceedings of the 12th International Symposium on Distributed Computing (DISC'98), 1998.

....of di#erent techniques to capture and express obsolescence relations is given in [30] 3.2 Optimistic Total Order We now give another example of an adaptation of a basic abstraction to create a more powerful, and e#cient, new abstraction. This new abstraction, called optimistic total order [27] has been proposed to address the performance limitations of totally ordered multicast protocols. It can be shown that the problem of ordering multicast messages in a total order is, in asynchronous systems, equivalent to the consensus problem [8] On the other hand, there are also known lower ....

....termination of the algorithm; Ensure that in most runs the estimated order is accurate with regard to the final order. The first proposed algorithms supporting optimistic delivery were based on characteristics specific to concrete classes of systems. For instance, the algorithm described in [27] assumes the availability of a local area network, as the estimated order is based on the network spontaneous order. On the other hand, the algorithm proposed in [36] can operate over large scale networks but the estimate order is based on an evaluation of the network delay, and there is a ....

F. Pedone and A. Schiper. Optimistic atomic broadcast. In Proceedings of the 12th International Symposium on Distributed Computing (DISC'98), 1998.

....behind operations that need to be performed anyway. This is accomplished by processing messages in an optimistic manner and hoping that most decisions will be correct although in some cases transactions might need to be aborted. This approach builds upon recent work in optimistic multicast [PS98] and a more aggressive version of optimistic delivery proposed in the context of Postgres R [KPAS99] and later used to provide high performance eager replication in clusters [PJKA00] We use an optimistic uniform multicast that delivers messages in two steps. In the rst step messages are ....

F. Pedone and A. Schiper. Optimistic Atomic Broadcast. In S. Kutten, editor, Proc. of 12th Distributed Computing Conference, volume LNCS 1499, pages 318{ 332. Springer, September 1998.

....optimistically, which are later committed when the final definitive order is established. The goal is to execute some application steps in parallel with the communication steps of the total order algorithm. Although several algorithms that support optimistic delivery have been proposed before [23, 10], these are specialized to some specific network types [23] or interaction patterns [10] Our approach is more generic because, in stable conditions, both the optimistic and uniform order are derived from the output of a fully fledged non uniform total order algorithm. The principles of our ....

....order is established. The goal is to execute some application steps in parallel with the communication steps of the total order algorithm. Although several algorithms that support optimistic delivery have been proposed before [23, 10] these are specialized to some specific network types [23] or interaction patterns [10] Our approach is more generic because, in stable conditions, both the optimistic and uniform order are derived from the output of a fully fledged non uniform total order algorithm. The principles of our approach are the following. An e#cient algorithm that provides ....

[Article contains additional citation context not shown here]

F. Pedone and A. Schiper. Optimistic atomic broadcast. In Proc. of the 12th International Symposium on Distributed Computing (DISC'98), 1998.

....the degenerate situation where a single process is multicasting and it can assume the sequencer role. is often very similar, therefore, similar to the final ordering decided by the sequencer. Nevertheless, delivery incurs always in the additional latency. The optimistic atomic broadcast protocol [22] takes this in consideration to improve average delivery latency of a consensus based total order protocol. Further latency improvements can be obtained if the application itself can take advantage of a tentatively ordered delivery. This is called optimistic delivery [17, 26] as it rests on the ....

F. Pedone and A. Schiper. Optimistic atomic broadcast. In Proc. the 12 Int'l Symp. on Dist. Computing, Sept. 1998.

....must support replication of objects in a geographically dispersed system, in which the nodes communicate through the Internet. This prevents the use of protocols that make used of speci c network properties (such as the low latency or network order preservation properties of local area networks [18]) RDBMS independence: a variety of commercial databases should be supported as the underlying data storage technology. This prevents the use of solutions that require adaptations to the database kernel. Protocol interchangeability : Copla must be exible enough to adapt to changing ....

....order protocol, i.e. a protocol that ensures that if two messages are delivered by a given order to a process (even if this process crashes) they are delivered in that order to all correct processes. The new protocol also supports the optimistic delivery of (tentative) total order indications [8, 18]. Given that the order established by the (non uniform) total order protocol is the same as the nal uniform total order in most cases (these two orders only di er when crashes occur at particular points in the protocol execution) this order can be provided to the consistency layer as a tentative ....

F. Pedone and A. Schiper. Optimistic atomic broadcast. In Proceedings of the 12th International Symposium on Distributed Computing (DISC'98), Andros, Greece, September 1998.

....optimistically, which are later committed when the nal de nitive order is established. The goal is to execute some application steps in parallel with the communication steps of the total order algorithm. Although several algorithms that support optimistic delivery have been proposed before [23, 10], these are specialized to some speci c network types [23] or interaction patterns [10] Our approach is more generic because, in stable conditions, both the optimistic and uniform order are derived from the output of a fully edged non uniform total order algorithm. The principles of our ....

....nitive order is established. The goal is to execute some application steps in parallel with the communication steps of the total order algorithm. Although several algorithms that support optimistic delivery have been proposed before [23, 10] these are specialized to some speci c network types [23] or interaction patterns [10] Our approach is more generic because, in stable conditions, both the optimistic and uniform order are derived from the output of a fully edged non uniform total order algorithm. The principles of our approach are the following. An ecient algorithm that provides ....

[Article contains additional citation context not shown here]

F. Pedone and A. Schiper. Optimistic atomic broadcast. In Proceedings of the 12th International Symposium on Distributed Computing (DISC'98), 1998.

....must support replication of objects in a geographically dispersed system, in which the nodes communicate through the Internet. This prevents the use of protocols that make used of specific network properties (such as the low latency or network order preservation properties of local area networks [18]) RDBMS independence: a variety of commercial databases should be supported as the underlying data storage technology. This prevents the use of solutions that require adaptations to the database kernel. Protocol interchangeability: COPLA must be flexible enough to adapt to changing ....

....total order for the messages sent by passive nodes and for their own messages. They do so by applying a symmetric total order protocol based on an underlying terminating uniform reliable broadcast layer. The protocol also supports the optimistic delivery of (tentative) total order indications [8, 18]. Given that the order established by the (non uniform) total order protocol is the same as the final uniform total order in most cases (these two orders only differ when crashes occur at particular points in the protocol execution) this order can be provided to the consistency layer as a ....

F. Pedone and A. Schiper. Optimistic atomic broadcast. In Proceedings of the 12th International Symposium on Distributed Computing (DISC'98), Andros, Greece, Sept. 1998.

....needed. The broadcast used to send update transactions to all sites needs total order semantics. This order determines the serialization order of the transactions. We use an aggressive version, optimistic delivery total ordered broadcast [13] of the optimistic total ordered broadcast presented in [18]. Optimistic delivery in total ordered broadcast takes advantage of the fact that in a local area network, messages are often spontaneously totally ordered. This optimistic delivery broadcast is defined by three primitives. To broadcast(m) broadcasts the message m to all the sites in the system. ....

F. Pedone and A. Schiper. Optimistic Atomic Broadcast. In Proc. of DISC, 1998.

....The broadcast used to send update transactions to all sites needs total order semantics. This order determines the serialization order of the transactions. We use an aggressive version, optimistic delivery total ordered broadcast [KPAS99] of the optimistic total ordered broadcast presented in [PS98] Optimistic delivery has been applied successfully to total order multicast in the context of replication [PJKA00] and to uniform multicast in the context of non blocking atomic commitment [JPAA01] Optimistic delivery in total ordered multicast takes advantage of the fact that in a local area ....

F. Pedone and A. Schiper. Optimistic Atomic Broadcast. In S. Kutten, editor, Proc. of 12th Distributed Computing Conference, volume LNCS 1499, pages 318332. Springer, September 1998.

....eager replication protocols that guarantee consistency and increase fault tolerance. Such results are especially suitable for clusters of computers and large collections of shared nothing databases. Although some initial optimizations have been suggested [KPAS99a] based on optimistic techniques [PS98], existing protocols have still two major drawbacks. One is the amount of 2 redundant work performed at all sites. The other is the high abort rates created when consistency is enforced. In this paper, we address these two issues. First, we present a protocol that minimizes the amount of ....

....system [BR93] where all group members perceive membership (view) changes at the same virtual time, i.e. two sites deliver exactly the same messages before installing a new view. 3 In this paper we use an aggressive version [KPAS99b] of the optimistic total order broadcast presented in [PS98]. Each message corresponds to a transaction. Messages are optimistically delivered as soon as they are received and before the definitive ordering is established. With this the execution of a transaction can overlap with the calculation of the total order. If the initial order is the same as the ....

F. Pedone and A. Schiper. Optimistic Atomic Broadcast. In S. Kutten, editor, Proc. of 12th Distributed Computing Conference, volume LNCS 1499, pages 318--332. Springer, September 1998.

....multicast behind operations that need to be performed anyway. This is accomplished by processing messages in an optimistic manner and hoping that most decisions will be correct although in some cases transactions might need to be aborted. This idea builds upon recent work in optimistic multicast [4], which was proposed to reduce the latency of total ordered multicast. In here, we follow the more aggressive version of optimistic delivery proposed in the context of Postgres R [2] and later used to provide high performance eager replication in clusters [3] In addition, in any atomic ....

F. Pedone and A. Schiper. Optimistic Atomic Broadcast. In Proc. of 12th Distributed Computing Conference, volume LNCS 1499, pages 318--332. Springer, Sept. 1998.

....protocols run very fast if no malicious adversary is at work and all messages are delivered promptly. If a problem is detected (typically because liveness is violated) they may switch into a more secure mode using protocols that guarantee progress. This idea is quite common in the literature [31, 11]. In our Byzantine context, one has to make sure that safety is never violated, though. Acknowledgments This paper re ects the results of many discussions with Klaus Kursawe, Frank Petzold, Jonathan Poritz, Victor Shoup, Michael Waidner; I am grateful for all their contributions. 15 This work ....

F. Pedone and A. Schiper, \Optimistic atomic broadcast," in Proc. 12th International Symposium on Distributed Computing (DISC), 1998.

No context found.

PEDONE, F. AND SCHIPER, A. 1998. Optimistic atomic broadcast. In Proc. 12th Intl. Symp. on Distributed Computing (DISC). Number 1499 in Lecture Notes in Computer Science. Andros, Greece, 318--332.

....to be expensive and to lead to a significant overhead. The same holds for Atomic Broadcast, which was often criticized as being too expensive. In spite of that, designing optimistic Atomic Broadcast algorithms or optimistic active replication techniques was largely ignored until recently [17], even though optimistic algorithms were known since several years in the context of concurrency control [1] and file system replication [12] In the context of active replication, Pedone distinguishes two dimensions of optimism [16] i) Optimism at the level of the Atomic Broadcast algorithm. ....

....[12] In the context of active replication, Pedone distinguishes two dimensions of optimism [16] i) Optimism at the level of the Atomic Broadcast algorithm. ii) Optimism at the level of the treatment of the client request by the replicated service. The Optimistic Atomic Broadcast algorithm [17] is an example of (i) The algorithm makes the optimistic assumption that in a LAN messages are spontaneously received in total order with high probability, which is experimentally confirmed. If this assumption is met, the algorithm delivers messages faster than known Atomic Broadcast algorithms. ....

[Article contains additional citation context not shown here]

F. Pedone and A. Schiper. Optimistic Atomic Broadcast. In 12th. Intl. Symposium on Distributed Computing (DISC'98), pages 318--332. Springer Verlag, LNCS 1499, September 1998.

....frequently, our Generic Broadcast algorithm may become less etficient than an Atomic Broadcast algorithm, while if conflicts are rare, then our Generic Broadcast algorithm leads to smaller costs compared to Atomic Broadcast algorithms. 5 An exception is the Optimistic Atomic Broadcast algorithm [8], which can deliver messages with delivery latency equal to 2 if the spontaneous total order property holds. Propositions 6 and 7 assess the cost of the Generic Broadcast algorithm when messages do not conflict. In order to simplify the analysis of the delivery latency, we concentrate our results ....

F. Pedone and A. Schiper. Optimistic Atomic Broadcast. In Proc. of 12th International Symposium on Distributed Computing, pages 318 332, September 1998.

....to implement high performance, scalable applications on top of group communication primitives. Recently, a new approach has been proposed for atomic broadcast which, based on optimistic assumptions about the communication system, reduces the average delay for message delivery to the application [22]. The protocol takes advantage of the fact that in a LAN, messages normally arrive at the different sites exactly in the same order. Roughly speaking, this protocol considers the order messages arrive at each site as a first optimistic guess, and only if a mismatch of messages is detected, further ....

....27] all of them require some coordination between sites to guarantee that all messages are delivered in the same order at the different sites. However, when network broadcast (e.g. IP multicast) is used, there is a high probability that messages arrive at all sites spontaneously totally ordered [22]. If this happens most of the time, it seems to be a waste of resources to delay the delivery of a message until the sites agree to this same total order. One could, for example, optimistically process the messages as they are received. If a message is processed out of order, one has to pay the ....

[Article contains additional citation context not shown here]

F. Pedone and A. Schiper. Optimistic atomic broadcast. In Proceedings of the 12 th International Symposium on Distributed Computing (DISC'98, formerly WDAG), September 1998.

....to be expensive and to lead to a significant overhead. The same holds for Atomic Broadcast, which was often criticized as being too expensive. In spite of that, designing optimistic Atomic Broadcast algorithms or optimistic active replication techniques was largely ignored until recently [17], even though optimistic algorithms were known since several years in the context of concurrency control [1] and file system replication [12] In the context of active replication, Pedone distinguishes two dimensions of optimism [16] i) Optimism at the level of the Atomic Broadcast algorithm. ....

....[12] In the context of active replication, Pedone distinguishes two dimensions of optimism [16] i) Optimism at the level of the Atomic Broadcast algorithm. ii) Optimism at the level of the treatment of the client request by the replicated service. The Optimistic Atomic Broadcast algorithm [17] is an example of (i) The algorithm makes the optimistic assumption that in a LAN messages are spontaneously received in total order with high probability, which is experimentally confirmed. If this assumption is met, the algorithm delivers messages faster than known Atomic Broadcast algorithms. ....

[Article contains additional citation context not shown here]

F. Pedone and A. Schiper. Optimistic Atomic Broadcast. In 12th. Intl. Symposium on Distributed Computing (DISC'98), pages 318--332. Springer Verlag, LNCS 1499, September 1998.

No context found.

F. Pedone and A. Schiper, "Optimistic atomic broadcast," in Proc. 12th International Symposium on Distributed Computing (DISC), 1998.

No context found.

Andre Schiper Fernando Pedone. Optimistic atomic broadcast. In Proceedings of the 12 th International Symposium on Distributed Computing (DISC'98, formerly WDAG), September 1998.

No context found.

Andre Schiper Fernando Pedone. Optimistic atomic broadcast. In Proceedings of the 12 th International Symposium on Distributed Computing (DISC'98, formerly WDAG), September 1998.

No context found.

F. Pedone and A. Schiper. Optimistic Atomic Broadcast. Proc. of Distributed Computing Conf. (DISC), 1998.

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