D. Dolev, S. Kramer, and D. Malki. Early delivery totally ordered multicast in asynchronous environments. In Int. Symposium on Fault-Tolerant Computing, pages 544-553, Toulouse, France, June 1993. IEEE Comput. Soc.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... detector and, among these, those that provide only non uniform delivery [17, 2] In this setting, is is possible to provide total order delivery in a single communication step: a sequencer based algorithm [17] can deliver the messages from the sequencer in a single step, and a symmetric algorithm [20, 7, 2, 9] can also deliver messages in a single step when all processes are transmitting. Under the same assumptions, uniform delivery can be provided at the cost of one additional communication step, for a total of two communication steps. Unfortunately, if the assumptions do not hold, these algorithms ....

....an algorithm that, in stable periods, is as e#cient as the algorithms that assume a perfect failure detector. Therefore, we have decided to select an existing algorithm in that class to serve as the basis for our new algorithm. Although several algorithms have been described in the literature [4, 6, 7, 17, 19, 2], few were specifically targeted to operate in (geographically) large scale systems. In a large scale network processes tra#c patterns are usually heterogeneous. The same applies to the network links: some processes will be located within the same local area network whereas others will be ....

[Article contains additional citation context not shown here]

D. Dolev, S. Kramer, and D. Malki. Early delivery totally ordered multicast in asynchronous environments. In Digest of Papers, The 23th International Symposium on Fault-Tolerant Computing, pages 544--553, Toulouse, France, June 1993.

.... detector and, among these, those that provide only non uniform delivery [17, 1] In this setting, is is possible to provide total order delivery in a single communication step: a sequencer based algorithm [17] can deliver the messages from the sequencer in a single step, and a symmetric algorithm [20, 6, 1, 9] can also deliver messages in a single step when all processes are transmitting. Under the same assumptions, uniform delivery can be provided at the cost of one additional communication step, for a total of two communication steps. Unfortunately, if the assumptions do not hold, these algorithms ....

....an algorithm that, in stable periods, is as ecient as the algorithms that assume a perfect failure detector. Therefore, we have decided to select an existing algorithm in that class to serve as the basis for our new algorithm. Although several algorithms have been described in the literature [3, 5, 6, 17, 19, 1], few were speci cally targeted to operate in (geographically) large scale systems. In a large scale network processes trac patterns are usually heterogeneous. The same applies to the network links: some processes will be located within the same local area network whereas others will be ....

[Article contains additional citation context not shown here]

D. Dolev, S. Kramer, and D. Malki. Early delivery totally ordered multicast in asynchronous environments. In Digest of Papers, The 23th International Symposium on Fault-Tolerant Computing, pages 544-553, Toulouse, France, June 1993. IEEE.

....Guerraoui and Schiper [GS97a] have shown the impossibility of genuine implementation with unreliable failure detectors even if only one process may crash. This result explains why the known atomic multicast algorithms either ensure only local order atomic multicast [BSS91] either are not genuine [DKM93, AMAC95, EMS95], either require a reliable failure detection [Jia95, GMS91] or a membership service [BSS91, BJ87] By restricting multicast to sets of non intersecting process group, we circumvent this impossibility result. Moreover, this type of atomic multicast seems well adapted in the area of the ....

Danny Dolev, Shlomo Kramer, and Dalia Malki. Early delivery totally ordered multicast in asynchronous environment. In IEEE Proceedings of the 23th International Symp on Fault-tolerant computing (FTCS-23), pages 544--553, June 1993.

....distributed applications. For instance, total delivery order is a requirement for the implementation of replicated statemachines [86] which is a general paradigm for implementing fault tolerant distributed applications. Although several protocols have been described in the literature [4, 16, 13, 26, 36, 50, 51, 55, 62, 63, 71], few were speci cally targeted to operate in (geographically) large scale systems. In a large scale network processes trac patterns are usually heterogeneous. The same applies to the network links: some processes will be located within the same local area network whereas others will be connected ....

D. Dolev, S. Kramer, and D. Malki. Early delivery totally ordered multicast in asynchronous environments. In Digest of Papers, The 23th International Symposium on Fault-Tolerant Computing, pages 544-553. IEEE, 1993. 74

....corresponding implementations that are optimal in terms of resiliency. We also give an interesting application of our results regarding the implementation of atomic broadcast. 1 Introduction Atomic broadcast is a well known building block of fault tolerant distributed applications (e.g. see [7, 4, 9, 8, 10, 3, 2]) Informally, this communication primitive ensures that all messages broadcast are delivered in the same order. In a recent paper, Pedone and Schiper noted that for some applications some messages do not conflict with each other, and hence they can be delivered by different processes in ....

D. Dolev, S. Kramer, and D. Malki. Early delivery totally ordered multicast in asynchronous environment. In IEEE Proceedings of the 23th International Symp on Fault-tolerant computing (FTCS-23), pages 544--553, June 1993.

....group) Guerraoui and Schiper [11] have shown the impossibility of genuine implementation with unreliable failure detectors even if only one process may crash. This result explains why the known atomic multicast algorithms either ensure only local order atomic multicast [3] either are not genuine [6, 1, 7], either require a reliable failure detection [16, 10] or a membership service [3, 2] By restricting multicast to sets of non intersecting process group, we circumvent this impossibility result. Moreover, this type of atomic multicast seems well adapted in the area of the replication in which ....

D. Dolev, S. Kramer, and D. Malki. Early delivery totally ordered multicast in asynchronous environment. In IEEE Proceedings of the 23th International Symp on Fault-tolerant computing (FTCS-23), pages 544--553, June 1993.

.... process wishes to join or leave the group, and no process crashes) These differences appear to make group membership weaker than Consensus, and in fact the first one has been widely cited as a reason why group membership is solvable in asynchronous systems while Consensus is not [RB91, ADKM92, DKM93, DMS94, EMS95] In this paper we prove that this is not so: We define a problem called WGM (for Weak Group Membership) that allows the removal of erroneously suspected processes from the group, and is subsumed by any reasonable definition of group membership, and show that WGM cannot be solved in ....

....may only crash. It is important to note that this impossibility result applies to group membership services that are allowed to remove from the group, and even kill, an arbitrary number of non faulty processes that did not wish to be removed. Thus, contrary to a widespread view [RB91, ADKM92, DKM93, DMS94, EMS95] allowing the removal or killing of processes that are suspected to have crashed is not sufficient to make the primary partition group membership problem solvable. Note that the proof that WGM cannot be solved in asynchronous systems with failures hinges on the following liveness ....

Danny Dolev, Shlomo Kramer, and Dalia Malki. Early delivery totally ordered multicast in asynchronous environment. In Proceedings of the 23th Annual International Symposium on Fault-Tolerant Computing, pages 544--553, Toulouse, June 1993.

....form the causal order from acknowledgments piggybacked on messages. The Total protocol [Melliar Smith et al. 1990; Moser et al. 1993] implemented on top of Trans has the interesting characteristic that it continues to order messages even in the presence of faulty processors. The Toto protocol [Dolev et al. 1993] implemented on top of Lansis has a similar characteristic but, unlike the Total protocol, uses additional messages for voting. All of the other totally ordered multicast protocols discussed here block in the presence of processor faults until a Fig. 1. The Totem multiple ring protocol hierarchy. ....

DOLEV, D., KRAMER, S., AND MALKI, D. 1993. Early delivery totally ordered multicast in asynchronous environments. In Proceedings of the 23rd IEEE International Symposium on Fault-Tolerant Computing (Toulouse, France, June). IEEE Computer Society Press, Los Alamitos, CA, 544--553.

....symmetric, two phase and centralized solutions under this paradigm. Centralized ordering could also be classified as a star geometry, but the central node is typically chosen ad hoc based on some election or tokenpassing scheme among all nodes. 4.1. 1 Symmetric Ordering In symmetric schemes (S) [6, 11, 27], all hosts partake in the ordering process in a decentralized way, analogous to a voting process, using message stability properties. SN disseminate messages reliably to all hosts, which assign a timestamp to each message and place it in a pending buffer; for each message m, participant hosts (SN ....

....decentralized approaches are based on the total order solution by Lamport [19] where timestamps are assigned to messages at broadcast time. The algorithm by Chandra and Toueg [6] executes reliable broadcast in four communication steps with a weak failure detector, and the solution by Dolev et al. [11] is based on a majority consensus protocol. Both schemes have O(n 2 ) message complexity. Newer approaches such as the TO multicast protocol [17] or Scalatom [27] incur message complexity O(r 2 ) Hybrid algorithms using distributed messaging in conjunction with centralized sequencing [26] ....

D. Dolev, S. Kramer, and D. Malki. Early delivery totally ordered multicast in asynchronous environments. In Int. Symposium on Fault-Tolerant Computing, pages 544--553, Toulouse, France, June 1993. IEEE Comput. Soc.

....send messages) These algorithms are also known as symmetric algorithms, as all participants have peer roles. These algorithms are especially efficient when all participants periodically produce messages, with approximately similar rates. There are several known protocols is this category [9, 5]. Another class of algorithms is based on the selection of a special process in the system which is made responsible for establishing the ordering between messages. This process works as a sequencer of all messages and is often called the token site . A number of algorithms based on this ....

D. Dolev and S. Kramerand D. Malki. Early delivery totally ordered multicast in asynchronous environments. In Digest of Papers, The 23th International Symposium on Fault-Tolerant Computing, pages 544--553, Toulouse, France, June 1993. IEEE.

....consistent view of the order in which events (such as message delivery, process failures) have taken place. Design and development of fault tolerant group communication protocols for distributed systems satisfying certain order properties has been therefore an active area of research (e.g. 4][6]12] 17] We present a contribution to this area that makes use of the concept of logical clocks [10] We begin by describing the motivation behind our work and describe the novel features. In section three we present the basic system model and definitions and in the fourth section we develop the ....

....order of message numbers. 6. Comparison with Related Work Psync Consul [15, 17] is one of the best known protocol suite that implements causal and symmetric total delivery protocols; however, it has no support for multiple (overlapping) process groups. The Trans and Transis family of protocols [1, 6, 12] use elegant symmetric solutions for providing total order delivery, but are not quite general purpose, as they rely on network level broadcast communication; further, like Psync Consul, the issue of a process belonging to multiple process groups has not been addressed. ISIS was the first system ....

Dolev, D., Kramer, S. and Malki, D., "Early Delivery Totally Ordered Multicast in Asynchronous Environment", Digest of Papers, FTCS-23, Toulouse, pp. 544-553, June 1993.

....protocol that uses piggybacked ACKs to provide causal ordering of messages and detection of dropped packets. However, both it and the similar Trans [21] and Lansis [15] protocols require that all of the members of the group regularly transmit messages. The Trans protocol and the ToTo protocol [14] implemented on top of Lansis both provide total ordering of messages. These algorithms require that at least a majority of the group members be heard from before a message can be delivered, which causes latency to increase by at least an order of magnitude. For example, for the ToTo protocol to ....

D. Dolev S. Kramer and D. Malki. Early delivery totally ordered multicast in asynchronous environments. In 23rd Annual International Symposium on FaultTolerant Computing (FTCS), pages 544--553, June 1993.

....of this report were published in the proceedings of the 16th International Conference on Distributed Computing Systems, Hong Kong, May, 1996. which is a general paradigm for implementing fault tolerant distributed applications. Although several protocols have been described in the literature [2, 3, 4, 5, 7, 11, 12, 13, 15, 16, 18], few were specifically targeted to operate in (geographically) large scale systems. In a large scale network processes traffic patterns are usually heterogeneous. The same applies to the network links: some processes will be located within the same local area network whereas others will be ....

....in Section 4 for static topologies. Section 5 presents the switching protocol and Section 6 shows how it is used with dynamic topologies. Concluding remarks appear in Section 7. 2 Related work Among the several algorithms for implementing total ordering, the token site [5, 12] and symmetric [18, 7] are the most used approaches 1 .Both methods have ad 1 Other solutions exist but will not be considered in this paper due to their performance limitation in large scale environments. The class of algorithms known as Replica Generated Identifiers [3, 22] computes total order in two phases. ....

[Article contains additional citation context not shown here]

D. Dolev, S. Kramer, and D. Malki. Early delivery totally ordered multicast in asynchronous environments. In Digest of Papers, The 23th International Symposium on Fault-Tolerant Computing, pages 544--553. IEEE, 1993.

.... Further work on the Trans protocol at the Hebrew University of Jerusalem in Israel has led Dolev et al. to develop the Transis communication subsystem [3] 4] In Transis, a partial order is derived by the Lansis protocol, and this partial order is converted to a total order by the Toto protocol [14]. Similar to Trans from which it was derived, Lansis provides a partial order on messages using piggybacked acknowledgments, but in contrast, these acknowledgments are temporarily withheld pending deliverability of the message to the higher layers of their protocol stack. Thus, the acknowledgments ....

D. Dolev, S. Kramer, and D. Malki, "Early Delivery Totally Ordered Multicast in Asynchronous Environments." In 23rd Annual International Symposium on Fault-Tolerant Computing, pp. 544-553, Toulouse, France, June 1993.

....in which (future) view its multicast request will be serviced. Finally, note that view synchrony as implemented in Relacs guarantees nothing about the relative order of messages delivered during a given view. Applications that require ordering guarantees such as uniform [20] causal [8] or total [13] will have to rely on layers built on top of Relacs. 4 The Application Interface Applications that require Relacs services invoke them through a small set of library functions. The proposed interface is an attempt to maximize flexibility while minimizing complexity. The following is an informal ....

D. Dolev, S. Kramer and D. Malki. Early Delivery Totally Ordered Multicast in Asynchronous Environments. In Proc. 23nd Annual International Symposium on Fault-Tolerant Computing, pages 544--553, June 1993.

....protocol that uses piggybacked ACKs to provide causal ordering of messages and detection of dropped packets. However, both it and the similar Trans [25] and Lansis [20] protocols require that all of the members of the group regularly transmit messages. The Trans protocol and the ToTo protocol [19] implemented on top of Lansis both provide total ordering of messages. These algorithms require that at least a majority of the group members be heard from before a message can be delivered, which causes latency to increase by at least an order of magnitude. For example, for the ToTo protocol to ....

....partitions. To solve this problem, Rmp offers four levels of fault tolerant guarantees: atomic delivery within partitions, K resilient atomic between partitions, agreed delivery between partitions, and safe delivery between partitions. The exact semantics of agreed and safe delivery are defined in [19]. All of these guarantees rely on a method of failure detection based on timeouts. If communication to one or more group members fails for an extended period of time (say 15 30 seconds) the Rmp failure membership algorithm will remove them from the group. If this is due to a temporary ....

D. Dolev S. Kramer and D. Malki. Early delivery totally ordered multicast in asynchronous environments. In 23rd Annual International Symposium on Fault-Tolerant Computing (FTCS), pages 544--553, June 1993.

....m.s is currently in the process s view for m.g . It is possible to enhance Newtop to provide a stronger atomicity property, known as uniform atomicity, which ensures that if any process, whether crashed or not, is delivered m , then all the functioning processes are delivered m (in Transis papers [Amir92a, Amir92b, Dolev93], atomic and uniform 9 atomic deliveries have been termed agreed and safe respectively) To save space, we will not discuss the provision of uniform atomicity in this paper. Transport Layer Membership Service Logical Clock System View Installation Total Order Delivery Atomic Delivery Fig. 5: ....

Dolev, D., Kramer, S. and Malki, D., "Early Delivery Totally Ordered Multicast in Asynchronous Environment", Digest of Papers, FTCS-23, Toulouse, pp. 544-553, June 1993.

....distributed applications. For instance, total delivery order is a requirement for the implementation of replicated statemachines [23] which is a general paradigm for implement of fault tolerant distributed applications. Although several protocols have been described in the literature [5, 3, 19, 17, 13, 12, 4, 16, 7, 1, 11], few were specifically targeted to operate in large scale systems. Among the several algorithms to implement total ordering, the token site [5, 12] and symmetric [19, 7] are the most used approaches. In the token site approach, one (or more) sites are responsible for ordering the messages on ....

....distributed applications. Although several protocols have been described in the literature [5, 3, 19, 17, 13, 12, 4, 16, 7, 1, 11] few were specifically targeted to operate in large scale systems. Among the several algorithms to implement total ordering, the token site [5, 12] and symmetric [19, 7] are the most used approaches. In the token site approach, one (or more) sites are responsible for ordering the messages on behalf of the other processes in the system. In the symmetric approach, ordering is established by all processes in a decentralized way, using information about message ....

[Article contains additional citation context not shown here]

D. Dolev, S. Kramer, and D. Malki. Early delivery totally ordered multicast in asynchronous environments. In Digest of Papers, The 23th International Symposium on Fault-Tolerant Computing, pages 544--553, Toulouse, France, June 1993. IEEE.

No context found.

D. Dolev, S. Kramer, and D. Malki. Early delivery totally ordered multicast in asynchronous environments. In Int. Symposium on Fault-Tolerant Computing, pages 544-553, Toulouse, France, June 1993. IEEE Comput. Soc.

No context found.

D. Dolev, S. Kramer and D. Malki, Early delivery totally ordered multicast in asynchronous environments, Proc. FTCS, Toulouse, France (June 1993), 544--553.

No context found.

DOLEV, D., KRAMER, S., AND MALKI, D. 1993. Early delivery totally ordered multicast in asynchronous environments. In Proc. 23rd Intl. Symp. on Fault-Tolerant Computing (FTCS-23). Toulouse, France, 544--553.

No context found.

D. Dolev, S. Kramer, D. Malki. "Early Delivery Totally Ordered Multicast in Asynchronous Environments." 23rd Annual International Symposium on Fault-Tolerant Computing (FTCS). (Toulouse, France, June, 1993). pp. 544-553.

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