Esti Yeger Lotem, Idit Keidar, and Danny Dolev. Dynamic voting for consistent primary components. In 16th Symp. on Princ. of Distr. Comp. (PODC), pages 63--71, Santa Barbara, CA, USA, August 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....to operate collectively as a group, using the services to multicast messages to group members. Examples of group communication services are found in Isis [5] Transis [9] Totem [19] Newtop [12] Relacs [2] Horus [21] and Ensemble [4] Examples of recent work dealing with primary groups are [7, 17]. An example of an application using a group communication service for load balancing is by Fekete, Khazan and Lynch [13] To evaluate the effectiveness of partitionable group communication services, Sussman and Marzullo [24] proposed a measure (cushion) precipitated by a simple partition aware ....

E. Y. Lotem, I. Keidar, and Danny Dolev, "Dynamic Voting for Consistent Primary Components," Proc. of the 16th Annual ACM Symp. on Principles of Distributed Computing, pp. 63-71, 1997.

....in order to permit enough information ow between successive primary views to achieve coherence. For example, each primary view might have to contain at least a majority of the processes in the previous primary view. Several dynamic voting schemes have been developed to de ne primaries adaptively [12, 15, 21, 26, 33]. In particular, Lotem, Keidar, and Dolev [26] have described an implementation of a group membership service that yields only primary views, according to a dynamic notion of primary. An interesting feature of their work is that it points out various subtleties of implementing such a membership ....

....views to achieve coherence. For example, each primary view might have to contain at least a majority of the processes in the previous primary view. Several dynamic voting schemes have been developed to de ne primaries adaptively [12, 15, 21, 26, 33] In particular, Lotem, Keidar, and Dolev [26] have described an implementation of a group membership service that yields only primary views, according to a dynamic notion of primary. An interesting feature of their work is that it points out various subtleties of implementing such a membership service in a distributed manner subtleties ....

[Article contains additional citation context not shown here]

E. Lotem, I Keidar and D. Dolev, \Dynamic voting for consistent primary components", in Proc. of the 16 Annual ACM Symposium on Principles of Distributed Computing, Santa Barbara, CA, August 1997, pp. 63-71.

....Paths Quorum System In this section we suggest a quorum system that operates in the dynamic model, where processors may join and leave. Previous constructions of dynamic quorums focused on designing algorithms that allowed a group of processors to form a new quorum in a consistent way ( 8] [9], 21] The quorums themselves are usually assumed to be weighted voting. We focus on the combinatorial properties of dynamic quorums. Our goal is to design dynamic quorums that enjoy low load, high availability, low probe complexity and that scale gracefully in respect to these parameters. The ....

Esti Yeger Lotem, Idit Keidar, and Danny Dolev. Dynamic voting for consistent primary components. In Symposium on Principles of Distributed Computing, pages 63--71, 1997.

....membership of the group. The view includes a list of the processors that are members of the group. Views can change and may become di erent at di erent processors. There is a substantial amount of research dealing with speci cation and implementation of GCSs and grouporiented applications, e.g. [1, 2, 13, 7, 10, 16, 19, 21], and veri cation of GCSs and group oriented systems, e.g. 5, 15, 11] When developing group oriented, and especially partition aware applications, it is also important to understand the e ectiveness of group communication services [23] and the eciency bene ts that can be expected when using ....

....located in a fault prone network to operate collectively as a group, using the services to multicast messages to group members. Examples of GCS include Isis [2] Transis [7] Totem [19] Newtop [10] Relacs [1] Horus [21] and Ensemble [13] Examples of recent work dealing with primary groups are [5, 16]. An example of an application using a GCS for load balancing is by Fekete, Khazan and Lynch [15] To evaluate the e ectiveness of partitionable GCSs, Sussman and Marzulo [23] proposed the measure (cushion) precipitated by a simple partition aware application. Our de nition of work follows that ....

E. Y. Lotem, I. Keidar, and Danny Dolev, \Dynamic Voting for Consistent Primary Components," Proc. of the 16th Annual ACM Symp. on Principles of Distributed Computing, pp. 63-71, 1997.

....y changes not only to the set of quorums, but also to the very intersection property that de nes those quorums. Second, our clients (which we assume to be potentially numerous and transient) do not communicate between themselves and cannot inform one another of changes. Our work di ers from [BB98, LKD97] as well, in that our protocols can take hints on the possible level of failures in the system in order to adjust the resilience threshold, rather than relying on speci c detection of server failures. Some work in Byzantine agreement considers dynamic threshold changes during the run of the ....

E. Lotem, I. Keidar, and D. Dolev. Dynamic voting for consistent primary components. In Proceedings of the 16th ACM Symposium on Principles of Distributed Computing (PODC), August 1997.

....One or more client processes link to each Deno server, which communicate through pair wise information propagation. The servers are not necessarily ever fully connected. Deno uses bounded voting [27] in which the total currency in the system is fixed at a system wide value. Voting schemes [2, 4, 20, 25, 31, 33, 41] allow a quorum of all replicas to commit an update. Quorums are distinct sets that can each commit an update, provided that all replicas of the quorum agree. Serialization of updates is accomplished by requiring that any two potential quorums must share at least one replica. Hence, ....

E. Y. Lotem, I. Keidar, and D. Dolev, Dynamic Voting for Consistent Primary Components, in 17th ACM Symposium on Principles of Distributed Computing, June 1997.

....also requires that all but at most one of the concurrent sub groups execute Ei, and the one sub group (if any) that does not execute Ei has the quorum. Managing the merging of sub groups in quorum based environments needs to be aware of certain subtle problems, which are outlined and solved in [Lotem97, Ezhilchelvan99] Primary Partition Membership Service While partitionable membership service can be implemented in a non blocking manner and cannot be specified ruling out trivial implementations, primary partition membership service can be specified precisely but cannot be implemented in a ....

E Y Lotem, I Keidar and D Dolev. Dynamic Voting for Consistent Primary Components. Proceedings of ACM Symposium on Principles of Distributed Computing (PODC), pp. 63-71, 1997.

....that enable processors to autonomously adapt their quorum data structures. Examples of coordinated quorum adaptation are given in [4,12,13] These protocols identify quorums implicitly by gathering votes whose assignment can be adjusted dynamically. The coordinated quorum adaptation approaches in [16,18] enable changes to distributed data structures that explicitly identify quorums. In each of these coordinated approaches, multi phase updates ensure proper system wide ordering of communication with quorums in older and newer installed quorum assignments. Permission to install a new quorum ....

Lotem, E., Keidar, I., and Dolev, D., "Dynamic Voting for Consistent Primary Components," Proc. 16th Symp. on Principles of Distributed Computing, Santa Barbara, California, Aug. 1997.

....in the system. This mechanism is generally based on requiring a majority (or quorum) of processes to change the membership. More complex implementations of dynamic voting protocols that de ne quorums adaptively, depending on a dynamically changing set of processes, have also been proposed [27] [34] and rely on the existence of underlying group membership algorithms. Split brain behavior is immediately solved in a primary partition model since, in the event of a partition, only one set of processes in the system is allowed to remain operational. Any group membership algorithm can be easily ....

....only one set of processes in the system is allowed to remain operational. Any group membership algorithm can be easily extended with a primary component mechanism, based simply on a majority requirement or on more complex algorithms. For example, the dynamic quorums algorithm of Lotem et al. [34] can be implemented on top of our group membership algorithm. The rationale underlying the primary partition model is that it is impossible to guarantee consistency if both sides of a partitioning failure are permitted to remain live. The drawback of this approach is that in large networks, ....

E. Y. Lotem, I. Keidar and D. Dolev. \Dynamic Voting for Consistent Primary Components ". Proceedings of the Sixteenth ACM Symposium on Principles of Distributed Computing, ACM Press, pp. 63-71, August 1997.

....processes to be logically connected and to have a leader. This approach has been used for algorithms implemented on systems described in [1] 9] Articulate dynamic voting protocols that define quorums adaptively, depending on a dynamically changing set of processes, have also been proposed [8] [12]. Such protocols are always based on underlying group mechanisms, depend from their liveness properties and may be implemented on top of our algorithm. Finally, if our protocol is run on a partitionable network, with no mechanism for maintaining a primary component, in the case of a partition the ....

E. Y. Lotem, I. Keidar and D. Dolev. "Dynamic Voting for Consistent Primary Components". Proceedings of the Sixteenth ACM Symposium on Principles of Distributed Computing, ACM Press, pp. 63-71, August 1997.

....changes not only to the set of quorums, but also to the very intersection property that defines those quorums. Second, our clients (which we assume to be potentially numerous and transient) do not communicate between themselves and cannot inform one another of changes. Our work differs from [4, 9] as well, in that our protocols can take hints on the possible level of failures in the system in order to adjust the resilience threshold, rather than rely on specific detection of server failures. Some work in Byzantine agreement considers dynamic threshold changes during the run of the ....

E. Lotem, I. Keidar, and D. Dolev. Dynamic voting for consistent primary components. In Proceedings of the 16th ACM Symposium on Principles of Distributed Computing (PODC), August 1997.

....changes not only to the set of quorums, but also to the very intersection property that defines those quorums; second, our clients (which we assume to be potentially numerous and transient) do not communicate between themselves and cannot inform one another of changes. Our work differs from [4, 9] as well, in that our protocols can take hints on the possible level of failures in the system in order to adjust the resilience threshold, rather than rely on specific detection of server failures. There is some work in Byzantine agreement that considers dynamic changes during the run of the ....

E. Lotem, I. Keidar, and D. Dolev. Dynamic voting for consistent primary components. In Proceedings of the 16th ACM Symposium on Principles of Distributed Computing (PODC), August 1997.

....consistently. Traditionally, a primary view was defined as one containing a majority of all possible sites, but other, dynamic, definitions are possible, based on intersection properties between successive primary views. Several methods have been developed to define primary views adaptively, e.g. [6, 13, 14, 17, 27, 35, 41, 43, 47]. Producing good specifications for view oriented group communication services is difficult, because these services can be complicated, and because different such services provide different guarantees about safety, performance, and fault tolerance. Examples of specifications for group membership ....

....about safety, performance, and fault tolerance. Examples of specifications for group membership services and view oriented group communication services appear in [4, 5, 10, 12, 16, 21, 22, 25, 26, 36, 39, 42, 44] Extending these definitions to specify dynamic primary views was the focus of [14, 47]. 3 In this paper we combine the notion of dynamic primary view with that of a quorum system, and call the result a configuration. We integrate this with a group communication service, resulting in a dynamic primary configuration group communication service. The main difficulty in combining ....

[Article contains additional citation context not shown here]

E. Yeger Lotem, I. Keidar, and D. Dolev. Dynamic voting for consistent primary components. In Proceedings of the Sixteenth Annual ACM Symposium on Principles of Distributed Computing, pages 63--71, 1997.

....almost simultaneously, or any other transient turbulence in the network. When interrupted, dynamic voting algorithms differ in the length of their blocking period: some of the suggested algorithms (e.g. 7, 1] may block until all the members of the last primary become reconnected; others (e.g. [9, 12, 5, 8]) can make progress whenever a majority of the last primary becomes reconnected. Algorithms also differ in how long it takes them to resolve the outcome of interrupted attempts to form a primary component, and in their ability or inability to pipeline multiple such attempts. We focus on the ....

....the last primary becomes reconnected. Algorithms also differ in how long it takes them to resolve the outcome of interrupted attempts to form a primary component, and in their ability or inability to pipeline multiple such attempts. We focus on the dynamic voting algorithm of Yeger Lotem et al. [12], hereafter called YKD. We compare its availability with that of four variations on it one which removes some memory saving optimizations, a second variation due to De Prisco et al. 5] a third variation which is similar (although not identical) to the dynamic voting algorithms suggested in ....

[Article contains additional citation context not shown here]

E. Yeger Lotem, I. Keidar, and D. Dolev. Dynamic voting for consistent primary components. In 16th ACM Symposium on Principles of Distributed Computing (PODC), pages 63--71, August 1997. 43

....to service, or any other transient turbulence in the network. When interrupted, dynamic voting algorithms differ in their resilience: some of the suggested algorithms (e.g. 8, 1] may block until all the members of the last attempt to form a primary component become reconnected. Others (e.g. [12, 16, 5, 11]) can make progress whenever a majority of the members of the last attempt to form a primary component are present. Algorithms also differ in how long it takes them to resolve the outcome of interrupted attempts to form a primary component, and in their ability or inability to pipeline multiple ....

....are present. Algorithms also differ in how long it takes them to resolve the outcome of interrupted attempts to form a primary component, and in their ability or inability to pipeline multiple such attempts. We study four dynamic voting algorithms: The first algorithm is due to Yeger Lotem et al. [16]. The second is a variation on the first, due to De Prisco et al. 5] The third is based on the idea of two phase commit, similar to the algorithms suggested in [8, 1] The fourth resembles three phase commit, similar to ideas presented in [11, 12] As a control, we also compare the algorithm ....

[Article contains additional citation context not shown here]

E. Yeger Lotem, I. Keidar, and D. Dolev. Dynamic voting for consistent primary components. In 16th ACM Symposium on Principles of Distributed Computing (PODC), pages 63--71, August 1997. 8

....or any other transient turbulence in the network. When interrupted, dynamic voting algorithms di er in their resilience: some of the suggested algorithms (e.g. JM90, Ami95] may block until all the members of the last attempt to form a primary component become reconnected. Others (e.g. [MS95, YLKD97, DPFLS98, Lam98]) can make progress whenever a majority of the members of the last attempt to form a primary component are present. Algorithms also di er in how long it takes them to resolve the outcome of interrupted attempts to form a primary component, and in their ability or inability to pipeline multiple ....

....are present. Algorithms also di er in how long it takes them to resolve the outcome of interrupted attempts to form a primary component, and in their ability or inability to pipeline multiple such attempts. We study four dynamic voting algorithms: The rst algorithm is due to Yeger Lotem et al. [YLKD97]. The second is a variation on the rst, due to De Prisco et al. DPFLS98] The third is based on the idea of two phase commit, similarly to the algorithms suggested in [JM90, Ami95] The fourth resembles three phase commit, similar to ideas presented in [Lam98, MS95] As a control, we also ....

[Article contains additional citation context not shown here]

E. Yeger Lotem, I. Keidar, and D. Dolev. Dynamic voting for consistent primary components. In 16th ACM Symposium on Principles of Distributed Computing (PODC), pages 63-71, August 1997. 11

....concept. A quorum system is a collection of sets (quorums) such that any two sets intersect. Using such a quorum system, a network component can become the primary one if it contains a quorum. The concept of quorums may be further generalized to allow dynamic adjustment of the quorum system. In Yeger Lotem et al. 1997, we present a dynamic voting protocol for maintaining the primary component in the system; this protocol may be used in conjunction with COReL. 1.1 THE PROBLEM The Atomic Broadcast problem defined in Hadzilacos and Toueg, 1993 deals with consistent message ordering. Informally, Atomic Broadcast ....

....defines quorums adaptively. When a partition occurs, a majority of the previous quorum may chosen as the new primary component. Thus, a primary component must not necessarily a majority of the processors. Dynamic voting may introduce inconsistencies, and therefore should be handled carefully. In Yeger Lotem et al. 1997 we suggest an algorithm for consistently maintaining a primary component using dynamic voting. This algorithm may be easily incorporated into COReL, optimizing it for highly unreliable networks. In Keidar, 1994 we prove the correctness of the COReL algorithm. Acknowledgments The authors are ....

Yeger Lotem, E., Keidar, I., and Dolev, D. (1997). Dynamic voting for consistent primary components. In 16th ACM Symposium on Principles of Distributed Computing (PODC), pages 63--71.

....FLS97, BBD96, CS95, JFR93] HS95] first presents the specifications of a primary partition membership service and then shows how to extend them to specifications of a partitionable one. Some application semantics require that only one partition remains active following a partition [HS95, YLKD97] For such applications some partitionable GCSs provide a primary partition notification to the members of exactly one partition [FvR95, HS95] 6.2.3 The initial view All send and delivery events always occur in some view. FLS97] assumes that processes start up with some default view. ....

....of views that include p. The latter circumstance renders the membership protocol more complicated and less efficient, because it demands at least two rounds of communication. Thus, only a few GCSs (e.g. Transis [DMS96] and Relacs [BDM97] support this requirement, even in the absence of crashes. YLKD97] is also concerned with this requirement implementation. 6.2.6 Non triviality of membership Preciseness (Requirement 5.1.5) is one of the most fundamental properties of a membership service. The main goal of a membership service is to reflect the actual connectivity situation of the physical ....

E. Yeger Lotem, I. Keidar, and D. Dolev. Dynamic Voting for Consistent Primary Components. In Annual ACM Symp. on Principles of Distributed Computing, number 16, August 1997. To appear. 78

No context found.

Esti Yeger Lotem, Idit Keidar, and Danny Dolev. Dynamic voting for consistent primary components. In 16th Symp. on Princ. of Distr. Comp. (PODC), pages 63--71, Santa Barbara, CA, USA, August 1997.

No context found.

Esti Yeger Lotem, Idit Keidar, and Danny Dolev. Dynamic voting for consistent primary components. In Symposium on Principles of Distributed Computing (PODC), pages 63--71, 1997.

No context found.

E. Yeger Lotem, I. Keidar, and D. Dolev. Dynamic voting for consistent primary components. In Proceedings of the Sixteenth Annual ACM Symposium on Principles of Distributed Computing, pages 63-71, 1997.

No context found.

E. Lotem, I. Keidar, and D. Dolev. Dynamic voting for consistent primary components. In Proceedings of the 16th ACM Symposium on Principles of Distributed Computing (PODC), August 1997.

No context found.

E.Y. Lotem, I. Keidar, and D. Dolev, "Dynamic voting for consistent primary components," in 16th ACM Symposium on Principles of Distributed Computing, Santa Barbara, CA, Aug. 1997, pp. 63--71.

No context found.

E. Lotem, I. Keidar, and D. Dolev, "Dynamic Voting for Consistent Primary Components", in Proc. ACM Symp. on Principles of Distributed Systems, 1997.

No context found.

E. Lotem, I. Keidar, and D. Dolev, \Dynamic Voting for Consistent Primary Components", in Proc. 16 ACM Symp. on Principles of Distributed Computing, pp. 63-71, 1997.

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