R. Ladin, B. Liskov, and L. Shira, "Lazy Replication: Exploiting the Semantics of Distributed Services," Proc. 1st Workshop on Replicated Data, Houston, TX, Nov. 1990, pp. 31--34.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....isolated clients. Improving performance by providing weaker consistency may lead to more complicated semantics. While in practice, replicated systems are often incompletely or ambiguously speci ed, it remains important to provide formal consistency guarantees. Ladin, Liskov, Shrira, and Ghemawat [13] de ne a replicated data service. They specify general conditions for such a service, and present an algorithm based on lazy replication, in which operations received by each replica are gossiped in the background. Building on the work of Ladin et al. Fekete et al. 7] developed a exible ....

....(given by prev sets) to be preserved. The bene t of such relaxation is improved system performance. At the same time, the system is assured of reaching a globally consistent state, as speci ed and proven in [7] This is a property of ESDSImpl that di erentiates it from the implementation in [13], in which additional measures must be taken by an application designer to ensure that replica states do not diverge irrevocably. We have added several optimizations to ESDSAlg, producing an optimized abstract algorithm ESDSOptAlg and a corresponding ESDSOptImpl implementation. ESDSOptImpl xes ....

R. Ladin, B. Liskov, L. Shrira, and S. Ghemawat. Lazy replication: Exploiting the semantics of distributed services. ACM Transactions on Computer Systems, 10(4):360391, Nov. 1992.

....that have used weak consistency are Internet news, air traffic control, airline reservation and stock exchanges. Grapevine [SBN84] and the Global Name Service [Lam86] were among the first systems to use weak consistency. Other weak consistency protocols were presented in [DGH 87, DGP90, Gol92, LLS90, QP93, WB84] These protocols are useful and interesting; however, they require a node to communicate with all other nodes. This would be acceptable for small networks with a few replicas, but it will incur a large communication overhead for wide area networks like the Internet. This paper ....

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: exploiting the semantics of distributed services. In Proceedings of the 9th ACM symposium on Principles of Distributed Computing, pages 43--57, Quebec City, CA, August 1990.

....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 ....

....token holder) Thus, token based approaches are ine# cient in face of large network delays. In the symmetric approach, ordering is established by all processes in a decentralized way, using information about message stability. This approach usually relies on logical clocks [20] or vector clocks [4, 24, 19]: messages are delivered according to their partial order and concurrent messages are totally ordered using some deterministic algorithm. Symmetric algorithms have the potential for providing low latency in message delivery when all processes are producing messages. In fact, using a technique ....

R. Ladin, B. Liskov, L. Shrira, and S. Ghemawat. Lazy replication: Exploiting the semantics of distributed services. In Proc. of the Ninth Annual ACM Symposium of Principles of Distributed Computing, pages 43--57, 1990.

....operations in a partial order that gravitates over time towards a total order. It provides clear and unambiguous guarantees about the immediate and long term behavior of the system. To demonstrate its utility, we present an algorithm, based on one of Ladin, Liskov, Shrira, and Ghemawat [15], that implements this specification. Our algorithm exports the interface of the abstract service, and generalizes their algorithm by allowing general operations and greater flexibility in specifying consistency requirements. This work was supported by ARPA contract F19628 95 C 0118, ....

....of implementations. In practice, replicated systems are often incompletely or ambiguously specified. 1. 2 Background for our Work: Lazy Replication As it is important that our specification be applicable for real systems, we build heavily on the work of Ladin, Liskov, Shrira, and Ghemawat [15] on highly available replicated data services. They specify general conditions for such a service, and present an algorithm based on lazy replication, in which operations received by each replica are gossiped in the background. Responses to operations may be out of date, not reflecting the effects ....

[Article contains additional citation context not shown here]

R. Ladin, B. Liskov, L. Shrira, and S. Ghemawat. Lazy replication: Exploiting the semantics of distributed services. ACM Transactions on Computer Systems, 10(4):360--391, Nov. 1992.

....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 ....

....the token holder) Thus, token based approaches are inecient in face of large network delays. In the symmetric approach, ordering is established by all processes in a decentralized way, using information about message stability. This approach usually relies on logical clocks [20] or vector clocks [3, 24, 19]: messages are delivered according to their partial order and concurrent messages are totally ordered using some deterministic algorithm. Symmetric algorithms have the potential for providing low latency in message delivery when all 7 RTO1 Agreement: Consider RTO broadcast(m) If a correct ....

R. Ladin, B. Liskov, L. Shrira, and S. Ghemawat. Lazy replication: Exploiting the semantics of distributed services. In Proceedings of the Ninth Annual ACM Symposium of Principles of Distributed Computing, pages 43-57, 1990.

....a probabilistic protocol results in an application programming model which, unlike ours, differs significantly from conventional view synchrony. Although message semantics is here used to relax reliability, it has often been used for relaxing the ordering of messages. For instance lazy replication [17] relies on message semantics to relax causal order. Generic broadcast [19] is a relaxation of total order based on message semantics captured as a binary relation. The work on Optimistic Virtual Synchrony [25] also uses semantic information to alleviate the cost of view changes but, unlike our ....

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: Exploiting the semantics of distributed services. ACM SIGOPS Operating Systems Review, 25(1):49--54, January 1991.

....which the channel may deliver the multicast messages. Many existing frameworks provide support for FIFO, causal and total ordering semantics [BvR94, MMSA 96, ADKM92, KS98] Application semantics has been exploited in group communication frameworks to improve performance [MG97, Sch88, JSM97, LLS90] For example, replicas of a database can be maintained consistent by delivering the operations to all replicas in the same order. However, in some cases, the application entities may be able to accept operations in a sequence satisfying a weaker ordering requirement and still maintain ....

....G; i) then deliver(m1; G; j) deliver(m2; G; j) Several efficient implementations of the FIFO, causal and total ordering semantics have been proposed. Proposals have been made to weaken the causal and total ordering semantics by taking application semantics into account [Sch88, KS98, JSM97, LLS90] In the following, we discuss some examples to motivate the need for additional application specific information: ffl In active replication, the replicas of an object form a group and all operations are broadcast in total order to the group [Sch90, SAA98, FG00] Consider an application ....

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: Exploiting the semantics of distributed services. In Proceedings of the ACM Symposium on Principles of Distributed Computing, 1990.

....is called a server. For fault tolerance, the ambassador may be replicated. In this case, two or more ambassadors will serve remote invocations, and each ambassador will keep a local image of the object. Clients will interact with the group of ambassadors, through replicated remote invocations 1 [6,11] (see figure 3) 1 When objects are accessed exclusively through remote invo Ambassador Group Remote Client Group (1 element only) Data Data Node C Node D Node B Node A C1 C2 CR2 CR1 CR2 CR1 GRIP Matching Ambassador Client Client Repres. Node Data Copy Data CR1 C1 Figure ....

....invocations and local copy access. To our knowledge, the work in this area is very limited. However, each of these methods have been intensively studied in separate. Remote invocations (and remote function calls) to replicated objects has been supported by several systems as troupes [6] ARGUS [11], or DELTA 4 [15] More recent research projects include the ELECTRA [12] object oriented toolkit that also provides abstractions for remote method calling or the Object Groups [10] prototype that uses the ISIS system. However, most of these systems preclude local access to replicas and provide ....

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: Exploiting the semantics of distributed services. In Proceedings of the Workshop on the Management of Replicated Data, pages 31--34, Houston - USA, November 1990. IEEE.

....requirements. This is done by selectively relaxing reliability based on a message obsolescence relation dictated by the application. Although for different purposes, application semantics has been used before to optimize group communication protocols. For instance, to relax causal order [8], total order [9, 10] and ordering of message deliveries with view changes [17] We have presented the definition of a Semantically View Synchronous Multicast 12 primitive and have shown how it can be used in strongly consistent replication using a modified primary backup protocol. The practical ....

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: Exploiting the semantics of distributed services. ACM SIGOPS Operating Systems Review, 25(1):49--54, January 1991.

....that is, in a blocking, request con rmation manner, usually performed through RPC [14] or in open loop, that is, in an unblocking manner, usually performed through group communication, pioneered by Cheriton [27] Birman [16] and Cristian [28] and followed by many others. From previous works [17, 55], it is known to be dicult to scale service access when clients need to be strongly coupled (for instance, when all messages need to be ordered) However, there are a number of services, especially in large scale systems, where the coupling among clients, and between a client and the server, can ....

....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 ....

[Article contains additional citation context not shown here]

R. Ladin, B. Liskov, L. Shrira, and S. Ghemawat. Lazy replication: Exploiting the semantics of distributed services. MIT/LCS/TR 84, MIT Laboratory for Computer Science, 1990.

....is that by being a low level, low cost mechanism, all interprocess communication may be reasonably routed through it, thereby preserving causality across all communication. Some higher level research has focused on the problem of constructing replicated services. The lazy replication method of [LLSG90] is intended for the construction of replicated services supporting query and update operations. This approach includes protocols satisfying the causality, order, and atomicity properties. 96 8.2 The Meta Implementation The ISIS toolkit, by providing virtually synchronous behavior, greatly ....

Rivka Ladin, Barbara Liskov, Liuba Shrira, and Sanjay Ghemawat. Lazy replication: Exploiting the semantics of distributed services. Technical Report MIT/LCS/TR-484, M.I.T. Laboratory for Computer Science, July 1990.

....through RPC [Birrell Nelson 1984] or in open loop, usually performed through group communication, pioneered by Cheriton [Cheriton Zwaenepoel 1985] Birman [Birman Joseph 1987] and Cristian [Cristian et al. 1985] and followed by many others. From previous works [Birman et al. 1991] [Ladin et al. 1990], it is known to be difficult to scale service access when clients need to be strongly coupled (for instance, when all messages need to be ordered) However, there are a number of services, especially in large scale systems, where the coupling among clients, and between a client and the server, ....

R. Ladin, B. Liskov, L. Shrira and S. Ghemawat, Lazy Replication: Exploiting the Semantics of Distributed Services, Technical Report NMIT/LCS/TR-84, MIT Laboratory for Computer Science, 1990.

....and network partition. Another approach to improve the performance in a replicated database is to relax the serializability requirement. Works on quasi copies ( Alonso et al. 1988; Alonso et al. 1990; Barbara and Garcia Molina 1990] lazy replication ( Ladin et al. 1988; Ladin et al. 1992; Ladin et al. 1990]) and bounded ignorance ( Krishnakumar and Bernstein 1991] fall in this category. In contrast, as we show in section 3, the adaptive replication algorithms that we propose here can be combined with a concurrency control algorithm to preserve 1 copy serializability. Recently, a few works that ....

Ladin, R., Liskov, B., and Shrira, L. 1990. Lazy replication: Exploiting the semantics of distributed services. In Proceedings of the Workshop on Management of Replicated Data, pp. 31--34.

....local transactions which access conceptually replicated data items. Prior multidatabase frameworks and associated correctness criteria, such as QSR, do not manage these types of transactions. Lastly, algorithms to manage replicated data efficiently, such as Quasi copy [1] and Lazy Replication [26], have relevance to this work. In the quasi copy approach, updates are made at a pre defined primary site or set of sites. Inconsistency is allowed between the primary site and other replicated sites, and can be restored by sending the updated image from the primary site to the replicated sites. ....

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: Exploiting the semantics of distributed services. In Proceedings of the Ninth Annual ACM Symposium on Principles of Distributed Computing, Quebec City, Canada, August 1990. 24

....in the transaction tree and once more time for the root Clearly, this would be unacceptably costly. To avoid this problem, Liskov s ARGUS group proposed an approach in which commit decisions are deferred, so that only the top level commit protocol is actually executed as a multiphase protocol [LS83, LLSG90, LCJS87]. Intermediate commits are optimistically assumed successful, while aborts are executed directly by informing the commit participants of the outcome. Now, the issue arises of how to handle an access by a subtransaction to a lock held by a sibling subtransaction or to a data item updated by a ....

....system was an early leader among transactional computing systems that considered transactions on abstract objects. Developed by a team lead by Liskov at MIT, the Argus system consists of a programming language and an implementation that was used primarily as a research and experimentation vehicle [LS83, LCJS87, LLSG90]. Many credit the idea of achieving distributed reliability through transactions on distributed objects to this project, and it was a prolific source of publications on all aspects of transactional computing, theoretical as well as practical, during the decade or so of peak activity, The basic ....

Rivka Ladin, Barbara Liskov, Liuba Shrira and S. Ghemawat. Lazy Replication: Exploiting the Semantics of Distributed Services. In Proceedings of the 10th ACM Symposium on Principles of Distributed Computing, (Quebec, CA; Aug. 1990). ACM. 4358.

....consistent. However, this solution is limited because replica processing is blocked in the case of network or node failure [PHN87] In addition performance degrades as the number of nodes increases. An alternative solution relaxes the mutual consistency requirement by using lazy replication [Lad90], Gol95] In this case, each time an update transaction updates its replica copy it is committed locally and afterwards all other replica copies are updated in separate refresh transactions. Since mutual consistency is relaxed, the concept of freshness is used to measure the degree of deviation ....

Rivka Ladin. Lazy replication: Exploiting the semantics of distributed services. Proceedings of the Workshop on Management of Replicated Data, pages 31--34, 1990.

....with weak coherency requirements don t necessarily have to participate in replication control algorithms. Consider for example a snapshot: An update on the database doesn t have to contact the snapshot for update synchronization. Another operational approach is described by Ladin et al. [16]. Lazy replication allows consistency specification by introducing different types of operations, depending on the ordering requirements of an operation. So far we have only classified different approaches according to the techniqe of consistency specification. This classification is summarized ....

....updates are not allowed. The last class of weak consistent replication approaches to be mentioned are those that allow independent updates and try to enforce convergence by defining a globally unique order of update operations or by exploiting the semantics of different types of update operations [1, 7, 9, 16]. Independent updates, of course, increase site autonomy and allow best performance results for update operations. Furthermore, protocols supporting independent updates typically are very robust in the presence of failures like network partitions because updates don t need to be propagated and ....

Ladin, R.; Liskov, B.; Shrira, L.: Lazy Replication: exploiting the semantics of distributed services. Operating Systems Review, 25(1), January 1991, pp. 49-55

....supports semantics dependent rules for automatic recovery of specific applications such as directories. No general rules are supplied. In addition to directory file system propagation, asynchronous communication techniques have also been developed. An example of the latter is Lazy Replication [18]. It provides three classes of ordered messages, plus unordered delivery. This facility is at a much lower level of abstraction than transactions and ETs. Another example is Isis [6] which provides four kinds of multicast and broadcast facilities, including causal broadcasts. These communications ....

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: Exploiting the semantics of distributed services. In Proceedings of the Ninth ACM Symposium on Principles of Distributed Computing, Quebec City, August 1990. ACM/SIGACT-SIGOPS.

....with concurrency control, we describe it in Section 5.1. Concurrency Replica Control Communications Control Mapping Consistency Other Mechanism None ROWO 1 Copy Equivalence Static Mapping totally ordered broadcast Virtual Synchrony with ABCAST 3.1. 2 Lazy Replication Lazy replication [30] is a way to preserve consistency by exploiting the semantics of the service s operations to relax the constraints on ordering. Since some operations need not be ordered identically at all replicas, lazy replication guarantees one copy equivalence with improved response time and availability. Lazy ....

Rivka Ladin, Barbara Liskov, and Liuba Shrira. Lazy replication: Exploiting the semantics of distributed services. In Proceedings of the 1990 Symposium on Principles of Distributed Computing, pages 43--57, 1990.

.... drawbacks: In order to determine the current document, read operations usually must access more than one replica and atomic updates are required in order to preserve one copy serializability as consistency notion [Bernstein et al. 1985] The second approach is to weaken the consistency criterion [Ladin et al. 1990, Theel and Raynal 1997] Assuming that updates eventually reach every replica, it possible to allow read operations be performed in the past. This means that read operations return a valid, although not necessarily the most current version of the replicated document. The advantage of this ....

R. Ladin, B. Liskow, and L. Shira. Lazy Replication: Exploiting the Semantics of Distributed Services. In Proc. of the 9th Annual ACM Symposium on Principles of Distributed Computing, pages 43-- 57, 1990.

....that result from using prev sets. We substitute the multipart timestamp technique in place of prev sets to keep track of system dependencies. The multipart timestamp technique is based on Lamport s logical clocks [16] The approach is similar to the multipart timestamp implementation in [18]. 2 ESDS Overview This section is an overview of the work upon which our project is based. Section 2.1 describes the abstract algorithm from [1] Section 2.2 briefly summarizes the prototype implementation from [2] 2.1 ESDS Algorithm ESDS is a distributed data service based on the lazy ....

....an operation has several attributes that specify its semantics. First, the operation defines the transformation to be applied to the data object and the value to be returned to the user. Second, the operation may be causally dependent on other operations previously processed by the system [18]. To capture this dependence, the operation s state contains a prev component, a set of operation uids that must be executed before it. Despite allowing transient inconsistency among the replicas, ESDS provides provable guarantees that in the limit an eventual total order is established on all ....

[Article contains additional citation context not shown here]

R. Ladin, B. Liskov, L. Shrira, and S. Ghemawat. Lazy replication: Exploiting the semantics of distributed services. ACM Transactions on Computer Systems, 10(4):360-391, Nov. 1992.

....Improving performance by providing weaker consistency guarantees may lead to more complicated semantics. While in practice, replicated systems are often incompletely or ambiguously specified, it remains very important to provide formal consistency guarantees. Ladin, Liskov, Shrira, and Ghemawat [18] define one highly available replicated data service. They specify general conditions for such a service, and present an algorithm based on lazy replication, in which operations received by each replica are gossiped in the background. Responses to operations may be out of date, not reflecting the ....

....service, and present an algorithm based on lazy replication, in which operations received by each replica are gossiped in the background. Responses to operations may be out of date, not reflecting the effects of operations that have not yet been received by a given replica. Building on the work of [18], Fekete et al. 1] specify a flexible eventially serializable data service that we use in this work. 1.2 Experimental ESDS Implementation The ESDS algorithm is specified as a composition of I O Automata. I O Automata [2] are specified as state machines using a declarative description language. ....

[Article contains additional citation context not shown here]

R. Ladin, B. Liskov, L. Shrira, and S. Ghemawat. Lazy replication: Exploiting the semantics of distributed services. ACM Transactions on Computer Systems, 10(4):360391, Nov. 1992.

....list of consistency contracts. In contrast, Munin proposes the release consistency contract to its clients. This consistency contract is too strong for certain applications e.g. cooperative editing. Programmers of these applications might benefit from the support of weaker consistency contracts [1, 14, 16]. Second, Munin considers two access types: read and write. Our system proposes a generic approach that permits a consistency manager to exploit the semantics of object methods and hence, a richer set of access types. 4.2 The Shadows tool box of basic building blocks Shadows [5] provides a ....

Rivka Ladin, Barbara Liskov, and Liuba Shrira. Lazy replication: Exploiting the semantics of distributed services. Operating Systems Review, 25(1):49--55, January 1991. INRIA Adaptable Replicated Objects in Distributed Environments 23

....or gossip mechanisms [22] 37] Another approach that avoids two phase commit is given in [29] transactions commit at a primary site and are propagated asynchronously. A partial order among transactions is defined for controlling update propagation; the use of a partial order is also proposed in [27] and in [15] in the context of concurrency control for groupware systems) Finally, some systems allow even more flexibility. In [13] a system is described for maintaining weak consistency among various database systems. Updates are accepted at any site, and are guaranteed to be reflected at all ....

....both [16] 37] do not consider a reconciliation phase, which is one of the main component of our approach. Instead of using a specific reconciliation mechanism, a site can synchrously send part of its log and matrix to other sites. The use of vectors for update propagation is also suggested in [27]; this is based on gossip messages, where update propagation is enforced by two kinds of messages: update messages through which events are propagated, and ack messages by which a site acknowledges the reception of updates. 1.2. Outline The paper is organized as follows. In Sec. 2 we describe our ....

[Article contains additional citation context not shown here]

R. Ladin, B. Liskov, and L. Shira, "Lazy Replication: Exploiting the Semantics of Distributed Services," Proc. 1st Workshop on Replicated Data, Houston, TX, Nov. 1990, pp. 31--34.

....applications can be simplified if messages are received in order of logical precedence. Since extra complexity would be added to such applications, should the communication subsystem not provide causal delivery, several algorithms have been proposed to implement this ordering discipline [8, 20, 1, 5, 13, 7]. Nevertheless, despite its advantages, the use of causal communication has been somehow limited by the overhead incurred by existing implementations. A major cost of protocols that preserve logical precedence is the size of history information that needs to be stored and exchanged to maintain ....

....potentially causally related [20] To avoid the disadvantages of logical clocks, a new set of algorithms has been proposed, where the information required to precisely define causal relations is piggybacked on the messages exchanged. These approaches are based on causal histories or vector clocks [1, 13, 7]. While some work in this area assumed that all messages were addressed to all processes in the system [19, 13] recent systems allow messages to be addressed exclusively to a subset of the existing processes [1, 7, 12] These subsets may be structured in groups that may sometimes overlap. Group ....

[Article contains additional citation context not shown here]

Rivka Ladin, Barbara Liskov, Liuba Shrira, and Sanjay Ghemawat. Lazy Replication: Exploiting the Semantics of Distributed Services. Technical Report MIT/LCS/TR-84, MIT Laboratory for Computer Science, 1990.

....that ordering comes at low cost. Our work on the Transis project started at 1990. At that time, there were already several projects under way in this area: notably, the Isis system [13, 16] from Cornell University, has been in wide spread use since around 1987. Also, Psync [49] Lazy Replication [37], the Trans and Total protocols [43] Independently and roughly at the same time, two other projects in this area were launched: The Horus project [59, 58] rebuilds Isis from scratch, and the Amoeba operating system project integrates a group communication facility within the kernel of the ....

....in industrial and in academic settings. The Isis package supports a variety of multicast communication services, and provides the application programmer with the virtually synchronous model of programming [10, 14] Isis mandates the primary component consistency model. The Gossip project at MIT [40, 37] is based on a novel approach called lazy replication. It is suited for replicating a single service, and lets the client of the service specify the potential causal dependency among requests. In this way, unrelated requests of different clients do not incur any latency delay due to communication, ....

R. Ladin, B. Liskov, L. Shrira, and S. Ghemawat. Lazy Replication: Exploiting the Semantics of Distributed Services. In 9th Ann. Symp. Principles of Distributed Computing, pages 43--58, August 90.

....complex and highly concurrent interactions between several participants, whose membership may be largely dynamic. Striking examples are found in the domains of computer supported cooperative work, and distributed computer control. These and other applications exemplified in the literature [7,10], require complementing paradigms. One such paradigm gaining increasing acceptance is reliable group communication or multicasting, concerning the dissemination of information to a group of participants in a system. Algorithms and protocols to solve these problems have been presented in the last ....

....supply a range of functionality let us call it quality of service or QOS, a terminology very used in the communications community including addressing modes, group management support, and delivery properties. There exist a number of solutions providing varying degrees of order, such as [3,10,7,14]. Our subsystem, besides a range of order properties, from total and causal to FIFO, provides different agreement and synchronism properties, such as best effort and at least, or loose and tight synchrony 3 . Their combination, according to user requirements, yields the different qualities of ....

Rivka Ladin, Barbara Liskov, Liuba Shrira, and Sanjay Ghemawat. Lazy Replication: Exploiting the Semantics of Distributed Services. Technical Report MIT/LCS/TR-84, MIT Laboratory for Computer Science, 1990.

....has been a research issue for more than two decades. Although often limited in earlier systems, support for fault tolerance and replication has been subject to a growing interest in the last few years. Several approaches have been proposed, including many in the area of remote invocation [7,9,8]. However, most existing systems provide support for at most a limited set of pre defined replication strategies. Approaches like problem oriented shared memory [6,16,11] which allow different replication strategies to be applied to different objects are a key factor for efficiency in distributed ....

....to enforce reliable multicast communication between clients and replicated services we only require virtually synchronous communication among replicas. Additionally, we add the concept of unpropagated messages to maintain causality in the presence of weak replication schemes. Lazy replication [9] is similar to our scheme in that it relies on inexpensive pointto point interactions between clients and the replicated service. Global consistency is achieved by collecting and exchanging multipart timestamps; we achieve the same effect implicitly by relying on the services of the underlying ....

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: Exploiting the semantics of distributed services. In Proceedings of the Ninth Annual ACM Symposium of Principles of Distributed Computing, pages 43--57, Quebec City -- Canada, August 1990. ACM.

....(both physical and logical) on each DPQ. Our major goal is the design and analysis of concurrent methodologies which point out the relationship between efficiency and scalability. The next phase, not considered here, consists of supporting Fault tolerance (through replication and lazy update [9, 14]) and topologically dynamic network, i.e. with a growing or decreasing number of processors [13] Simultaneously, the DPQ is used to solve some scientific problems requiring high performance computing, e.g. Branch and Bound for Combinatorial Optimization problems, Event Set model for Discrete ....

R. Ladin, B. Liskov, and L. Shira. Lazy replication: exploiting the semantics of distributed services. In Proceedings of the 9th Symposium on Principles of Distributed Computing (PODC'90), pages 43--57, 1990.

....next section extend this model to a distributed environment. 2.2 Replicating channels Replicated channel types offer a generic interface for replicating data and operations, and for enforcing consistency between replicas. Well known consistency semantics are strong consistency, causal consistency [9], weak consistency and release consistency [3] All these consistency semantics are well suited to certain classes of application. Each replicated channel implements also a particular protocol used to synchronize replicas (e.g. update, invalidate) The class of data structure in use influences ....

Rivka Ladin, Barbara Liskov, and Liuba Shrira. Lazy replication: Exploiting the semantics of distributed services. In IEEE Computer Society Technical Committee on Operating Systems and Application Environments, volume 4, pages 4--7. IEEE, IEEE Computer Society, 1990.

....[HHW89] propose a lower overhead solution, although garbage collection is more conservative. The acknowledgment algorithm presented in Section 6.5 borrows several ideas from Wuu. Consistency algorithms that use data semantics provide high data availability. Garcia [GAB83] Sarin [Sar86] and Ladin [LLS90] generalize the directory solution for other data types. Garcia uses a data patch tool that automates the data integration step. The correct final state of a database and corrective measures to reach it are predefined for various potential non serializable operations. These are used to reach ....

....with the mobile user by capturing their effective state at one site and then restarting the captured state at the next site. This solution is application specific and does not address issues related to file consistency in a replicated mobile environment. 8. 3 Causal Consistency Ladin, et al. [LLS90] support causal ordering of data reads, and causal, forced and immediate ordering for updates. This work borrows some of its ideas from previous work at ISIS [BJ87] Since Ladin develops a more efficient solution for replicated systems, we will only discuss their work. Causal ordering is similar ....

Rivka Ladin, Barbara Liskov, and Liuba Shrira. "Lazy Replication: Exploiting the Semantics of Distributed Services." In Proceedings of the Workshop on Management of Replicated Data, pp. 31--34. IEEE, November 1990.

....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 principle have been published with different degrees of fault tolerance [4, 8, 6, 3]. Finally, there is a class of algorithms known by the name of Replica Generated Identifiers [12] In this technique, total order is computed in two phases. In the first phase, message recipients propose candidate unique identifiers for the message. In the second phase one candidate is selected ....

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: exploiting the semantics of distributed services. In Proceedings of the Workshop on the Management of Replicated Data, pages 31--34, Houston - USA, November 1990. IEEE.

....have used weak consistency are internet news, air traffic control, air line reservation and stock exchanges. Grapevine [SBN84] and the Global Name Service [Lam86] were among the first systems to use weak consistency. Other weak consistency protocols were presented in [DGH 87, DGP90b, Gol92, LLS90, QP93, WB84] However, these protocols assumed that any node could communicate with any other node. This assumption, though appropriate for small networks with a few replicas, is unrealistic for wide area networks like the Internet. This paper presents a Hierarchical Asynchronous Replication ....

....The flooding approach suffers from incurring a large number of messages and lots of redundancy which causes waste in network bandwidth. The second approach is based on probabilistic techniques and has been used by most of the previous weak consistency protocols [DGH 87, DGP90b, Gol92, LLS90, SBN84, QP93] These schemes rely on delayed propagation, where nodes exchange status and new messages periodically in sessions. Partners in these sessions are selected to include either all nodes or a subset of nodes drawn from a probabilistic function (e.g. randomly or such that communication ....

[Article contains additional citation context not shown here]

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: exploiting the semantics of distributed services. In Proceedings of the 9th ACM symposium on Principles of Distributed Computing, pages 43--57, Quebec City, CA, August 1990.

....instead of atomic transactions. Second, we deal with objects replicated over multiple sites. That is, our emphasis is on increasing concurrency of independent operations over multiple sites rather than increasing concurrency among transactions on a single site. Finally, we compare our work with [29]. Here, lazy replication has been proposed as a way to preserve consistency by exploiting the semantics of the service s operations to relax the constraints on ordering. Three kinds of operations are supported: operations for which the clients define the required order dynamically during the ....

....collection of different kinds of operations, some requiring total ordering and some requiring partial ordering with respect to one another. Our approach performs much better in a case such as this when mixture of these different operations needs to be applied to an object. The approach proposed in [29] must resort to a total ordering in such a case. 7.2 Membership Membership protocols have been proposed for both synchronous and asynchronous environment. Membership protocols in synchronous systems include [10 , 16, 27, 33, 43] All these protocols make use of synchronized clocks to maintain a ....

R. Ladin, B. Liskov, L. Shrira, and S. Ghemawat. Lazy replication: Exploiting the semantics of distributed services. Technical Report MIT/LCS/TR-484, MIT Laboratory for Computer Science, Cambridge, MA, Jul 1990.

.... style called virtual synchrony for replicated services: The events in the system are delivered to all the components in a consistent order, allowing them to undergo the same changes as if the events are synchronous ( 8, 7] Another system providing high availability services is described in [13, 15]. They show how to replicate a service efficiently using a lazy asynchronous form of updating. However, the information required for the ordering of updates is carried by the user requests. Our service definitions are greatly influenced by the ISIS experience and the virtual synchrony concept. ....

R. Ladin, B. Liskov, L. Shrira, and S. Ghemawat. Lazy replication: Exploiting the semantics of distributed services. In Ann. Symp. Principles of Distributed Computing, number 9, pages 43--58, August 90.

....Gaggle invocations may be dealt with by different receivers is not a concern. Causally Consistent Name Service. Although the weak consistency of a Lampson style name service helps to ensure high availability and low latency, Ladin has observed that sometimes this semantics is inadequate [13]. For example, suppose a system administrator wishes to create a new sub directory and then to populate it with information about printers. If the update that creates the directory goes to one server and the request to add the first printer goes to another, the operation to add the printer may ....

Ladin, R., Liskov, B. and Shrira, L. "Lazy Replication: Exploiting the Semantics of Distributed Services". Proc. of the 9th ACM Symp. on Prin. of Distributed Computing, Quebec City, Quebec, August 1990, pp.43-57.

....in Amoeba to implement the directory service. In [14] Spiewak describes a framework for replicating objects in Spring OS using primary backup algorithm. Here the replicas are periodically updated. Thus the replicas will not always be in a mutually consistent state. The reader can refer to [15, 16, 17, 18] for additional reading on replication. 2.1 Motivations for this Work The foregoing discussion mainly dealt with various algorithms for object replication. These algorithms basically keep the copies of replicated objects in a mutually consistent state. However we are interested in building a ....

R. Ladin, B. Liskov, L. Shrira, and S. Ghemawat, "Lazy replication: Exploiting the semantics of distributed services," Technical Report MIT/LCS/TR-484, Massachusetts Institute Of Technology, July 1990.

....protocol for managing replicated data asynchronously. This method is especially suited for the bulletin board application mentioned above. Grapevine [17] and the Global Name Service [15] were among the first systems to use weak consistency. Other weak consistency protocols were presented in [7, 8, 12, 14, 16, 18]. These protocols are useful and interesting; however, they require a node to send messages to all other nodes. This would be acceptable for small networks with a few replicas, but it will incur a large communication overhead for wide area networks like the Internet. Moreover, a node sends not ....

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: exploiting the semantics of distributed services. In Proceedings of the 9th ACM symposium on Principles of Distributed Computing, pages 43--57, Quebec City, CA, August 1990.

....processes, inter site protocols run among processors (even if on behalf of processes) 35] Figure 1 makes explicit which are the site and the participant level modules. Network Infrastructure A number of distributed protocols in the recent years have been designed to be network independent [19,5,16]. However, in trying to be generic and scalable, they do not take advantage of the existence and the emergence of network technologies such as LANs and MANs. In consequence, while these will perform well either in local or wide areas, synchronous or asynchronous environments, etc. users will be ....

....to accommodate grades of agreement, order and synchronism; ffl supply a range of services each formed by a combination of some of the properties above; ffl select those services to fulfil user needs, i.e. making a good use of the end to end argument. Rationale for this exercise can be found in [19,5,27,24]. 0.1 Site group membership The problem of membership is the problem of knowing who or what belongs to a system or group, or is present in an activity. At this point, it is important to draw a distinction not taken into account by most group oriented systems we know of: the difference between ....

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: Exploiting the semantics of distributed services. In Proceedings of the Workshop on the Management of Replicated Data, pages 31--34, Houston - USA, November 1990. IEEE.

....list of consistency contracts. In contrast, Munin proposes the release consistency contract to its clients. This consistency contract is too strong for certain applications e.g. cooperative editing. Programmers of these applications might benefit from the support of weaker consistency contracts [1, 15, 17]. Second, Munin considers two access types: read and write. Our system proposes a generic approach that permits a consistency manager to exploit the semantics of object methods and hence, a richer set of access types. 4.2 The Shadows tool box of basic building blocks Shadows [5] provides a ....

Rivka Ladin, Barbara Liskov, and Liuba Shrira. Lazy replication: Exploiting the semantics of distributed services. Operating Systems Review, 25(1):49--55, January 1991.

....orders. A total ordering means that all processes will see the same messages in the same order, though that order will not necessarily be the order messages were sent. Causal ordering implies that any messages with a potential causal relation will be delivered in the same order at all replicas [8, 9]. Messages with no causal relation can be delivered in different orders at different processes. A bound inconsistency ordering ensures that the database at one site never differs from the correct global value by more than a constant [10, 11] Weaker orderings include a per process or FIFO channel ....

R. Ladin, B. Liskov, and L. Shrira, "Lazy replication: exploiting the semantics of distributed services," Operating Systems Review, vol. 25, pp. 49--55, January 1991.

....requiring complex and highly concurrent interactions between several participants, whose membership may be largely dynamic. Striking examples are found in the domains of computer supported cooperative work, and distributed computer control. These and other examples exemplified in the literature [14,18,5], require complementing paradigms. One such paradigm gaining increasing acceptance is reliable group communication (multicasting) concerning the dissemination of information to a group of participants in a system. The implementation of this paradigm meets a number of problems, due to natural ....

....supply a range of functionality let us call it quality of service or QOS, a terminology very used in the communications community including addressing modes, group management support, and delivery properties. There exist a number of solutions providing varying degrees of order, such as [7,18,14,24]. Our subsystem, besides a range of order properties, from total and causal to FIFO, provides different agreement and synchronism properties, such as best effort and at least, or loose and tight synchrony 3 . Their combination, according to user requirements, yields the different qualities of ....

Rivka Ladin, Barbara Liskov, Liuba Shrira, and Sanjay Ghemawat. Lazy Replication: Exploiting the Semantics of Distributed Services. Technical Report MIT/LCS/TR-84, MIT Laboratory for Computer Science, 1990.

....can tolerate some inconsistency and reconciliation methods should be available to resolve conflicts. Typical applications that have used weak consistency are internet news, air traffic control, airline reservation and stock exchanges. Several weak consistency protocols were presented in [10, 11, 15, 19, 22, 23]. However, they require a node to communicate with all nodes which will incur a large communication overhead for wide area networks like the Internet. This paper evaluates the performance of HARP, a hierarchical replication protocol that scales well for thousands of replicas while ensuring ....

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: exploiting the semantics of distributed services. In Proceedings of the 9th ACM symposium on Principles of Distributed Computing, pages 43--57, Quebec City, CA, August 1990.

....ESDS guarantees eventual consistency of the replicated data, while allowing the clients to choose between strict operations with guaranteed consistent responses and non strict operations that return responses consistent with some ordering of operations. ESDS builds on the work of Ladin et al. [3], who defined a data service with relaxed consistency and presented an algorithm based on lazy replication. The ESDS paper [1] includes a service specification and an abstract distributed algorithm; both are given in terms of I O automata of Lynch and Tuttle [2] The abstract algorithm is shown to ....

R. Ladin, B. Liskov, L. Shrira, and S. Ghemawat. Lazy replication: Exploiting the semantics of distributed services. ACM Transactions on Computer Systems, 10(4):360-391, Nov. 1992.

.... Centralized Server This scheme makes use of a logically centralized service that tracks all inter node references [LL92] The implementation of the service may actually be distributed, and may use replication for high availability and reliability, but it appears as if it were run by one server [LLSG90]. Nodes communicate with the service to provide it with information of their outlists and translists (Section 2.4.2) typically, soon after they have performed a local collection. They also query it for the accessibility of their inlist entries. More precisely, a node queries whether an inlist ....

R. Ladin, B. Liskov, L. Shrira, S. Ghemawat. Lazy Replication: Exploiting the Semantics of Distributed Services. Proceedings of the 9th ACM Symposium on the Principles of Distributed Computing, ACM, Canada, Aug 1990.

No context found.

R. Ladin, B. Liskov, and L. Shira, "Lazy Replication: Exploiting the Semantics of Distributed Services," Proc. 1st Workshop on Replicated Data, Houston, TX, Nov. 1990, pp. 31--34.

No context found.

R. Ladin, B. Liskov, and L. Shrira. Lazy replication: Exploiting the semantics of distributed services. In ACM Symposium on Principles of Distributed Computing, 1990.

No context found.

Rivka Ladin, Barbara Liskov, and Liuba Shrira. Lazy replication: Exploiting the semantics of distributed services. In IEEE Computer Society Technical Committee on Operating Systems and Application Environments, volume 4, pages 4--7. IEEE Computer Society, 1990.

No context found.

Ladin, R., Liskov, B. and Shrira, L. "Lazy Replication: Exploiting the Semantics of Distributed Services". Proc. of the 9th ACM Symp. on Prin. of Distributed Computing, Quebec City, Quebec, August 1990, pp.43-57.

No context found.

Ladin, R., Liskov, B., Shrira, L.: Lazy replication: exploiting the semantics of distributed services. Proc. 9th ACM Symp. on the Principles of Distributed Computing, (1990).

First 50 documents  Next 50

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