O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In S. J. Mullender, editor, Distributed Systems, pages 55--96. Addison-Wesley, 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....this thesis will only discuss time driven discrete simulation models. A distributed system is a collection of sequential processes P 1 , P 2 , and P n , and a network of unidirectional communication channels. Each channel connects a pair of processes and delivers messages between them [3]. Distributed simulations typically provide more overall host processing power and more storage space than sequential simulation systems, and thus support more simulation entities and more detailed appearance and behaviors of entities within the system. The distributed structure also allows ....

....then e f if and only if [34] 1. a process executes both e and f , e occurs before f ; or 2. e = send(m) and f = receive(m) for some message m; or 3. there exists an event h, such that e h and h f . Then the causal ordering of events is defined as satisfying the following requirement [3]: send i (m) send j (m ) deliver k (m) deliver k (m ) for all messages m, m , sending processes p i , p j and destination process p k . 55 A widely used example to prove the necessity of maintaining causal order in a distributed system is shown in Figure 4.6. Since deliver 2 (m ....

[Article contains additional citation context not shown here]

Ozalp Babaoglu and Keith Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In Sape Mullender, editor, Distributed Systems. ACM Press, 1993.

....established process migration model, and how these abstractions can be applied in our problem domain. 2.2 Distributed systems and distributed applications We will first determine how a distributed system can be defined. In the literature, a variety of explanations exist. Babaoglu and Marzullo [3] defines a distributed system as consisting of a collection of processes and a network capable of implementing communication channels between these processes. Identifying these two major components is important both during the system design phase, and when evaluating the final product. The ....

....is necessary. In his classic paper [33] Lamport shows how logical clock values can be used to synchronize the events of a system. The suggestion is to apply the events taken by the message exchange. These events may be used for synchronizing the sequential processes. Babaoglu and Marzullo [3] employs a general abstraction of FIFO delivery to show that the ordering of the messages can be maintained for all causally related messages, even if they are sent by independent processes. The resulting property is called causal delivery (CD) and may be expressed as: CD: send m send m deliver ....

[Article contains additional citation context not shown here]

Babaoglu, Ozalp and Marzullo, Keith: "Consistent global states of distributed systems: Fundamental concepts and mechanisms", in "Distributed systems, 2. ed.", Mullender (ed.), ACM Press, 1993.

....may be found in [15] Aside from practical implementation work, much research has been devoted to the study of problems arising in distributed debugging. As noted before, we classify these problems into those that deal with observation and those that deal with control. Predicate Detection [1, 6] is the main problem involved in observing distributed computations. It involves detecting whether a specified global property ever occurs in a distributed computation. For example, one might wish to check that the validity of mutually exclusive sections of code is maintained. Predicate detection ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: fundamental concepts and mechanisms. In S. Mullender, editor, Distributed Systems, chapter 4. Addison-Wesley, 1993.

....have to be computed. If the property is stable (once true it remains true forever) a snapshot algorithm can be used to detect it [3] When the property is not stable, all the global states, through which the computation could have passed, have to be considered. These states constitute a lattice [1, 15]; a node of the lattice represents a possible global state of the distributed computation and an edge represents an event that changes the global state of the computation. A general method to detect such unstable properties consists in building the lattice associated with the distributed execution ....

....[2, 12] this can be done on the fly by pipelining the construction and the traversal of the lattice with the execution. The basic problem with the detection of general (unstable) properties on global states is that the size of the lattice can be exponential with respect to the number of processes [1, 15]. However for some specific properties on global states such as conjunction of local predicates [9] relational global predicate [16] or inevitable global states [7, 9] the lattice construction is not necessary. The subject of this paper is the second class of properties, namely the ones on ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: fundamental concepts and mechanisms, in Distributed Systems, chapter 4, pages 55--93. ACM Press, Frontier Series, (S.J. Mullender Ed.), 1993.

....algorithms that run on the system with mobile computing devices therefore require some modi cations to compensate for these factors. In this paper, we consider causal message ordering required in many distributed applications such as management of replicated data [8, 9] distributed monitoring [6], resource allocation [18] distributed shared memory [3] multimedia systems [2] and collaborative work [19] The protocols to implement causal message ordering in systems with static hosts have been presented in [14, 9, 16, 18, 20, 21] These protocols can be executed by every mobile host with ....

O. Babaoglu and K. Marzullo. Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms. In Sape Mullender, editor, Distributed Systems, pages 55-96. Addison-Wesley, 1993.

....objects or actors which are garbage based on that graph. Since there is no omniscient observer in a distributed system, a global snapshot is usually constructed by taking a consistent cut of the global state and accounting for any messages in transit in the network. The reader is referred to [5, 34] for details. However, such a global snapshot approach requires synchronization among all the hosts in the system which raises questions about scalability. For this reason, most algorithms try to avoid taking a global snapshot. 35 There are several other algorithms available in the literature ....

Ozalp Babaoglu and Keith Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In S. Mullender, editor, Distributed Systems, pages 55--96. Addison-Wesley, 1993.

....perform the protocol. synchronous rounds: minimum number of synchronous rounds required by the protocol, presuming that every party is allowed to send arbitrarily many messages with every time tick and to receive arbitrarily many messages sent by other parties at the beginning of a round (cf. [BM93] p. 133) simple rounds: minimum number of required synchronous rounds, presuming that every party sends and receives at most one message per round (cf. ABM87] Whereas the total number of messages of a protocol measures the number of packages sent, the number of exchanges can be considered ....

Babaoglu, O., Marzullo, K., Consistent Global State of Distributed Systems: Fundamental Concepts and Mechanisms. In S. Mullender (ed.), Distributed Systems. NY: ACM Press, 1993, 55--145.

....perform the protocol. synchronous rounds: minimum number of synchronous rounds required by the protocol, presuming that every party is allowed to send arbitrarily many messages with every time tick and to receive arbitrarily many messages sent by other parties at the beginning of a round (cf. [BM93] p. 133) simple rounds: minimum number of required synchronous rounds, presuming that every party sends and receives at most one message per round (cf. ABM87] Whereas the total number of messages of a protocol measures the number of packages sent, the number of exchanges can be considered ....

Babaoglu, O., Marzullo, K., Consistent Global State of Distributed Systems: Fundamental Concepts and Mechanisms. In S. Mullender (ed.), Distributed Systems. NY: ACM Press, 1993, 55--145.

....of dependent events and observation [5] In many applications, causal order based on logical time is not enough to capture the dependencies needed by the application s logic. Vector clocks bring progress over logical (scalar) timestamps by inducing an order that exactly reflects causal precedence [6 8]. Matrix clocks [9, t0] extend vector clocks by cap turing a global property, namely what A knows about what B knows about C . Such shared knowledge is needed in many instances involving close cooperation, such as replica update management and collaborative work. However, matrix clocks require ....

O. Babaoglu and K. Marzullo. Consistent Global States of Distributed Systems: Fundamental Consepts and Mechanisms, chapter 3. S. Mullenderr, addison-wesley edition, 1993.

....entities that have significant nonlinear movements and action vocabularies. A distributed system is a collection of sequential processes P 1 , P 2 , and P n , and a network of unidirectional communication channels. Each channel connects a pair of processes and delivers messages between them [2]. Distributed simulations typically provide more overall host processing power and more storage space than sequential simulation systems, and thus support more simulation entities and more detailed appearance and behaviors of entities within the system. The distributed structure also allows ....

Ozalp Babaoglu and Keith Marzullo: "Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms" In Sape Mullender, editor, Distributed Systems. ACM Press, 1993.

....is changed for a move out event, a message is sent which notifies the next cell of the change. In our implementation, the GVT (global virtual time) computation is performed by means of a coloring algorithm [22] to determine the set of all the messages in transit before a certain consistent cut [2] of the distributed computation. A token message circulates among all processes in Stage 2, along a predefined and fixed virtual ring of processes. The computation of each process is cyclically colored with one of three colors (white, black, and red) and the current color is given to all messages ....

O. Babaoglu and K. Marzullo, Consistent global states of distributed systems: fundamental concepts and mechanisms, in: Distributed Systems (Addison-Wesley, 1995)

....to visualize concurrent execution, the statetransition approach requires a global snapshot of algorithm state. Unfortunately, in distributed environments global snapshots are costly due to distributed state and asynchrony, and may not correspond to any state entered by the underlying execution [23]. Moreover, semantically equivalent execution behavior may yield different state transitions. As a result, the statetransition approach is costly to implement and does not effectively abstract over the relevant behavior in distributed systems. A more natural model for visualizing distributed ....

....on visualization actions: Causal Connection Restriction: The invocation order of visualization actions must preserve the causal order of actor events which trigger them. Note that under the casual connection restriction, the resulting visualization always corresponds to a consistent cut [23] of the triggering events in the underlying execution. A B C A 1 B 1 B 2 C 1 C 2 Figure 2.2: Event Diagram. Three actors A, B and C are shown together with their visualization events. Figure 2.2 illustrates how the causal connection restriction affects visualization. The event ....

[Article contains additional citation context not shown here]

Ozalp Babaoglu and K. Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In S. Mullender, editor, Distributed Systems, chapter 4, pages 55--96. ACM Press, New York, NY, 1994.

....have been moved to CLP1. 5 Open Issues A number of challenging issues have to be addressed before this approach is realised. Techniques are required to obtain global snapshots of the distributed simulation and approximate the spheres of influence at any instant, e.g. Chandy Lamport 1985, Babaoglou Marzullo 1993). Furthermore, algorithms for redistributing the state and reorganising the tree to approximate the spheres of influence and balance the load to achieve high simulation performance must be developed; in this context appropriate performance metrics and cost functions need to be defined which will ....

Babaoglou, O. & Marzullo, K. (1993), Consistent global states of distributed systems: Fundamental concepts and mechanisms, Technical Report UBLCS-93-1, Laboratory for Computer Science, University of Bologna.

....algorithms that run on the system with mobile computing devices therefore require some modifications to compensate for these factors. In this paper, we consider causal message ordering required in many distributed applications such as management of replicated data [8, 9] distributed monitoring [6], resource allocation [18] distributed shared memory [3] multimedia systems [2] and collaborative work [19] The protocols to implement causal message ordering in systems with static hosts have been presented in [14, 9, 16, 18, 20, 21] These protocols can be executed by every mobile host with ....

O. Babaoglu and K. Marzullo. Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms. In Sape Mullender, editor, Distributed Systems, pages 55--96. Addison-Wesley, 1993.

....algorithms that run on the system with mobile computing devices therefore require some modifications to compensate for these factors. In this paper, we consider causal message ordering required in many distributed applications such as management of replicated data [5, 6] distributed monitoring [4], resource allocation [11] distributed shared memory [2] multimedia systems [1] and collaborative work [12] The protocols to implement causal message ordering in systems with static hosts have been presented in [7, 6, 9, 11, 13, 14] These protocols can be executed by every mobile host with ....

O. Babaoglu and K. Marzullo. Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms. In Sape Mullender, editor, Distributed Systems, pages 55--96. Addison-Wesley, 1993.

....and part of the state is feasible. However, a number of challenging issues have to be addressed before this approach is realised. Techniques are required to obtain global snapshots of the distributed simulation and approximate the spheres of influence at any instant, e.g. Chandy Lamport 1985, Babaoglou Marzullo 1993) Furthermore, algorithms for redistributing the state and reorganising the tree to approximate the spheres of influence and balance the load to achieve high simulation performance must be developed; in this context appropriate performance metrics and cost functions need to be defined which will ....

Babaoglou, O. & Marzullo, K. (1993), Consistent global states of distributed systems: Fundamental concepts and mechanisms, Technical Report UBLCS-93-1, Laboratory for Computer Science, University of Bologna.

....for every state S i in the observation the next state S i 1 is the result of a single event on one process. In this case we say that S i precedes S i 1 . It can be shown that the set of all consistent global states together with the transitive closure of the precedes relation forms a lattice [1, 13]. Every consistent observation corresponds to a path through the lattice starting at the top element, so observations are not observer independent [13] Given some predicate and by using this lattice, it is possible to define two observer independent notions of observation in the following way ....

....up arrays at different monitors by a superscript, i.e. up j refers to the array on monitor b j . The correspondence is as follows: Process p i has crashed iff up[i] false. Assuming n = 3, the predicate stating that exactly two processes have crashed can be formalized as ( up[1] :up[2] up[3] up[1] up[2] up[3] up[1] up[2] up[3] A question related to this way of formalization is the following: Consider a process p i which has a local variable x initially set to 1. The predicate which we want to detect is x 6= 1. As long as p i is up and ....

[Article contains additional citation context not shown here]

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In S. Mullender, editor, Distributed Systems, chapter 4, pages 55--96. Addison-Wesley, second edition, 1993.

....a computation: Although script driven probing and fault injection has been applied to several complex communication protocols and distributed applications, we need to develop a formal framework for orchestrating a computation. Past work on global snapshots or global 20 predicate detection [4] is closely related to this problem. Furthermore, we need to identify a minimal set of primitives for injecting faults given a certain failure model. Acknowledgment This work is supported by a grant from the U.S. Office of Naval Research. We are very grateful to Jerry Toporek at Mentat for ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanism. In Sape Mullender, editor, Distributed Systems. Addison Wesley, 1993. Second Edition.

....system (a snapshot) at any particular moment. In sequential programming this is straightforward. In distributed simulations however snapshots are not easily obtainable; to determine the system state, all the different local process as well as channel states need to be taken into account [Chan85] [Baba93]. The monitoring system should be able to correlate the histories of the different processes and put them in a global temporal perspective. This is a complicated issue related to the problem of dealing with causality in a distributed environment, discussed in section 3.4. An approach for ....

Babaoglou, O., Marzullo, K., "Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms", Technical Report UBLCS-93-1, Laboratory for Computer Science, University of Bologna, January 1993.

.... j : s i s s j ) and : s j s s i ) A global state Sigma of the distributed execution is a n uple of local states one from each process: Sigma = s 1 ; s 2 ; Delta Delta Delta ; s n ) Such a global state is consistent if it could have been passed through by the distributed execution [1, 12], i.e. CGS Sigma consistent , 8i 6= j : s i k s j 3.3 Weak precedence and strong concurrency The set of local states is also structured by a weaker precedence relation, called weak precedence and denoted w . Informally s i w s j if s i began before s j . In that case it is possible, if ....

.... v(s i ) j] v(s j ) j] F5 s i w s j ) v(s i ) j] v(s j ) j] F6 s i k s j and s i w s j , v(s i ) i] v(s j ) i] and v(s i ) j] v(s j ) j] 3 v i : max(v i ; v(m) j 8x 2 1 Delta Delta Delta n : v i [x] max(v i [x] v(m) x] 4 This relation has been also established in [1]. Irisa Local states in distributed computations: a few relations and formulas 9 4.3 Proofs In order to use results of [4, 11] we associate with each relevant event r i the timestamp of the local state s i produced by this event. As for local states, the timestamp of any relevant event r is ....

[Article contains additional citation context not shown here]

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: fundamental concepts and mechanisms, in Distributed Systems, chapter 4. ACM Press, Frontier Series, (S.J. Mullender Ed.), 1993.

....property for protocols that operate at this level. Of course, we can still use real time for other purposes that demand lesser degrees of accuracy, and will reintroduce it later, but for the time being, we accept this perspective. Babaoglu and Marzullo discuss some uses of consistent cuts in [BM93]. Potential causality is a useful tool for reasoning about a distributed system, but it also has more practical significance. There are several ways to build logical clocks with which causal relationships between events can be detected, to varying degrees of accuracy. p 0 a f e p 3 b p 2 ....

....the algorithm. By introducing such weaker orderings, a group can be made more likely to tolerate failure and can gain a significant performance improvement. 13.16 Related Readings On logical notions of time: Lam78b, Lam84] Causal ordering in message delivery: BJ87a, BJ87b] Consistent cuts: [CL85, BM93]. Vector clocks: Fid88, Mat89] used in message delivery: SES89, BSS91, LLSG92] Optimizing vector clock representations [Cha91, MM93] compression using topological information about groups of processes: BSS91, RVR93, RV95] Static groups and quorum replication: Coo85, BHG87, BJ87a] ....

[Article contains additional citation context not shown here]

Ozalp Babaoglu and Keith Marzullo. Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms. In Distributed Systems (2nd Edition), S.J. Mullender, editor. ACM Press (Addison-Wesley), 1993.

....the technicalities, in the rest of this introduction we discuss informally the structure of the frames, and its impact on the logic presented here, especially with respect to the kind of properties of the computations that can be meaningfully expressed. Asynchronous systems are characterized in [1] by the following properties: there are no bounds on the relative speeds of processes as well as on message delays. Therefore, they rule out synchronized local clocks, and reasoning on global time: message passing is the only possible mechanism for communication. This restricts the ways in which ....

....i j= F Example 2. As an example, the structure (each world is labelled with the propositions holding in it) 1) p ## p; q ## # # # # # # ## p ## p ## p; q ## ## # # # # # # (2) p; q ## q ## r ## s ## ## # # # # # # t ## ## # # # # # # (3) r ## s ## q ## is a model of the formulae: [1] p : p holds in all states of component 1 ( h1ip is false instead) 1] p [3] r : p holds in the initial state of component 1, and r holds in the initial state of component 3; h2i q ( h1i p h3i q) Any state in 2 satisfying q is eventually followed by a state in 1 satisfying p, and ....

[Article contains additional citation context not shown here]

Ozalp Babaoglu and Keith Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In S. Mullender, editor, Distributed Systems, pages 55-96. Addison-Wesley, 1993.

....in the rest of this introduction we discuss informally: the structure of the frames; its impact on the logic adtl presented here, especially with respect to the kind of properties of the computations that can be meaningfully expressed. Asynchronous systems have been characterized in [1] by the following properties: there are no bounds on the relative speeds of processes as well as on message delays. Therefore, they rule out synchronized local clocks, and reasoning on global time: message passing is the only possible mechanism for communication. This restricts the ways in which ....

....3. As an example, the structure (each world is labelled with the propositions holding in it) 1) p ## p; q ## N N N N N ## p ## p ## p; q ## ## M M M M M (2) p; q ## q ## r ## s ## ## L L L L L t ## ## r r r r r (3) r ## s ## q ## when suitably extended, is a model of the formulae: h1i [1] p : p holds in all states of component 1 ( h1ip is false instead) h1i p : p holds in the initial state of component 1; h2i q ( h1i p h3i q) Any state in 2 satisfying q is eventually followed by a state in 1 satisfying p, and by a state in 3 satisfying q; h1i q hxi h2i true) ....

[Article contains additional citation context not shown here]

Ozalp Babaoglu and Keith Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In S. Mullender, editor, Distributed Systems, pages 55-96. Addison-Wesley, 1993.

....of GC consistent cuts, and introduces the two other above mentionned kinds of cuts. It compares the 2 Chandy and Lamport [8] introduced a global state recording algorithm which implements one kind of causal cut; then, Mattern [17] defined causal cuts in the general case. Babaoglu and Marzullo [5] describe causal cuts in detail. INRIA GC consistent Cuts of Databases 5 properties of GC consistent cuts with those of causal cuts. Section 6 summarizes our contribution. Appendix A contains proofs. 2 Fundamental definitions and assumptions In this section we recall some concepts related to ....

....time and causal precedence: having the events s i incomparable by causal precedence is like having them simultaneous. It is beyond the scope of this report to describe the exact extent to which causal cuts are consistent, or to explain in a better than intuitive way why they are consistent; see [5]. 5.1.2 Causal precedence in databases We consider parts in a database as analogous to processes in a distributed system. The contents of a part corresponds with the state of a process. We define causal precedence in a database execution to be a relation between snapshots, i.e. between ....

Ozalp Babaoglu and Keith Marzullo. Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms. In BROADCAST First Year Report, volume 1 (Fundamental Concepts) of LNCS, St. Malo, France, October 1993. which one ?

.... seminal work in this area is the global snapshot algorithm [3] which is used to detect stable bugs (which re main true once they become true) Since then, detection algorithms have been designed for many different classes of bugs such as: race conditions [11] predicates on single global states [1], predicates based on sequences of global states [5] Research in replaying trace computations have focussed on reducing the size of the trace by determining which events are necessary for successful replaying [9] Our approach focusses on adding a control mechanism to the debugging process to ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: fundamental concepts and mechanisms. In S. Mullender, editor, Distributed Systems, chapter 4. Addison-Wesley, 1993.

....of such a distributed execution can only see a sequence of all these events that respects the partial order. Using this sequence of events an observer can reconstruct, starting from the initial state of the computation, a sequence of global states through which the computation may have progressed [1, 17]. One important question is then: are there global states seen by all observations of a distributed computation The detection of such global states (called inevitable in the following) can be very helpful to solve some problems, as an inevitable global state has necessarily been passed through ....

....is delivered to P i , v i is updated to max(v i ; v(m) 1 . 2.4.2 Formulas to detect local states precedence Let v(s i ) and v(s j ) be the timestamps respectively associated with local states s i of P i and s j of P j . From results of [6, 15] on the timestamping of events and results of [1] on the timestamping of local states we can deduce the following relations about the precedence of local states. Other relations and formulas concerning local states can be found in [7] 1 v i : max(v i ; v(m) 8k 2 1 Delta Delta Delta n : v i [k] max(v i [k] v(m) k] RR n2194 10 ....

[Article contains additional citation context not shown here]

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: fundamental concepts and mechanisms, in Distributed Systems, chapter 4. ACM Press, Frontier Series, (S.J. Mullender Ed.), 1993.

....respectively, strictly before the sequences of requests ) i q l and ) j q l were sent to those objects. However, we don t know how requests in ) i p l were interleaved in time with requests in ) j p l . This lack of time ordering information is typical of asynchronous computing systems [BM93]. Therefore we don t know the sequence of state transitions that the view object O underwent between state identifiers 1 p h and p h . If the compression function c produces the same result for any of the possible sequences of state transitions in the view object O , the value of p l is ....

Babaoglu, O. and Marzullo, K. Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms. Distributed Systems, S. Mullender, ed., Addison-Wesley, 1993.

....message by the same process is overtaken in delivery by a later message. If not provided by the network layer, FIFO delivery over non FIFO channels can be implemented by having the source process add a sequence number to its messages and let destinations deliver according to such sequence numbers [4]. Finally, we assume that a reliable, unordered multicast protocol is running at every host providing reliable delivery of a message to all operational hosts in a target multicast group. Ordered multicast should be host minimal, i.e. no other hosts should be affected by multicast of a message ....

O. Babaoglu and K. Marzullo. Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms, chapter 4, pages 55--96. In: S. Mullender (Ed.) - Distributed Systems, Addison-Wesley, 1993.

....reduction in time for solving the general problem. Our predicate detection algorithms can be applied to distributed debugging when processes have independent events, as in multi threaded processes. 1 Introduction A fundamental problem in distributed systems is that of predicate detection [1, 6] detecting whether a global condition occurs while running a distributed program. Its main application is in the testing, debugging and monitoring of distributed programs. Predicate detection is usually specified in a model of distributed computation based on a happened before relation [8] ....

....order models [8, 10] Our work builds on the happened before model [8] by extending the same idea to allow concurrency or independence between events within a local process. We are aware of no other study of such an extended happened before model. Predicate detection is a widely studied problem [1, 6]. Approaches to solving predicate detection are divided into three categories: global snapshot based [2] lattice construction based [4] and predicate restriction based [7] approaches. The first approach can detect only stable predicates (which remain true once they become true) and the second ....

[Article contains additional citation context not shown here]

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: fundamental concepts and mechanisms. In S. Mullender, editor, Distributed Systems, chapter 4. Addison-Wesley, 1993.

....checkpointing. It is obviously desirable to avoid the domino effect and therefore several techniques have been developed to prevent it. One such technique is to perform coordinated checkpointing in which processes coordinate their checkpoints in order to save a system wide consistent state [16]. This consistent set of checkpoints can then be used to bound rollback propagation. Alternatively, communication induced checkpointing forces each process to take checkpoints based on information piggybacked on the application messages it receives from other processes [50] Checkpoints are taken ....

....messages. Rollback recovery protocols generally assume that the communication network is immune to partitioning but differ in the assumptions they make about network reliability. Some protocols assume that the communication subsystem delivers messages reliably, in First In First Out (FIFO) order [16], while other protocols assume that the communication subsystem can lose, duplicate, or reorder messages [28] The choice between these two assumptions usually affects the complexity of recovery and its implementation in different ways. Generally, assuming a reliable network simplifies the design ....

[Article contains additional citation context not shown here]

O. Babaoglu and K. Marzullo. "Consistent global states of distributed systems: Fundamental concepts and mechanisms." Distributed Systems, Ed. S. Mullender, Addison-Wesley, pp. 55---96, 1993.

....exists an ffl such that for all t and any actors a and b, jC a (t) Gamma C b (t)j ffl. Second, method execution on each actor is atomic and non preemptive. Finally, method execution time is subsumed by communication and scheduling delay. These assumptions are realistic and implementable (cf. [24, 13, 21]) With these assumptions, our real time computational model is simplified in two ways. First, because local clocks are synchronized, a global clock in the system is obtained. Second, because actor invocations are scheduled atomically (non preemptively) the message processing time is predictable ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In Sape Mullender, editor, Distributed Systems, pages 55--94. ACM Press, 1993.

....policy to avoid the above situation. Avoiding Fast Delete A careful examination of Figure 7 shows that the POL message creates a causal dependency between the create message and the delete message (the thick arrows in the figure) Any of the well known techniques for causal delivery of messages [4] will therefore ensure that the create message will not be overcome by the corresponding delete. Larchant RDOSS sends the create, POL, and delete messages using the Isis cbcast primitive. 5 Related Work Multiprocessor and concurrent GC algorithms [2, 10] are not directly applicable, because ....

Ozalp Babaoglu and Keith Marzullo. Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms, chapter 4, pages 55--93. Addison-Wesley, ACM Press, second edition edition, 1993.

....Bouge87] Any stable property can be detected by taking global snapshots periodically. This approach does not work for an unstable predicate which may turn true only between two snapshots and not at the time when the snapshot is taken. An entirely different approach is required for such predicates [WalGar91, CooMar91, GarWal92, SchMat92, GarWal94, BabMar93, TomGar93, HPR93]. We have earlier presented an approach to detect a class of unstable predicates called weak predicates [GarWal94] In this paper, we continue our investigation of detection for a different class of unstable predicates. The reader is referred to [GarWal94] for a discussion of related work and the ....

O. Babaoglu and K. Marzullo, "Consistent global states of distributed systems: fundamental concepts and mechanisms", In S. Mullender, editor, Distributed Systems, 2nd edition, ACM Press, Frontier Series, 1993.

....because the predicate may become true and then false again between two snapshots. An entirely different approach is required for unstable predicates. 8 2.1.2 Unstable Global Predicates There has been much work in observing global states of distributed computations. Babaoglu and Marzullo [3] and Schwartz and Mattern [49] both survey recent work on detecting consistent global states in a distributed system. Recently, Chase and Garg [8] have shown that unstable global predicate detection is an NP complete problem. In that paper they define the property of linearity and show that there ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: fundamental concepts and mechanisms, in Distributed Systems, chapter 4. ACM Press, Frontier Series. (S. J. Mullender Ed.), 1993.

.... dynamicity (new agents appear, old agents disappear) dynamic changing set of agents composing a family (agents enter and retreat into from families) ffl Global state of a distributed system How is it possible to determine the global state of a distributed system, when we know only local states [2, 21] This paper firstly presents primitives for both agents and multi agents systems together with a brief list of theories of agents and mas. It then mentions some uses of category theory in diverse fields of computer science. Finally it clarifies what we intend by agents and families of agents, ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. Technical Report UBLCS93 -1, Laboratory for Computer Science, University of Bologna, 1993.

....a computation: Although script driven probing and fault injection has been applied to several complex communication protocols and distributed applications, we need to develop a formal framework for orchestrating a computation. Past work on global snapshots or global predicate detection [24] is closely related to this problem. Furthermore, we need to identify a minimal set of primitives for injecting faults given a certain failure model. ....

O. Babaoglu and K. Marzullo, "Consistent global states of distributed systems: Fundamental concepts and mechanism," in Distributed Systems, S. Mullender, editor, Addison Wesley, 1993. Second Edition.

....defined on a single cut of the distributed computation. Other researchers have also considered predicates that involve multiple cuts. For example, MC88, GW92] discuss linked predicates, HPR93, BR94] discuss atomic sequences, and [FRGT94] discuss regular patterns. We refer the reader to [BM93, SM94] for surveys of stable and unstable predicate detection. This paper is organized as follows. Section 2 describes our model of execution of a distributed program. We use the notion of a deposet to model an execution. Section 3 proves the intractability of the general problem. Section 4 describes ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: fundamental concepts and mechanisms, in Distributed Systems, chapter 4. ACM Press, Frontier Series. (S.J. Mullender Ed.), 1993.

.... (e) C 2 (e) SC2 For every event e with an immediately succeeding event e 0 on the same process, C 1 (e) C 1 (e 0 ) and C 2 (e) C 2 (e 0 ) 3 Generic Theory of Consistent Global States Predicate detection in asynchronous systems is based on the notion of consistent global states (CGSs) [BM93]. Informally, a global state is consistent if it could have occurred during the computation. Recall that an ideal of a partial order hS; OEi is a set I S such that (8x 2 I : 8y 2 S : y OE x ) y 2 I) Ideals of hEv(c) e hb i are usually called consistent cuts. Recall that for any partial ....

....I : 8y 2 S : y OE x ) y 2 I) Ideals of hEv(c) e hb i are usually called consistent cuts. Recall that for any partial order, the set of its ideals ordered by inclusion ( forms a lattice [DJR93] Furthermore, the lattice of CGSs ordered by G is isomorphic to the lattice of consistent cuts [SM94, BM93]. This isomorphism has an important consequence for detection algorithms; specifically, it implies that a minimal increase with respect to G corresponds to advancing one process by one event, and hence that the lattice of CGSs can be explored by repeatedly advancing one process by one event. ....

[Article contains additional citation context not shown here]

¨ Ozalp Babaoglu and Keith Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In Sape Mullender, editor, Distributed Systems, chapter 5, pages 97--145. Addison Wesley, 2nd edition, 1993.

....for detecting strong conjunctive predicates appear in [GW92] 2. 2 Non global state based approaches A general method for detecting non global state based predicates is to construct the lattice associated with a distributed execution [CM91, DJR93] and then analyze it to detect the behavior [BM93, BR94, JJJR94]. A node of this lattice represents a possible global state 1 of the distributed computation and an edge represents an event that changes the global state of the computation. This approach can detect a very broad class of behaviors, including global state based predicates. The drawback is that ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: fundamental concepts and mechanisms, in Distributed Systems, chapter 4. ACM Press, Frontier Series. (S.J. Mullender Ed.), 1993.

....used to visualize concurrent execution, the state transition model requires a global snapshot of algorithm state. Unfortunately, in distributed environments global snapshots are costly due to distributed state and asynchrony, and may not correspond to any state entered by the underlying execution [4]. Moreover, semantically equivalent execution behavior A Visualization Model for Concurrent Systems 3 may yield different state transitions. As a result, the state transition model is costly to implement and does not effectively abstract over the relevant behavior in distributed systems. In ....

....can be used to reason about global behavior. Thus, we require implementation mechanisms which maintain a representation of the causal order of events and ensure that invoked visualization actions are causally delivered. Causal delivery can be guaranteed by implementing a vector clock protocol [4]. In particular, we wish to guarantee that invocations of visualization actions are causally delivered to the visualization monitor and hence executed in causal order. A Visualization Model for Concurrent Systems 15 Observer Algorithm Component System Message Passing,new, become Message ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In S. Mullender, editor, Distributed Systems, chapter 4, pages 55--96. ACM Press, New York, NY, 1994.

....a global predicate is a fundamental problem in distributed computing. This problem arises in many contexts such as designing, testing and debugging of distributed programs. Previous work has described algorithms for detecting stable and unstable global predicates [2, 3, 5, 7, 6, 10, 11, 13] See [1, 12] for surveys of stable and unstable predicate detection. This paper is most closely related to [7] 6] and [3] Garg and Waldecker [7] present a method for detecting weak conjunctive predicates (WCP) A WCP is a global predicate formed as the conjunction of n local predicates. Many problems in ....

O Babaoglu, and K. Marzullo, "Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms," Distributed Systems , 2nd Edition, editor Sape Mullender, Addison Wesley, New York, NY. 1994, pp. 55-96.

....of the system. This synchronous solution, although very simple, reduces concurrency and interferes with the underlying computation. In contrast, in the Process and Message Model (PMM) consistent snapshots can be constructed asynchronously by a component that passively receives process states [1, 2, 13]. Fischer, Griffeth and Lynch [8] designed a global snapshot algorithm that is tailored for transaction based systems, but Institute of Computing, UNICAMP, Caixa Postal 6176, 13083 970 Campinas, SP. This work has been supported by FAPESP under grant no. 95 1983 8 for Islene Calciolari Garcia ....

....object states read locked by a; ffl LockW (a) object states write locked by a; ffl LockRW (a) LockR(a) LockW (a) ffl Out(a) object states promoted by a. The execution of an application is represented by a diagram similar to the space time diagram used traditionally in the PMM [1, 13] (Figure 2) In the diagram, horizontal lines represent sequential action structures with time flowing from left to right. Parallel horizontal lines represent concurrent action structures, which may fork or join depending on the global 4 I. C. Garcia and L. E. Buzato a oe 0 3 oe 0 1 oe 1 1 ....

[Article contains additional citation context not shown here]

O. Babaoglu. Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms. Technical Report UBLCS-93-1, University of Bologna, 1993.

....[17] but no specific protocols are provided. In [6] an active rules mechanism is proposed for maintaining replication consistency. Finally, we note that in the field of distributed (operating) systems there has been considerable work in the area of snapshots and consistent global states, e.g. [2]. Although this work appears highly related to the problem we address, there are two significant differences: 1) The conditions to be evaluated in the distributed system setting are stable, i.e. once a condition becomes valid, it stays valid. This property is not true of database integrity ....

....drift t D . Then the local clocks can be used instead of a global clock by incorporating t D into our protocols, similarly to the way we have incorporated t Delta to obtain the SRQ protocol from TRQ. In addition, we are investigating whether we can adapt our protocols so that logical clocks [2, 13] are sufficient. 3.3 TRT Protocol The TRQ and SRQ protocols add timestamping to the DRQ protocol in order to achieve accuracy. This approach corresponds to one dimension in the protocol design space. The next dimension we consider is transactions. Note that we are not proposing to add global ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In S. Mullender, editor, Distributed Systems. ACM Press, New York, 1993.

....an alternative approach by restricting the global predicate to one that can be efficiently detected, such as the conjunction and disjunction of local predicates. A well written chapter on monitoring distributed Sec. 18.8 Concluding Remarks 451 computations for asynchronous systems appears in [1]. 18.8 Concluding Remarks This chapter presented a framework for formal specification and monitoring of runtime constraints in real time systems. Although the focus has been on run time monitoring in a multi tasking shared memory environment, most of the concepts presented are directly applicable ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanism. In S. Mullender, editor, Distributed Systems, chapter 4, pages 55--96. Addison-Wesley, Reading, Mass., second edition, 1993.

No context found.

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In Sape Mullender, editor, Distributed Systems, pages 55--96. Addison-Wesley, 1993.

....As an alternative to proving properties a priori, they can be checked concurrently with an actual execution of the program through run time property detection. With this technique, conclusions that are drawn are not about all possible executions of the program, but about all possible observations [SM94, BM93] of an actual execution. The third alternative reasoning about program properties after an execution is called post mortem analysis and is similar to run time property detection. The basic difference is that the analysis has to be based on data collected during the execution (traces) and ....

....i is the global state that results after executing event e i in Sigma i Gamma1 . By construction, each path of the lattice starting at the minimal element and proceeding upwards corresponds to an observation of the computation and each observation corresponds to a path in the lattice [BM93, SM94]. In other words, the lattice of consistent global states represents all possible observations for the computation. Note that, internal to the computation, the actual sequence of global states that is produced cannot be known and this lattice represents the best information that is available. PI ....

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: fundamental concepts and mechanisms, in Distributed Systems, chapter 4, pages 55-- 93. ACM Press, Frontier Series, (S.J. Mullender Ed.), 1993.

No context found.

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In S. J. Mullender, editor, Distributed Systems, pages 55--96. Addison-Wesley, 1993.

No context found.

 O. Babaoglu and K. Marzullo. Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms, chapter 4, pages 55-96. In: S. Mullender (Ed.) - Distributed Systems, Addison-Wesley, 1993.

No context found.

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: Fundamental concepts and mechanisms. In S. J. Mullender, editor, Distributed Systems, pages 55--96. Addison-Wesley, 1993.

No context found.

O. Babaoglu and K. Marzullo. Consistent global states of distributed systems: fundamental concepts and mechanisms. In S. Mullender, editor, Distributed Systems, chapter 4. Addison-Wesley, 1993.

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