Jean-Michel Hélary

A communication-induced checkpointing protocol that ensures rollback-dependency trackability

Considering an application in which processes take local checkpoints independently (called basic checkpoints), this paper develops a protocol that forces them to take some additional local checkpoints (called forced checkpoints) in order that the resulting checkpoint and communication pattern satisfies the Rollback Dependency Trackability (RDT) property. This property states that all dependencies between local checkpoints are on-line trackable by using a transitive dependency vector. Compared to other protocols ensuring the RDT property, the proposed protocol is less conservative in the sense that it takes less additional local checkpoints. It attains this goal by a subtle tracking of causal dependencies on already taken checkpoints; this tracking is then used to prevent the occurrence of hidden dependencies. As indicated by simulation study, the proposed protocol compares favorably with other protocols; moreover it additionally associates on-the-fly with each local checkpoint C the minimum global checkpoint to which C belongs.

Computing global functions in asynchronous distributed systems with perfect failure detectors

A Global Data is a vector with one entry per process. Each entry must be filled with an appropriate value provided by the corresponding process. Several distributed computing problems amount to compute a function on a global data. This paper proposes a protocol to solve such problems in the context of asynchronous distributed systems where processes may fail by crashing. The main problem that has to be solved lies in computing the global data and in providing each noncrashed process with a copy of it, despite the possible crash of some processes. To be consistent, the global data must contain, at least, all the values provided by the processes that do not crash. This defines the Global Data Computation (GDC) problem. To solve this problem, processes execute a sequence of asynchronous rounds during which they construct, in a decentralized way, the value of the global data and eventually each process gets a copy of it. To cope with process crashes, the protocol uses a perfect failure detector. The proposed protocol has been designed to be time efficient: it allows early decision. Let t be the maximum number of processes that may crash, t<n where n is the total number of processes, and f be the actual number of process crashes (f/spl les/t). In the worst case, the protocol terminates in min(2f+2,t+1) rounds. Moreover, the protocol does not require processes to exchange information on their perception of crashes. The message size depends only on the size of the global data.

Deadlock Models and a General Algorithm for Distributed Deadlock Detection

This paper deals with the problem of deadlock detection in asynchronous message passing systems in a system model that covers unspecified receptions and non-FIFO channels. It presents a hierarchy of deadlock models and deadlock detection problems. It abstracts deadlocks by a general deadlock model that has the same modeling power as the OR-AND model; however, it has much concise expressive power. An abstract general definition of deadlocks in distributed systems is presented that defines deadlocks independently of the underlying deadlock model. This formulation can be used to design a single distributed deadlock detection algorithm which uniformly addresses all deadlocks in the context of various request models such as AND, OR, AND-OR, and k-out-of-n requests. A simple generalized deadlock detection algorithm that uses a circulating token is presented to illustrate the concept. The algorithm is formally described and proven correct. Moreover, possible refinements of the basic solution concerning improvements of token routing and parallel implementation are outlined and evaluated. Extensions to individual and global termination issues are also addressed. Since the proposed deadlock detection algorithm is designed around the abstract definition of deadlocks, it has some very favorable features.

A general scheme for token- and tree-based distributed mutual exclusion algorithms

In a distributed context, mutual exclusion algorithms can be divided into two families according to their underlying algorithmic principles: those that are permission-based and those that are token-based. Within the latter family, a lot of algorithms use a rooted tree structure to move the requests and the unique token. This paper presents a very general information structure (and the associated generic algorithm) for token- and tree-based mutual exclusion algorithms. This general structure not only covers, as particular cases, several known algorithms, but also allows for the design of new ones that are well suited for various topology requirements. >

A distributed algorithm for mutual exclusion in an arbitrary network

A distributed algorithm for mutual exclusion is presented. No particular assumptions on the network topology are required, except connectivity; the communication graph may be arbitrary. The processes communicate by using messages only and there is no global controller. Furthermore, no process needs to know or learn the global network topology. In that sense, the algorithm is more general than the mutual exclusion algorithms which make use of an a priori knowledge of the network topology (for example either ring or complete network). A proof of the correctness of the algorithm is provided. The algorithms complexity is examined by evaluating the number of messages required for the mutual exclusion protocol.

Virtual Precedence in Asynchronous Systems: Cencept and Applications

This paper introduces the Virtual Precedence (VP) property. An interval-based abstraction of a computation satisfies the VP property if it is possible to timestamp its intervals in a consistent way (i.e., time does not decrease inside a process and increases after communication). A very general protocol P that builds abstractions satisfying the VP property is proposed. It is shown that the VP property encompasses logical clocks systems and communication-induced checkpointing protocols. A new and efficient protocol which ensures no local checkpoint is useless is derived from P. This protocol compares very favorably with existing protocols that solve the same problem. This shows that, due the generality of its approach, a theory (namely, here VP) can give efficient solutions to practical problems (here the prevention of useless checkpoints).

Communication-induced determination of consistent snapshots

A classical way to determine consistent snapshots consists in using Chandy-Lamports (1985) algorithm. This algorithm relies on specific control messages that allow processes to synchronize local checkpoint determination and message recording in order that the resulting snapshot is consistent. This paper investigates a communication-induced approach to determine consistent snapshots. In such an approach, control information is carried by application messages. Two abstract necessary and sufficient conditions are stated: one associated with global checkpoint consistency, the other associated with message recording. A general protocol is derived from these abstract conditions. Actually, this general protocol can be instantiated in distinct ways, giving rise to a family of communication-induced snapshot protocols. This general protocol shows there is an intrinsic tradeoff between the number of forced checkpoints and the number of recorded messages. Finally, a particular instantiation of the general protocol is provided.

Early stopping in Global Data Computation

The Global Data Computation problem consists of providing each process with the same vector (with one entry per process) such that each entry is filled by a value provided by the corresponding process. This paper presents a protocol that solves this problem in an asynchronous distributed system where processes can crash, but equipped with a perfect failure detector. This protocol requires that processes execute asynchronous computation rounds. The number of rounds is upper bounded by min(f+2, t+1, n), where n, t, and f represent the total number of processes, the maximum number of processes that can crash, and the number of processes that actually crash, respectively. This value is a lower bound for the number of rounds when t<n-1. To our knowledge, this protocol is the first to achieve this lower bound. Interestingly, this protocol meets the same lower bound as the one required in synchronous systems.

Towards the construction of distributed detection programs, with an application to distributed termination

SummaryMethodological design of distributed programs is necessary if one is to master the complexity of parallelism. The class of control programs, whose purpose is to observe or detect properties of an underlying program, plays an important role in distributed computing. The detection of a property generally rests upon consistent evaluations of a predicate; such a predicate can be global, i.e. involve states of several processes and channels of the observed program. Unfortunately, in a distributed system, the consistency of an evaluation cannot be trivially obtained. This is a central problem in distributed evaluations. This paper addresses the problem of distributed evaluation, used as a basic tool for solution of general distributed detection problems. A new evaluation paradigm is put forward, and a general distributed detection program is designed, introducing the iterative scheme ofguarded waves sequence. The case of distributed termination detection is then taken to illustrate the proposed methodological design.

Interval Consistency of Asynchronous Distributed Computations

An interval of a sequential process is a sequence of consecutive events of this process. The set of intervals defined on a distributed computation defines an abstraction of this distributed computation, and the traditional causality relation on events induces a relation on the set of intervals that we call I-precedence. An important question is then, “Is the interval-based abstraction associated with a distributed computation consistent?” To answer this question, this paper introduces a consistency criterion named interval consistency (IC). Intuitively, this criterion states that an interval-based abstraction of a distributed computation is consistent if its I-precedence relation does not contradict the sequentiality of each process. More formally, IC is defined as a property of a precedence graph. Interestingly, the IC criterion can be operationally characterized in terms of timestamps (whose values belong to a lattice). The paper uses this characterization to design a versatile protocol that, given intervals defined by a daemon whose behavior is unpredictable, breaks them (in a nontrivial manner) in order to produce an abstraction satisfying the IC criterion. Applications to communication-induced checkpointing are suggested.