Réplication (informatique) | EPFL Graph Search

En informatique, la réplication est un processus de partage d'informations pour assurer la cohérence de données entre plusieurs sources de données redondantes, pour améliorer la fiabilité, la tolérance aux pannes, ou la disponibilité. On parle de réplication de données si les mêmes données sont dupliquées sur plusieurs périphériques. La réplication n'est pas à confondre avec une sauvegarde : les données sauvegardées ne changent pas dans le temps, reflétant un état fixe des données, tandis que les données répliquées évoluent sans cesse à mesure que les données sources changent. Réplication active/passive On distingue couramment la réplication passive et active. Lors de réplication active, les calculs effectués par la source (ou maître) sont répliqués, alors que lors de réplication passive, seul le serveur maître procède au calcul et il ne propage que les modifications finales de la mémoire à effectuer. Si à tout moment un unique maître est désigné pour effectuer toutes

Replication of non-deterministic objects

This thesis discusses replication of non-deterministic objects in distributed systems to achieve fault tolerance against crash failures. The objects replicated are the virtual nodes of a distributed application. Replication is viewed as an issue that is to be dealt with only during the configuration of a distributed application and that should not affect the development of the application. Hence, replication of virtual nodes should be transparent to the application. Like all measures to achieve fault tolerance, replication introduces redundancy in the system. Not surprisingly, the main difficulty is guaranteeing the consistency of all replicas such that they behave in the same way as if the object was not replicated (replication transparency). This is further complicated if active objects (like virtual nodes) are replicated, and these objects themselves can be clients of still further objects in the distributed application. The problems of replication of active non-deterministic objects are analyzed in the context of distributed Ada 95 applications. The ISO standard for Ada 95 defines a model for distributed execution based on remote procedure calls (RPC). Virtual nodes in Ada 95 use this as their sole communication paradigm, but they may contain tasks to execute activities concurrently, thus making the execution potentially non-deterministic due to implicit timing dependencies. Such non-determinism cannot be avoided by choosing deterministic tasking policies. I present two different approaches to maintain replica consistency despite this non-determinism. In a first approach, I consider the run-time support of Ada 95 as a black box (except for the part handling remote communications). This corresponds to a non-deterministic computation model. I show that replication of non-deterministic virtual nodes requires that remote procedure calls are implemented as nested transactions. Unfortunately, effects of failures are not local to the replicas of a virtual node: when a failure occurs, nested remote calls made to other virtual nodes must be undone. Also, using transactional semantics for RPCs necessitates a compromise regarding transparency: the application must identify global state for it cannot be determined reliably in an automatic way. Further study reveals that this approach cannot be implemented in a transparent way at all because the consistency criterion of Ada 95 (linearizability) is much weaker than that of transactions (serializability). An execution of remote procedure calls as transactions may thus lead to incompatibilities with the semantics of the programming language. If remotely called subprograms on a replicated virtual node perform partial operations, i.e., entry calls on global protected objects, deadlocks that cannot be broken can occur in certain cases. Such deadlocks do not occur when the virtual node is not replicated. The transactional semantics of RPCs must therefore be exposed to the application. A second approach is based on a piecewise deterministic computation model, i.e., the execution of a virtual node is seen as a sequence of deterministic state intervals. Whenever a non-deterministic event occurs, a new state interval is started. I study replica organization under this computation model (semi-active replication). In this model, all non-deterministic decisions are made on one distinguished replica (the leader), while all other replicas (the followers) are forced to follow the same sequence of non-deterministic events. I show that it suffices to synchronize the followers with the leader upon each observable event, i.e., when the leader sends a message to some other virtual node. It is not necessary to synchronize upon each and every non-deterministic event — which would incur a prohibitively high overhead. Non-deterministic events occurring on the leader between observable events are logged and sent to the followers just before the leader executes an observable event. Consequently, it is guaranteed that the followers will reach the same state as the leader, and thus the effects of failures remain mostly local to the replicas. A prototype implementation called RAPIDS (Replicated Ada Partitions In Distributed Systems) serves as a proof of concept for this second approach, demonstrating its feasibility. RAPIDS is an Ada 95 implementation of a replication manager for semi-active replication for the GNAT development system for Ada 95. It is entirely contained within the run-time support and hence largely transparent for the application.

EPFL1998

Agreement-related problems

Xavier Défago

Agreement problems constitute a fundamental class of problems in the context of distributed systems. All agreement problems follow a common pattern: all processes must agree on some common decision, the nature of which depends on the specific problem. This dissertation mainly focuses on three important agreements problems: Replication, Total Order Broadcast, and Consensus. Replication is a common means to introduce redundancy in a system, in order to improve its availability. A replicated server is a server that is composed of multiple copies so that, if one copy fails, the other copies can still provide the service. Each copy of the server is called a replica. The replicas must all evolve in manner that is consistent with the other replicas. Hence, updating the replicated server requires that every replica agrees on the set of modifications to carry over. There are two principal replication schemes to ensure this consistency: active replication and passive replication. In Total Order Broadcast, processes broadcast messages to all processes. However, all messages must be delivered in the same order. Also, if one process delivers a message m, then all correct processes must eventually deliver m. The problem of Consensus gives an abstraction to most other agreement problems. All processes initiate a Consensus by proposing a value. Then, all processes must eventually decide the same value v that must be one of the proposed values. These agreement problems are closely related to each other. For instance, Chandra and Toueg [CT96] show that Total Order Broadcast and Consensus are equivalent problems. In addition, Lamport [Lam78] and Schneider [Sch90] show that active replication needs Total Order Broadcast. As a result, active replication is also closely related to the Consensus problem. The first contribution of this dissertation is the definition of the semi-passive replication technique. Semi-passive replication is a passive replication scheme based on a variant of Consensus (called Lazy Consensus and also defined here). From a conceptual point of view, the result is important as it helps to clarify the relation between passive replication and the Consensus problem. In practice, this makes it possible to design systems that react more quickly to failures. The problem of Total Order Broadcast is well-known in the field of distributed systems and algorithms. In fact, there have been already more than fifty algorithms published on the problem so far. Although quite similar, it is difficult to compare these algorithms as they often differ with respect to their actual properties, assumptions, and objectives. The second main contribution of this dissertation is to define five classes of total order broadcast algorithms, and to relate existing algorithms to those classes. The third contribution of this dissertation is to compare the expected performance of the various classes of total order broadcast algorithms. To achieve this goal, we define a set of metrics to predict the performance of distributed algorithms.

EPFL2000

Group communications and database replication

Matthias Wiesmann

Databases are an important part of today's IT infrastructure: both companies and state institutions rely on database systems to store most of their important data. As we are more and more dependent on database systems, securing this key facility is now a priority. Because of this, research on fault-tolerant database systems is of increasing importance. One way to ensure the fault-tolerance of a system is by replicating it. Replication is a natural way to deal with failures: if one copy is not available, we use another one. However implementing consistent replication is not easy. Database replication is hardly a new area of research: the first papers on the subject are more than twenty years old. Yet how to build an efficient, consistent replicated database is still an open research question. Recently, a new approach to solve this problem has been proposed. The idea is to rely on some communication infrastructure called group communications. This infrastructure offers some high-level primitives that can help in the design and the implementation of a replicated database. While promising, this approach to database replication is still in its infancy. This thesis focuses on group communication-based database replication and strives to give an overall understanding of this topic. This thesis has three major contributions. In the structural domain, it introduces a classification of replication techniques. In the qualitative domain, an analysis of fault-tolerance semantics is proposed. Finally, in the quantitative domain, a performance evaluation of group communication-based database replication is presented. The classification gives an overview of the different means to implement database replication. Techniques described in the literature are sorted using this classification. The classification highlights structural similarities of techniques originating from different communities (database community and distributed system community). For each category of the classification, we also analyse the requirements imposed on the database component and group communication primitives that are needed to enforce consistency. Group communication-based database replication implies building a system from two different components: a database system and a group communication system. Fault-tolerance is an end-to-end property: a system built from two components tends to be as fault-tolerant as the weakest component. The analysis of fault-tolerance semantics show what fault-tolerance guarantee is ensured by group communication based replication techniques. Additionally a new faulttolerance guarantee, group-safety, is proposed. Group-safety is better suited to group communication-based database replication. We also show that group-safe replication techniques can offer improved performance. Finally, the performance evaluation offers a quantitative view of group communication based replication techniques. The performance of group communication techniques and classical database replication techniques is compared. The way those different techniques react to different loads is explored. Some optimisation of group communication techniques are also described and their performance benefits evaluated.

EPFL2002