Alfrânio Correia

GORDA: An Open Architecture for Database Replication

Database replication has been a common feature in database management systems (DBMSs) for a long time. In particular, asynchronous or lazy propagation of updates provides a simple yet efficient way of increasing performance and data availability and is widely available across the DBMS product spectrum. High end systems additionally offer sophisticated conflict resolution and data propagation options as well as, synchronous replication based on distributed locking and two-phase commit protocols. This paper presents GORDA architecture and programming interface (GAPI), that enables different replication strategies to be implemented once and deployed in multiple DBMSs. This is achieved by proposing a reflective interface to transaction processing instead of relying on-client interfaces or ad-hoc server extensions. The proposed approach is thus cost-effective, in enabling reuse of replication protocols or components in multiple DBMSs, as well as potentially efficient, as it allows close coupling with DBMS internals.

Testing the dependability and performance of group communication based database replication protocols

Database replication based on group communication systems has recently been proposed as an efficient and resilient solution for large-scale data management. However, its evaluation has been conducted either on simplistic simulation models, which fail to assess concrete implementations, or on complete system implementations, which are costly to test with realistic large-scale scenarios. This paper presents a tool that combines implementations of replication and communication protocols under study with simulated network, database engine, and traffic generator models. Replication components can therefore be subjected to realistic large scale loads in a variety of scenarios, including fault-injection, while at the same time providing global observation and control. The paper shows first how the model is configured and validated to closely reproduce the behavior of a real system, and then how it is applied, allowing us to derive interesting conclusions both on replication and communication protocols and on their implementations.

Revisiting 1-copy equivalence in clustered databases

Recently renewed interest in scalable database systems for shared nothing clusters has been supported by replication protocols based on group communication that are aimed at seamlessly extending the native consistency criteria of centralized database management systems. By using a read-one/write-all-available approach and avoiding the fine-grained synchronization associated with traditional distributed locking, one needs just a single distributed interaction step for each update transaction. Therefore the system can easily be scaled to a large number of replicas, especially, with read intensive loads typical of Web server support environments.In this paper we point out that 1-copy equivalence for causal consistency, which is subsumed by both serializability and snapshot isolation criteria, depends on basic session guarantees that are costly to ensure in clusters, especially in a multi-tier environment. We then point out a simple solution that guarantees causal consistency in the Database State Machine protocol and evaluate its performance, thus highlighting the cost of seamlessly providing common consistency criteria of centralized databases in a clustered environment.

AKARA: A Flexible Clustering Protocol for Demanding Transactional Workloads

Shared-nothing clusters are a well known and cost-effective approach to database server scalability, in particular, with highly intensive read-only workloads typical of many 3-tier web-based applications. The common reliance on a centralized component and a simplistic propagation strategy employed by mainstream solutions however conduct to poor scalability with traditional on-line transaction processing (OLTP), where the update ratio is high. Such approaches also pose an additional obstacle to high availability while introducing a single point of failure. More recently, database replication protocols based on group communication have been shown to overcome such limitations, expanding the applicability of shared-nothing clusters to more demanding transactional workloads. These take simultaneous advantage of total order multicast and transactional semantics to improve on mainstream solutions. However, none has already been widely deployed in a general purpose database management system. In this paper, we argue that a major hurdle for their acceptance is that these proposals have disappointing performance with specific subsets of real-world workloads. Such limitations are deep-rooted and working around them requires in-depth understanding of protocols and changes to applications. We address this issue with a novel protocol that combines multiple transaction execution mechanisms and replication techniques and then show how it avoids the identified pitfalls. Experimental results are obtained with a workload based on the industry standard TPC-C benchmark.

Group-Based replication of on-line transaction processing servers

Several techniques for database replication using group communication have recently been proposed, namely, the Database State Machine, Postgres-R, and the NODO protocol. Although all rely on a totally ordered multicast for consistency, they differ substantially on how multicast is used. This results in different performance trade-offs which are hard to compare as each protocol is presented using a different load scenario and evaluation method. In this paper we evaluate the suitability of such protocols for replication of On-Line Transaction Processing (OLTP) applications in clusters of servers and over wide area networks. This is achieved by implementing them using a common infra-structure and by using a standard workload. The results allows us to select the best protocol regarding performance and scalability in a demanding but realistic usage scenario.

On the use of a reflective architecture to augment Database Management Systems

The Database Management System (DBMS) used to be a commodity soft- ware component, with well known standard interfaces and semantics. However, the performance and reliability expectations being placed on DBMSs have increased the demand for a variety add-ons, that augment the functionality of the database in a wide range of deployment scenarios, offering support for features such as clustering, replica- tion, and self-management, among others. A well known software engineering approach to systems with such requirements is reflection. Unfortunately, standard reflective in- terfaces in DBMSs are very limited. Some of these limitations may be circumvented by implementing reflective features as a wrapper to the DBMS server. Unfortunately, these solutions comes at the expense of a large development effort and significant per- formance penalty. In this paper we propose a general purpose DBMS reflection architecture and interface, that supports multiple extensions while, at the same time, admitting efficient imple- mentations. We illustrate the usefulness of our proposal with concrete examples, and evaluate its cost and performance under different implementation strategies.

Serpentine: adaptive middleware for complex heterogeneous distributed systems

Adaptation of system parameters is acknowledged as a requirement to scalable and dependable distributed systems. Unfortunately, adaptation cannot be effective when provided solely by individual system components as the correct decision is often tied to the composition itself and the system as a whole. In fact, proper adaption is a cross-cutting issue: Diagnostic and feedback operations must target multiple components and do it at different abstraction levels. We address this problem with the SERPENTINE middleware platform. By relying on the industry standard JMX as a service interface, it can monitor and operate on a wide range of distributed middleware and application components. By building on a JMX-enabled OSGi runtime, SERPENTINE is able to control the life-cycle of components themselves. The scriptable stateless server and cascading architecture allow for increased dependability and flexibility.

Practical database replication

This chapter illustrates how the concepts and algorithms described earlier in this book can be used to build practical database replication systems. This is achieved first by addressing architectural challenges on how required functionality is provided by generally available software componentes and then how different components can be efficiently integrated. A second set of practical challenges arises from experience on how performance assumptions map to actual environments and real workloads. The result is a generic architecture for replicated database management systems, focusing on the interfaces between key components, and then on how different algorithmic and practical optimization options map to real world gains. This shows how consistent database replication is achievable in the current state of the art.

A Pragmatic Protocol for Database Replication in Interconnected Clusters

Multi-master update everywhere database replication, as achieved by protocols based on group communication such as DBSM and Postgres-R, addresses both performance and availability. By scaling it to wide area networks, one could save costly bandwidth and avoid large round-trips to a distant master server. Also, by ensuring that updates are safely stored at a remote site within transaction boundaries, disaster recovery is guaranteed. Unfortunately, scaling existing cluster based replication protocols is troublesome. In this paper we present a database replication protocol based on group communication that targets interconnected clusters. In contrast with previous proposals, it uses a separate multicast group for each cluster and thus does not impose any additional requirements on group communication, easing implementation and deployment in a real setting. Nonetheless, the protocol ensures one-copy equivalence while allowing all sites to execute update transactions. Experimental evaluation using the workload of the industry standard TPC-C benchmark confirms the advantages of the approach

Evaluating certification protocols in the partial database state machine

Partial replication is an alluring technique to ensure the reliability of very large and geographically distributed databases while, at the same time, offering good performance. By correctly exploiting access locality most transactions become confined to a small subset of the database replicas thus reducing processing, storage access and communication overhead associated with replication. The advantages of partial replication have however to be weighted against the added complexity that is required to manage it. In fact, if the chosen replica configuration prevents the local execution of transactions or if the overhead of consistency protocols offsets the savings of locality, potential gains cannot be realized. These issues are heavily dependent on the application used for evaluation and render simplistic benchmarks useless. In this paper, we present a detailed analysis of partial database state machine (PDBSM) replication by comparing alternative partial replication protocols with full replication. This is done using a realistic scenario based on a detailed network simulator and access patterns from an industry standard database benchmark. The results obtained allow us to identify the best configuration for typical online transaction processing applications.