Area : Scalability and Virtual Full Replication

To achieve better scalability in a fully replicated distributed main-memory database, we propose support for virtual full replication [6]. Full replication is often used to improve availability and predictability in distributed real-time applications. In a fully replicated database all updates are sent to all nodes regardless of where the data is going to be used there or not. In a fully replicated database with asynchronous replication, availability and predictability is improved at the cost of consistency. By allowing relaxed consistency and avoiding synchronous replication, a number of predictability problems connected to synchronization are avoided, such as distributed locking of objects and reliance on stable communication during transaction execution. Since nodes are autonomous during transaction execution, there is no need for full connectivity between nodes at all times. This system scales better, since nodes allow both simultaneous queries and simultaneous updates for different object replicas, without a distributed lock mechanism that involves communication between nodes where object replicas are allocated. This also improves availability even further, although it requires conflict detection and resolution of concurrent updates. Virtual full replication is a concept that improves scalability even further without changing the application’s assumption of having access to a fully replicated database. We support virtual full replication by segmenting the database and allowing segments to have individual degrees of replication, avoiding excess replication and excess main-memory usage. Partial replication is reasonable, since many applications actually use only a subset of the replicated data. We combine virtual full replication with asynchronous replication to achieve scalability. Our work focuses on scalability of replication effort and storage requirements, in terms of number of nodes in the system and the number of data objects in the database, while maintaining application’s image of full replication. Improving scalability allows more nodes to be added, the database size to be increased or deadlines to be tightened, without saturating the network or requiring excessive amount of main memory. We aim to show that predictable and scalable replication effort and storage requirement can be achieved, still supporting the same degree of fault-tolerance and data availability as that of a fully replicated database. We have developed concepts and solutions for central issues in segmentation [17]. Currently we focus on segmentation as a principle for improving scalability but also suggest further steps in the exploration of segments [18], like scheduled and prioritized replication of updates in distributed real-time databases [16], dynamic segment allocation, and segment allocation policies based on other application criteria than availability. Dynamic segment allocation will allow dynamic adjustment for where segments are allocated. The need for data is typically changing during the system of the system, following changes on operating modes or user input or requirements. Typical applications are simulators, where actors move and use different data at different nodes, depending on what actions the actor take. Another application area in within mobile communication, where terminals