Semantics-Driven Optimistic Data Replication: Towards a Framework Supporting Software Architects and Developers

Abstract

Optimistic data replication is an important enabling technology for high availability and scalability, but the disadvantage is that it can only guarantee eventual consistency. Inconsequence, apps might observe stale data and write conflicts. In practice, optimistic data replication is usually part of data base systems that implement detection and resolution of conflicts with purely syntactic techniques. On the downside, this can lead to conflicts that are impossible to detect and resolve correctly by using the database framework only, ultimately resulting in concurrency anomalies. In order to avoid concurrency anomalies, developers are forced to use error-prone workarounds, there by increasing application complexity significantly. An alternative to using existing database replication would be to implement a custom solution tailored to the application semantics that could minimize the occurrence of write conflicts and concurrency anomalies. However, the realization of custom solutions driven by application semantics is very difficult in practice, due to the complexity of concurrency control in distributed systems. In this position paper we demonstrate practical problems related to state of-the-practice optimistic data replication and motivate the need for semantics-driven solutions. Our research aims at simplifying the development of custom solutions by delivering a frame workfor semantics-driven optimistic data replication. This frame work consists of a technical part and accompanying concepts for the design of extended domain models optimized for optimistic data replication and conflict minimization

Tackling Consistency-related Design Challenges of Distributed Data-Intensive Systems

“…Regarding the design of domain objects, trivial aggregate shares between 37.5% and 60% were achieved. 4 We provide the complete, manually anonimized dataset: https://doi.org/10.6084/m9.figshare.14988228 The re-designed model was much clearer and easier to understand H App3 → Complexity…”

“…During inconsistency windows, applications might read stale data or even update the same data concurrently, resulting in concurrency anomalies and conflicts that need to be handled by the application itself. Implementation of custom code for conflict resolution is a huge source of human error [4,31], which is why some researchers have proposed exploiting "semantic tricks" like commutativity of update operations to prevent conflicts from happening in the first place, e.g., [24]. Weikum et al introduced Multilevel Transactions for increased concurrency in relational databases [46].…”