Deducing Operation Commutativity from Replicated Data Declaration

• Autor(a) principal: Fragoso, André Gonçalves
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/93771
• topic: replicated systems; consistency levels; data-centric concurrency control; language interpretation; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Distributed systems often resort to data replication not only to enhance their availability but also to reduce user-perceived latency by balancing the load between replicas and routing their requests accordingly. The choice of which consistency level that should be adopted by these replicated systems is critical for the fulfilment of their performance and correctness requirements. However, defining a strategy that strikes the right balance between these concerns in this type of environments is far from being a trivial task due to the related overheads that are amplified in distributed scenarios. Recognising the tension between latency and consistency, many systems allow multiple consistency levels to coexist. Nevertheless, the performance fine-tuning mechanisms supported by the existing hybrid solutions place a high burden on the programmer since the necessary input can be somehow complex requiring him to understand the semantics of each operation of the service he is developing in order to correctly instruct the system on how to handle concurrent updates. Thus, specifying operation dependencies, orderings and invariants to be preserved or even picking the right consistency level to be assigned to a certain data item is, generally, an error-prone task that hinders reasoning. To overcome this adversity, this work aims to reduce the effort spent by the programmer by only requiring the latter to introduce a simple and intuitive input at data declaration. Following this approach, reasoning is centralised and all accesses to replicated data are identified automatically. With all data accesses identified, it is then possible to deduce the side effects of each operation and determine, for each one of them, those with which it conflicts. In this context, this thesis also presents a compile-time analysis applied to the Java language able to evaluate operation pairwise commutativity from the input given at data declaration.