Ginger: A Transactional Middleware with Data and Operation Centric Mixed Consistency

• Autor(a) principal: Rocha, Luís Miguel Dias
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/119342
• topic: Distributed Systems; Consistency; Middleware; Replication; Data Store; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Many modern digital services to correspond to user demand need to offer high availability and low response times. To that end, a lot of digital services resort to geo-replicateddistributed systems. These systems are deployed closer to users, splitting latency acrossmultiple servers and allowing for faster access and communication. However, to accommodate these systems the data stores are also split up across multiple locations. Committing an operation is such systems requires coordination among the multiple replicas.These systems must allow data to be stored as fast as possible without breaking safety constraints of the developers systems.There are three main approaches to define the level of consistency to be guaranteed when accessing the data: over data, over operations or over transactions. The problem with approaches such as consistency over data or consistency over transactions is that they are very limited, as they can result in operations that could be executed in lower consistency levels to be executed at higher consistency levels. Our approach to this problemis the conciliation of executing transactions while expressing consistency in both data and operations. We instantiate this proposition in a middleware system, called Ginger,that is deployed between the user and the data stores. Ginger benefits from all the other approaches, allowing for execution of transactions, that include operations with different levels of consistency, over data with different levels of consistency. This provides the benefits of the isolation from transactions while also providing the performance and control,that consistency defined over operations and consistency defined over data provide.Our experimental results show that Ginger comparing to previously mentioned approaches, such as consistency over data and consistency over transaction, provides faster transaction committing speeds. Ginger serves as proof of concept that using consistency defined both over data and operations while using transactions is possible and may be aviable approach. Further development of the system will provide more functionalities,further evaluation, and a more in-depth comparison to other systems.