Publicação

Database replication for enterprise applications

Ver documento

Detalhes bibliográficos
Resumo:A common pattern for enterprise applications, particularly in small and medium businesses, is the reliance on an integrated traditional relational database system that provides persistence and where the relational aspect underlies the core logic of the application. While several solutions are proposed for scaling out such applications, database replication is key if the relational aspect is to be preserved. However, it is worrisome that because proposed solutions for database replication have been evaluated using simple synthetic benchmarks, their applicability to enterprise applications is not straightforward: the performance of conservative solutions hinges on the ability to conveniently partition applications while optimistic solutions may experience unacceptable abort rates, compromising fairness, particularly considering long-running transactions. In this thesis, we address these challenges. First, by performing a detailed evaluation of the applicability of database replication protocols based on conservative concurrency control to enterprise applications. Results invalidate the common assumption that real-world databases can be easily partitioned. Then, we tackle the issue of unacceptable abort rates in optimistic solutions by proposing a novel transaction scheduler, AJITTS, which uses an adaptive mechanism that by reaching and maintaining the optimal level of concurrency in the system, minimizes aborts and improves throughput.
Autores principais:Alonso, Ana Luísa Parreira Nunes
Assunto:Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática
Ano:2017
País:Portugal
Tipo de documento:tese de doutoramento
Tipo de acesso:acesso aberto
Instituição associada:Universidade do Minho
Idioma:português
Origem:RepositóriUM - Universidade do Minho
Descrição
Resumo:A common pattern for enterprise applications, particularly in small and medium businesses, is the reliance on an integrated traditional relational database system that provides persistence and where the relational aspect underlies the core logic of the application. While several solutions are proposed for scaling out such applications, database replication is key if the relational aspect is to be preserved. However, it is worrisome that because proposed solutions for database replication have been evaluated using simple synthetic benchmarks, their applicability to enterprise applications is not straightforward: the performance of conservative solutions hinges on the ability to conveniently partition applications while optimistic solutions may experience unacceptable abort rates, compromising fairness, particularly considering long-running transactions. In this thesis, we address these challenges. First, by performing a detailed evaluation of the applicability of database replication protocols based on conservative concurrency control to enterprise applications. Results invalidate the common assumption that real-world databases can be easily partitioned. Then, we tackle the issue of unacceptable abort rates in optimistic solutions by proposing a novel transaction scheduler, AJITTS, which uses an adaptive mechanism that by reaching and maintaining the optimal level of concurrency in the system, minimizes aborts and improves throughput.