Publicação
Convergent types for shared memory
| Resumo: | It is well-known that consistency in shared memory concurrent programming comes with the price of degrading performance and scalability. Some of the existing solutions to this problem end up with high-level complexity and are not programmer friendly. We present a simple and well-defined approach to obtain relevant results for shared memory environments through relaxing synchronization. For that, we will look into Mergeable Data Types, data structures analogous to Conflict-Free Replicated Data Types but designed to perform in shared memory. CRDTs were the first formal approach engaging a solid theoretical study about eventual consistency on distributed systems, answering the CAP Theorem problem and providing high-availability. With CRDTs, updates are unsynchronized, and replicas eventually converge to a correct common state. However, CRDTs are not designed to perform in shared memory. In large-scale distributed systems the merge cost is negligible when compared to network mediated synchronization. Therefore, we have migrated the concept by developing the already existent Mergeable Data Types through formally defining a programming model that we named Global-Local View. Furthermore, we have created a portfolio of MDTs and demonstrated that in the appropriated scenarios we can largely benefit from the model. |
|---|---|
| Autores principais: | Brandão, José Pedro Maia |
| Assunto: | Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática |
| Ano: | 2019 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | It is well-known that consistency in shared memory concurrent programming comes with the price of degrading performance and scalability. Some of the existing solutions to this problem end up with high-level complexity and are not programmer friendly. We present a simple and well-defined approach to obtain relevant results for shared memory environments through relaxing synchronization. For that, we will look into Mergeable Data Types, data structures analogous to Conflict-Free Replicated Data Types but designed to perform in shared memory. CRDTs were the first formal approach engaging a solid theoretical study about eventual consistency on distributed systems, answering the CAP Theorem problem and providing high-availability. With CRDTs, updates are unsynchronized, and replicas eventually converge to a correct common state. However, CRDTs are not designed to perform in shared memory. In large-scale distributed systems the merge cost is negligible when compared to network mediated synchronization. Therefore, we have migrated the concept by developing the already existent Mergeable Data Types through formally defining a programming model that we named Global-Local View. Furthermore, we have created a portfolio of MDTs and demonstrated that in the appropriated scenarios we can largely benefit from the model. |
|---|