Conversa LSD: Automated configuration of distributed storage systems

Olá pessoal,

estamos retomando as conversas lsd nesta semana. Nossa primeira conversa lsd será nessa quarta-feira, 29/02, às 16h, com direito à comemoração dos aniversariantes dos meses de janeiro e fevereiro e boas-vindas aos pesquisadores visitantes (Lauro, Mihai, Nitin e Josef).

O primeiro palestrante será Lauro Beltrão, aluno de doutorado da UBC. O título e resumo da sua palestra seguem abaixo.

Lívia
-------

AUTOMATED CONFIGURATION OF DISTRIBUTED STORAGE SYSTEMS

Distributed storage systems have evolved to employ optimization techniques (e.g., replication, data deduplication, data striping and caching policies) that enable trade-offs over interrelated success metrics such as response time, throughput, storage space, and energy consumption. Setting the configuration parameters of such optimization techniques to best match the success criteria involves exploring a large configuration space to tune the parameters for each of these techniques. Additionally, the values of the configuration parameters are workload-dependent, requiring exploration of the parameters for each different workload. Such exploration makes manually configuring the distributed storage system an undesirable if not unfeasible task.
My research addresses the following main question: How to configure a distributed storage system (i.e., enable/disable various optimization techniques and configure their parameters) to match the application requirements with minimal human intervention?
During my presentation, I will briefly present the motivation to automate the configuration of a distributed storage system and the requirements for an automated configuration solution. I will describe the architecture proposed for anautomated configuration solution and an initial use cases based on data deduplication. Finally, I will briefly present the next steps of my research focusing on the optimization techniques and success metrics that we plan to focus on.