Título: Main Memory Databases Instant Recovery

Data: 31/01/2019

Horário: 16:00h

Local: Sala de Seminários - Bloco 952

Resumo:

Similar to traditional disk-based systems, in-memory databases perform recovery activities (logging, checkpointing and restart) to restore a database to a consistent state after the system crash. The in-memory databases recover then-self offline, meaning that database only becomes available to new transactions after the full recovery process. Systems can keep database replicas for high availability. However, hardware improvements are not immune to some sources of failures that can cause multiple and shared failures. Thus, software techniques are necessary to avoid failures and repair failed systems as quickly as possible. This work proposes a instant recovery approach for in-memory databases, i.e. the system can process transactions immediately after the failure. We use identifying hot and cold data to load incrementally into memory the most frequently accessed tuples. Moreover, we can load tuples on demand to meet transactions as soon as they arrive.

Banca:

• Prof. Dr. José Maria da Silva Monteiro Filho (MDCC/UFC - Orientador)
• Prof. Dr. Ângelo Roncalli Alencar Brayner (MDCC/UFC)
• Prof. Dr. João Paulo do Vale Madeiro (Unilab)