Adaptive data management for in-memory database clusters |
Demand for high performance and high availability combined with
plummeting hardware prices has led to the widespread emergence of large
computing clusters. Such environments offer great potential for highly
efficient and fault-tolerant data management systems, but realizing this
vision requires revisiting several fundamental assumptions about
current systems. Web applications have become commonplace in most online businesses. These systems require fast interactive response times and the ability to serve a large number of clients uninterruptedly. At the core of a web application lies a database system responsible for storing the application state. As systems grow in number of clients, the database invariably becomes a performance bottleneck. Current solutions to increase the performance and the availability of database systems usually rely on specialized hardware and only scale to a few dozen servers. Sprint aggregates servers in a cluster and makes use of high performance techniques to overcome the latency limitations of traditional on-disk databases. Data partitioning and replication are combined for performance and high availability. Each server runs a local in-memory database system (IMDB). This means that the whole database state resides in the main memory of servers, although no single server is expected to contain the whole database image. Sprint will contribute to research on the design and implementation of future highly efficient and available adaptive data-management systems. It will revisit traditional data management algorithms from the perspective of emerging cluster technologies and seek to understand how their performance is affected by these technologies. Project leaflet (pdf). Also available in Italian (pdf). |
Traditional DB systems Sprint: cluster of IMDBs |
Project members: |
Fernando Pedone (Associate Professsor) |
Lásaro Camargos (PhD student) | |
Rodrigo Schmidt (PhD student) | |
Marcin Wieloch (PhD student) |
For further details: | Fernando Pedone |
Faculty of Informatics | |
University of Lugano | |
CH-6904 Switzerland |
|
E-mail: firstname.lastname AT unisi.ch |