Relational Cloud: A Database-as-a-Service for the Cloud

This CIDR 2011 paper provides a unified view of their ongoing projects. The main vision is to implement a new transactional database-as-a-service (DBaaS), Relational Cloud. The key golas of the system are efficient multi-tenancy, elastic scalability, and database privacy. To achieve these goals, the authors try to utilize workload awareness of the system. While the privacy issue seems to be more or less orthogonal, the paper illustrates how to partition, place, and migrate for the other two goals.

This paper is thought-provoking, because the conventional wisdom has been that the scalability of transactional RDBMS is difficult to achieve. Here are my random thoughts:

• How much scalability does the system try to achieve? If it is just tens of commodity machines, many commercial RDBMSs are already supporting it very well (either within a mainframe or based on commodity servers), and why not just simply use them on the cloud?
• This paper does not say anything about the scalability of the front-end server. In Figure 1 they notate front-end servers, but no further comment is made thereafter.
• Scalability without availability could be useless, especially with mission-critical storages. There is no mention about fault tolerance in this paper.
• Is multi-tenancy really desired? If the main goal of multi-tenancy is to consolidate lots of nearly inactive databases from tenants (independent billing entities) into a single server, I would rather happily pay for N+1 dedicated servers than paying for N multi-tenanted servers.