SQL Azure as a Self-Managing Database Service: Lessons Learned and Challenges Ahead

When SQL Azure was released in August 2009, it was the first database service of its kind along multiple axes, compared to other Cloud services: shared nothing architecture and log-based replication; support for full ACID properties; providing consistency and high availability; and by offering near 100% compatibility with on-premise SQL Server delivered a familiar programming model at cloud scale. Today, just over two years later, the service has grown to span six hosting regions across three continents; hosting large numbers of databases (in the order 100s of thousands), increasing more than 5x each year with 10s of thousands of subscribers. It is a very busy service, clocking more than 30 million successful logins over a 24 hour period. In this paper we reflect on the lessons learned, and the challenges we will need to face in future, in order to take the SQL Azure service to the next level of scale, performance, satisfaction for the end user, and profitability for the service provider.