Because the solution was on Windows Azure, we wanted to test different configurations and solution. We wanted to see what the performance is if we use 1, 2 and 3 instances of SQL endpoints. Not only this, we had different times of SQL endpoints –Azure SQL (SAS), Virtual Machine with SQL Server and SQL Azure Azure SQL Premium. (SAS but with dedicated resources).
The good part in our scenario was that the data don’t change very often. We could start from the assumption that the database will be updated only one time per day. Because of this we could use different methods to create a load balancing from our SQL instances.
Because the time was very limited and the only think that we needed was to distributed the load on all the machines we decided to replicate the data on all the SQL instances and create a simple load balancer from code. We divided the tick time (timpstamp) with the number of instances and based on this we selected the SQL instance that will be hit. A simple mechanism that can be very easily managed for the PoC. Of course the final solution will be more complicated, but for a PoC this solution was perfect. In 10 minutes we had a load balancer :-).
We checked the load of each machine and we can say that the queries were pretty good distributed. The load of each machine was almost the same (+/-10%)
What others solution we could use?
Master-Slave – In this case we have multiple instances of SQL servers. Each slave can be used for reading operation and the master will be used for writing operations.
Sharding – The solution that we used to resolve our problem
AlwaysOn – The base concept is the same one as for Master-Slave
Transaction Replication – Is based on a snapshot and when data is changing, all the subscribers are notified about the change. You can filter individual database objects and publish changes to subscribers
Peer-to-Peer Replication – Is based on Transaction Replication, but with some new features
Change Tracking – Is a very base and simple tracking mechanisms. The only thing that is notifing you is that a specific row had change. You don’t know any information related to the old or new value
Change Data Capture – Writes in the logs all the changes that are made over a table and are used to track and synchronize the rest of the database instances
We have a lot of solutions on the marker and for SQL Server. Based on our needs we needed to select the best one for us. I was surprised to see that a simple load balancer, written in 5 lines of code can do a pretty good stuff.