Last week, I played with it as part of a load test project, and I was amazed by its capability even though I presented these features individually many times before.
My test setting is a very typical LOB web application. ASP.NET web tier, SQL Database, plus IE or Visual Studio load test client running on laptop. I tried many different deployment options, and compared the performance differences. It's very easy to switch from one deployment to another. Just a few clicks on the Azure management portal or simply republish the project from Visual Studio. Here are my learning.
Overall design:
Deployment Options:
Website: Azure Website free tier, Azure Website reserved instances, Web Role instances, multiple number of instances
SQL: Azure SQL Database, SQL Server in Azure VM, multiple sizes of VMs
Load test client: laptop, Azure IaaS VM
Network: Virtual Network, or not
Performance findings:
- Azure website free tier, performance quota will soon be blown, and request will be rejected.
- Azure website reserved instance, and Web role instance. Performance is comparable between these two options. Both can scale pretty well, and performance is stable.
- SQL Server in VM. Tried different file placement. There is some perf improvement by moving log/temp files to other disks, however the improvement is not significant. Need further investigation.
- SQL Server in VM. Switching to bigger VM significantly improves performance.
- SQL Azure vs. SQL Server. This is the most interesting part. SQL Azure 1 GB Database achieved comparable performance as SQL Server in Middle sized VM. The price of SQL Azure is only 1/10 of that of SQL Server in this configuration. The result should be related to my testing workload (a mixture of read/update/insert against NopCommerce web), and should not be generalized without further investigation.
