I’ve been getting a few questions lately about whether or not we support TFS on Azure IaaS. The truth is that a number of us have set up and used it on IaaS but we have never really formally tested it. My biggest question was performance. We’ve just finished running some performance tests. Here’s the results we got:
| TFS config | Model | CPU | Memory | Disk | Active TPC |
| Dual server | AT: Medium Azure VM | AT: 2P | AT: 3.5GB | AT: | 30 |
| DT: Large Azure VM | DT: 4P | DT: 7GB | 1xVHD: OS | ||
| 1xVHD: VC Proxy files | |||||
| DT: | |||||
| 1xVHDs: TempDB files | |||||
| 1xVHD: Data files | |||||
| 1xVHD: Log files | |||||
| 1xVHD: OS |
In this configuration, we were able to achieve results supporting a team of 500 – 1,000 users. This is roughly consistent with the hardware sizing guidelines we give for dedicated hardware. Our results indicate that TFS on Azure IaaS will cap out ~1,500 users due to exhaustion of I/O bandwidth for the log and temp db.
Of course, Azure IaaS hasn’t RTMed yet so we can’t officially support it until it does but once it RTMs, this should provide some indication that it performs/scales pretty well.
Brian