While comparing traditional enterprise requirements against what is provided by AWS isn't typically very informative, VMware's response brought a few things to mind.
- While the infographic includes power and cooling, the bullet-ed list does not – it's important to make sure that you have a clear understanding of all of the costs required beyond the cost of the server. This includes power, cooling, network connectivity (bandwidth between sites isn't cheap), and cabling.
- When comparing the solutions, make sure you consider monitoring and automation in scope – often times, cloud offerings can be better and cheaper than what you currently have deployed on-premises.
- Also, be sure to quantify the security benefits of cloud instances over traditional infrastructure – Amazon Security Groups compared to VMware NSX for per VM traffic security, for example.
- The final often-overlooked bit is the cost of rolling out additional services – what is the difference between developing functionality on premises versus leveraging already developed features in the cloud. On Microsoft Azure, MSSQL clustering is a checkbox – how long would it take your DBA team to development, deploy, and maintain that functionality?
Ultimately, any generic TCO calculator is only going to tell a piece of the total story – take a hard look at your own environment and exactly what cloud-based IaaS and PaaS could provide before determining that "it is cheaper to do it yourself."
Succint post, accurate points. Good work!
In addition to the physical costs of an individual machine you have to consider the overall cost of capacity to the benefit of elasticity in the cloud. What does the cost per server become when I no longer have any capacity in my datacenter? That concept doesn’t exist in Azure or AWS where I can add almost a limitless number of instances.
Even better yet is that with a bit of monitoring and automation I only pay for instances when I need them and the minute I don’t, my costs drop dramatically. With an on premise environment even if I power down a hypervisor when not in use I still incur a tremendous cost for it while it sits idle.