Isn’t the Cloud more expensive?
Of course, higher-level services are more expensive per second of computing power compared to managing your own infrastructure. However, to avoid comparing apples and oranges, it is important to include all cost factors.
Understanding cost drivers
When considering costs, it is also important to consider the administrative costs of your instances and the opportunity costs of deploying the development team in more productive areas. Is the architecture really getting the most out of and optimizing higher-level services?
Especially the cost of higher stability has to be factored into this equation. Depending on higher level services and spreading your workloads across Datacenters is relatively straightforward with managed services, but more complicated and work intensive if you build it yourself. Automated health checks and self-healing is built into most higher level services but often has to be integrated into your own instances. This can be easier or harder depending on the types of services you’re using.
„Incorrect configurations can be costly.“
Management of existing instances
Security and access management is another area of worry that has to be thought through and factored into this calculation. The cost of running detection software on instances isn’t free, managing access to existing instances doesn’t happen at no cost and wrong configurations can be quite costly here.
Switch to low-level instances
Switching to low level instances and compute resources in AWS can make sense, after using and optimising higher level service and thoroughly understanding the use case you need those low level resources for. But of course this should not be the first step, especially as most workloads simply aren’t that difficult to run, don’t have that many special cases and can be prized into AWS and other Cloud providers to be competitive with low level instances.
But those calculations have to be done with scrutiny and taking opportunity costs, development complexity, management overhead and security implications into account.
Conclusion
And with AWS providing more and more cost reduction measures like SPOT for different types of compute its hard for this calculation to decide for self managed instances in the end.
We’re happy to help with that calculation.
