Once you’ve hit the limits of what Cloud can offer you, you need to make some decisions about what’s next. For everyone, their specific reasons are different- some need greater control, others need better performance, and still others need strict cost control.
What are your options?
Colocation
The less you ask the Cloud’s defaults for, the more control you have over your own environment. The classic Colocation offers space, power, and physical security. This leaves your network, your servers, and your deployments entirely within your own hands, but it also allows perfectly optimizing your platform architecture.
The more knowledge you have about your needs, the better your environment becomes. You can alleviate physical bottlenecks which alternatives literally cannot. You can build high availability without depending on a third party’s reliability. You can scale up or down and pursue compute density and storage density at a level which makes sense to you instead of what makes sense to a third party.
The downside is that this requires human expertise and talent. Understanding architecture is necessary to optimize architecture. There are no shortage of gains to be had, but the architect must have the vision to see them.
On-Prem
Many, but not all of the advantages of a Colocation can be had deploying on premises, within an existing office. Once again, platform architecture comes to the fore. Usually sizing, power, network bandwidth and cooling requirements place a bound to what can be deployed within such an environment.
That said, an architect retains the ability to optimize the workload at hand. Specific benefits include a tight integration between any front-office as well as back-office functionality. Anything local can remain local, so customized deployments become simpler.
Control also becomes easier, as any physical issue can be addressed from the office rather than utilizing a Colocation’s remote hands.
Managed Metal
In between Colocation and Cloud sits Managed Metal. It offers less customization than Colocation and forces some trade-offs between the Metal provider’s decisions and your organization’s workload needs.
The benefit of Managed Metal over Cloud is that it’s a physical environment instead of a virtualized environment, so there are no shared servers or shared workload environments. One customer’s workload cannot steal resources from another customer’s workload.
It also brings the managed overhead benefits of Cloud, in that the backbone network and physical deployments are handled via a third party, so any issue must be addressed by that third party. However, the key downside of Managed Metal is that you are stuck with the design decisions of the provider. The hardware specifications are chosen by them, not by you, and their network is designed by them, not by you, so it may not match well with your actual needs. Your level of control may not be as high as it is with Colocation, nor will the feature set be as broad as with Colocation, but you are within a tightly controlled and managed environment where Managed Metal can make more service guarantees than Colocation can.
Further, Managed Metal can offer a lot of programmability of Cloud. Infrastructure-as-a-service is core to the concept, where hardware deployments can be scripted, and the provider’s API can be pulled into automated workflows.
Conclusion
Getting the design and architecture right is a hard challenge, and making the right decision between Colocation, On-Prem, and Managed Metal requires a deep understanding of your own workload, its pain points and bottlenecks. Getting the balance right between customization and automation is not trivial, and much thought must be put into the specifics.
