Sinclair is CEO and cofounder of Apprenda.
The idea that hybrid cloud is the end state of enterprise computing is no longer controversial. Nearly all technologists, IT executives, and analysts subscribe to the idea that public cloud and on-premises computing both have a place in modern enterprise IT strategy.
A hybrid end state isn’t a bridging tactic or a strategic consolation prize, but a desirable outcome. In fact, a strong case could be made that a hybrid model allows for specialized optimization based on use cases – there are many scenarios both now and in the future that may map best to on-premises or public cloud.
There are two primary ways to implement a hybrid end state: asymmetric and symmetric.
1. Asymmetric – In asymmetric orientations, an enterprise consumes public cloud as one endpoint, and builds an on-premises cloud that is a distinctly separate, second endpoint. For example, we could look at the Infrastructure-as-a-Service (IaaS) layer and say that an enterprise could use OpenStack on-premises and AWS in public cloud, and use processes, operations, and a brokering abstraction across the two endpoints to help normalize the consumption of IaaS regardless of what side of the firewall it came from.
In asymmetric hybridity, the technology used on-premises is different than that used in the public cloud, resulting in the need for reconciliation and the need to accept a lossy factor (i.e. the two technologies may have different features and evolutionary paths) since points of differentiation between the two need to be ignored or marginalized to ensure consistency.
2. Symmetric – Symmetric hybridity means that an enterprises on-premises assets and public assets are using the same technology, and that technology reconciles the assets on both sides of the firewall as a single endpoint. An example of this would be a Platform-as-a-Service (PaaS) layer that can be installed on-premises that could use local OSes and OSes from one or more public clouds all under one logical instance of the PaaS.
The PaaS hides the fact that resources are coming from disparate providers and only exposes that fact where appropriate (e.g. at the policy definition level to shape deployments). In this case, the PaaS is the single endpoint where interaction happens, and resources on both sides are used as resource units by the PaaS. Any organizational processes and consumption processes would be ignorant to the idea that a border exists in the resource model.
Pros and cons of symmetric and asymmetric models
Symmetric models guarantee that anyone within the enterprise consuming cloud infrastructure is shielded from the distinction between on- and off-premises resources and capabilities. If the end user of cloud infrastructure (e.g. a developer or data scientist) is required to acknowledge any asymmetry, they will have to deal with it in their project. This explicit need to deal with a fractured cloud creates an immediate “tax” related to consuming infrastructure and it will generate consumption biases.
For example, if one side of the asymmetric deployment is easier to consume than the other, then an end user will prefer that even if the not-so-easy side is more aligned with the project, and will cause IT itself a number of headaches when it comes to operations related to that project.
It’s important to understand symmetry doesn’t mean the on-premises and public cloud side of a hybrid deployment must be equal. Certainly, workloads may need on-premises or public assets to satisfy certain requirements the other side couldn’t possibly satisfy.
What symmetry guarantees is that a workload that is indifferent to on-premises or public never be exposed to those concepts. Symmetry also ensures a workload with requirements that can only be satisfied by one part of a hybrid cloud or another is never exposed to the technical divide between the clouds. Instead, a workload communicates its preference in the language of requirements.
Asymmetric models might be good starting points or appropriate for certain layers of the infrastructure stack, but they’re not ideal as a final end state. Symmetric models are clearly superior in almost all other aspects.
In response to this, CIOs should pursue a strategy of ‘Eventual Symmetry.’ Eventual symmetry means that any cloud strategy must:
- Choose symmetric models over asymmetric models where possible
- If asymmetric is the only possible approach, ensure that the implementation lends itself to eventually being replaced by a symmetric model or that processes and technology be used to abstract the asymmetry into a perceived symmetric model
By establishing eventual symmetry as a core cloud strategy pillar, a CIO can guarantee that any disjointedness in their strategy will be resolved. He or she can also ensure consumers of IT resources are abstracted away from details related to on-premises and off-premises.