Designing the Private Cloud – Top Down or Bottom Up

In the quest to build a private cloud, it’s important to note how the individual parts fit.  Designed well, and you’ve created an automated, cloud optimized data center allowing the enterprise to respond to the dynamic pace of the changing business landscape.  Designed poorly, and you’ve added another layer of data center controls further emphasizing IT as a cost center, not a business enabler.   

There are two ways to go about the design of your private cloud.  Top down, choosing orchestration component like OpenStack and utilizing commodity parts to build out; or bottom up, utilizing hardware components already in place and determining the best orchestration layer to fit those components.  For many large enterprises, the typical choice is bottom up; utilizing existing architecture and infrastructure already in the production environment and centering on the corporate standard hypervisor and embedded staff expertise.  Others however have shed any preconceived notions of their existing infrastructure and are building from the ground up; based on open standards with the plan to migrate existing application (or building all new applications) to function in the open standards cloud environment.  For purposes of this article, I’m going to focus on the bottom up approach.  Information on building an OpenStack environment can be found at and will be discussed in a future article.

 Cloud Management Picture

Virtualization-enabled Core Infrastructure:  Embedded and entrenched infrastructure can be the cornerstone of your cloud build.  However, that infrastructure must be able to take advantage of some of the enhanced attributes of the hypervisor.  Capabilities like vMotion, Live Migration, Replication, VDS, vShield, Virtual Switch, etc. are only supported on specific vendor hardware.  If your storage, server, or network can’t take advantage of these capabilities, you pay for unutilized features in both CapEx and OpEx.  Or worse, you pay extra for software and services to make up for the deficiencies in your hardware.

Converged Infrastructure:  A recent trend in the industry is converged infrastructure.  In simple terms, a Converged Infrastructure (CI) allows you to manage you storage, networks and servers as a single unit, from a single console, gaining data center automation and management simplification.  With CI, you can provision a VM from a single console including all storage provisioning, memory allocation and network connections.  This solution though poses the very possible risk of vendor lock-in.  Many vendors sell their version of a converged infrastructure that ties you to their compute, network and storage choices.  They want you to purchase their version of the client-server mainframe.  Vendors need to recognize that customers have entrenched infrastructure that needs to be integrated into the CI, not excluded.  As the consumer, you need to ask how will my existing infrastructure integrate with the CI, and what happens in three years when I need to refresh.  Some vendors currently offer a more open CI while others are moving in that direction.  Caveat emptor.

Virtualization Layer:  The two leaders in virtualization for better or worse are Microsoft (Hyper-V) and VMware (ESXi).  To discuss which is better is to get into a political debate.  Alternative to these entrenched hypervisors, XEN and KVM stand out as excellent alternatives.  Although Hyper-V and ESXi are feature rich and industry leaders in hypervisor technology, XEN and KVM typically don’t have the same price tag associated with them and fit better into a more open standards environment.  For most enterprises, you’re already invested in the skill set and standardized around one of these four hypervisors making the decision easy.

Orchestration Layer:  Quite simply, the orchestration layer, or cloud layer if you will, provides for the five essential characteristics of the cloud.  That doesn’t mean that your cloud must include all 5 to be considered a cloud; only that cloud is made up of at least some of these characteristics.  I would argue though that a fully functioning private cloud comprises of the first three (On-Demand Self-Service, Broad Network Access, Resource Pooling) with only the fourth (Rapid Elasticity) and fifth (Measured Service) being optional.

Cloud Essential Characteristics (per NIST)

  • On Demand Self Service
  • Broad Network Access
  • Resource Pooling
  • Rapid Elasticity or Expansion
  • Measured Service

Cloud Management:  It is only when we reach Hybrid Clouds that Cloud Management becomes a critical construct.  Cloud Management allows for both the management of multiple clouds, and the movement of workloads between clouds, in an automated fashion.  This is really the wholly grail of a strategic, cloud optimized data centers.  The degree to which you can move workloads from one private cloud to another is made possible by the cloud management layer.


Ultimately, designing your private cloud is dependent on your entrenched infrastructure and software, the skill set of the team managing the data center, and your future vision.  Whether top down or bottom up, the only wrong choice is status quo.



Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: