Nearly all enterprises and a good percentage of midsize businesses believe they'll be deploying at least some private cloud technology in the next three years. Any business planning a cloud project quickly realizes there are many cloud software, or "stack," offerings, from different vendors that view the future of cloud in different ways. Picking the right cloud stack now could avoid costly retrenchment later.
The process of choosing a cloud stack starts by defining your private cloud deployment model, based on what's driving the private cloud interest in your company. Here are three common deployment models within enterprises and midsize companies.
- Virtual extension model. In this model, the main consideration is to preserve existing virtualization software, with the goal of supplementing it with cloud computing features and flexibility. With this type of model, a company will want to pick a cloud stack that's compatible with what they're already running.
- Public cloud to hybrid cloud model. These companies have already committed to a public cloud provider, but they are now aiming to bring some cloud projects back on-premises or create a hybrid cloud -- where public cloud resources act as a backup or as offload resources to a growing private cloud. Here, the primary goal of cloud stack selection will be compatibility with the public cloud service that's already in use.
- Greenfield model. In this case, the cloud consumer has no significant virtualization or cloud commitment and is just looking for the best cloud software for the long term. Many larger companies that have a strong commitment to service-oriented architecture (SOA) aim to bridge cloud and SOA into an environment in which the cloud acts as a platform and not a virtual machine host. These companies would align with the greenfield model.
Cloud stack shopping: The virtualization extension model
For companies that fit the virtual extension model, a major question is, "Is your existing IT infrastructure based on VMware?" VMware users typically select VMware's vCloud Director as their cloud software, and this is particularly true for those who use features like vMotion for VM migration. While other cloud stack options can be hypervisor-agnostic, they rarely exploit VMware's capabilities fully or easily.
The process of choosing a cloud stack starts by defining your private cloud deployment model, based on what's driving the private cloud interest in your company.
For companies without existing ties to VMware, the common cloud stack choices are Citrix CloudPlatform, Eucalyptus and OpenStack. All three claim a fair degree of hypervisor support, but the first step should always be to ensure your cloud choice works with your current hypervisor.
The second deciding factor is the cloud software commitment of your own primary IT vendor. If you're an HP shop, for example, you'll probably want to think about OpenStack, since HP's cloud is built on OpenStack technology.
Cloud stack shopping: The public cloud to hybrid model
For businesses that fit the public cloud to hybrid cloud model of private cloud deployment, the obvious first question is, "Which public cloud or clouds do we currently use?" While, there seems to be a market movement toward OpenStack as the framework for any new public cloud adoption, there is also growing competitive tension between the OpenStack community and Amazon Web Services (AWS), which may widen the gap between Amazon clouds and OpenStack software. AWS Elastic Compute Cloud (EC2) is most compatible with Citrix CloudPlatform and Eucalyptus, while Rackspace -- one of the founders of OpenStack -- is clearly more compatible with OpenStack.
An important foundation for members of this group is to know what cloud services they will use beyond basic Infrastructure as a Service (IaaS), such as OpenStack Nova or AWS EC2. AWS is adding more Platform as a Service (PaaS) capabilities every year, and most of these new services aren't directly supported in private clouds.
Planning to create an AWS-symbiotic private cloud will mean matching your plans to AWS services. If the goal is to migrate from public to private cloud, then the services need only be compatible, but if the goal is hybridization, then the services will have to be interoperable as well. This is the area where most users of this cloud deployment model get in trouble, so be careful.
Microsoft Azure users -- or users of Azure-compatible PaaS clouds from other vendors -- can build Windows Azure private clouds and use Microsoft Service Bus to hybridize public clouds with Azure. Be sure to use the latest version of Windows Server for this process. If you're running Windows Server on an IaaS cloud, you can still use the Windows Azure Service Bus tools for hybridization if you have the latest software running on all your internal and cloud servers.
Cloud stack shopping: The greenfield model
For companies with greenfield private clouds, the range of options is large, but so is the risk. The first step is to establish what cloud stack platform will support potential cloud applications. Generally, PaaS will produce the best results in terms of cost savings and operating simplicity. It will also likely make it easier to create a hybrid cloud down the road. A platform-specific hypervisor, such as KVM for Linux, may be the best option if candidate cloud applications all run on one platform and if there is no PaaS platform available.
SOA users should always explore whether a hypervisor is needed at all. SOA applications are componentized and often provide flexibility in hosting components on server pools. Some people have compared SOA to a "Component as a Service" model, and SOA can be hosted on compatible platforms with nothing more than multitasking capability. Often this provides better performance and resource efficiency.
There are other cloud stacks available, in addition to those already mentioned. Many public cloud providers and some large enterprises use the Xen Cloud Platform for its scalability and flexibility, but it's less of a turnkey option. Hadoop is a data-model-based cloud architecture that's ideal for "big data" applications and expanding to other areas, but it's too specialized to be a general cloud stack candidate.
No matter which cloud stack enterprises choose for a private cloud project, they should be prepared to tackle other common problems with workflow distribution and development. And the scale of a deployment can further complicate matters. The larger your cloud deployment plan is, the more critical it will be to explore all options and issues before making a decision. Going back on a cloud commitment is difficult and expensive.
About the author
Tom Nolle is president of CIMI Corp., a strategic consulting firm specializing in telecommunications and data communications since 1982.