In Part 1 of this article, we defined private clouds, talked about the differences between deploying server virtualization and implementing a private cloud, described the risks associated with deploying private clouds and listed the phases and steps involved in transitioning to a private cloud.
In this part we delve deeper into the technology choices needed for the virtualisation, management and automation required for a private cloud. We talk to some who have made the move to a private cloud, or are doing so.
In general, selecting the technologies to implement a private cloud is easier than figuring out the business rules and operational procedures you'll need. Regardless, choosing the software to virtualis your data centre and then picking the automation and orchestration management tools is very important.
While some view automation and orchestration tools as "extra" cloud management tools, implementers and experts say they're just as necessary as the basic tools for managing servers and storage. Without the "extra" tools, you will not be able to reduce the administration costs in private clouds.
How you go about building a private cloud depends on what you have to start with. The legacy of your environment may dictate what you do first. If you are starting from scratch, then you have to start by virtualizing your servers. Then you begin to virtualize your storage and your networks, and build out from there.
These steps are prerequisites if you want to fully realize the benefits of private clouds. You need to be able to provision hardware and software to customers who request it, and then deploy the hardware or services; you need a way to manage and control the environment. You also need to be able to manage the private-to-public paradigm -- that is, the ability to move workloads back and forth between private and public clouds.
So far, how private clouds are built differs from enterprise to enterprise.
When preparing for a private cloud, you have to ask and answer questions such as:
- What is going to be running in the private cloud and what is not?
- What applications can I scale well to take advantage of the cloud?
- If I have two data centres, to what extent can I migrate applications and share capacity between them? Where does cloud help? Where does it hurt?
These questions are part of an iterative process; businesses need to work their way toward mature business processes for their private clouds.
Paul Cameron, head of enterprise services at Suncorp, a major financial services provider in Brisbane, Australia, says that when his company began planning and strategizing for its private cloud, two of the first things it did was create a service-based operating model and create a service catalog. The service catalog contains the list of services being automated for internal use and is available to business users via a self-service portal.
First a framework, then a configuration database
Key to this catalog was the implementation of an ITIL framework that resulted in storing information around Suncorp's assets and business application relationships in a CMDB (configuration management database). All of Suncorp's major IT processes - incident, problem, asset and change -- leverage the CMDB.
Populating a service catalog can be time consuming. But if you are using IT service management and change management tools such as BMC Remedy or Service-now.com and have an existing CMDB in place, it can be easier. You can work through the appropriate services in the CMDB to provide the automated services listed in a service catalog. This is what Suncorp is doing with its BMC Remedy-based CMDB.
Cameron said that Suncorp is deploying a private cloud because it has to serve its customers better and take care of them more quickly. In traditional data centers, enterprises often take a week or even months to provision a server depending on how heavy IT staff workloads are and how long queues are for various tasks required by users.
Now, at Suncorp, a user goes to the self-service portal and requests resources and services. Once the requests are made, the fulfillment of these services is automated. Suncorp has now virtualized most of its data centers around servers, storage and so on, resulting in about 80% of its data center services now being covered by automated self-service portal(s).
Most enterprises that have private clouds use some type of method, such as chargeback or physical limits on the amount of capacity that users can request, to keep the lid on demand. Otherwise, users might just keep provisioning virtual servers and use up the capacity quickly.
Essential cloud components
Jeffrey Driscoll, a systems engineer at consultancy Precision IT, advises that when companies start building a private cloud, the basic building blocks are servers, storage such as a SAN, and virtualization software. "Then you start building a cluster," he says, and after that cluster is complete, "capacity planning becomes critical."
Capacity planning involves figuring out what happens when you add servers and other resources to the cluster as needed to keep up with business demand. Capacity planning is a major component of the cluster and the cloud's performance. If it's done wrong, you might end up with useless systems or have to shoehorn in traditional, non-cloud systems to keep things running.
Most organisations are not good at monitoring and keeping ahead of capacity. To be able to satisfy user demands, you always have to have some extra capacity on the data center floor, which means a certain amount of hardware sitting around in idle mode. Keeping a history of capacity usage in your enterprise can help you make sure that you have sufficient - but not too much - capacity.
One solution is to create a hybrid cloud environment and, when capacity is not available in the private cloud, move requests for capacity to public clouds such as Amazon Elastic Compute Cloud.
Once the cluster is up and running, you can start provisioning virtual servers. The result is a tiered architecture with a server layer, a network layer and a virtualisation layer. There is a management tool at each layer. "Now you can start thinking about automation," Driscoll says.