Cloud computing may be the next logical evolution of virtual data centers, but rushing into a private cloud design can create more problems than it solves.
Unfortunately, many companies (or at least C-level executives) fall into the trap set by software vendors and believe that they need a cloud yesterday -- and that building one is as simple as selecting a product.
The reason Amazon succeeded is that we can do everything we wish in their cloud. You have to copy that model
Ivan Pepelnjak, chief technology advisor at NIL Data Communications
Once on this path, it doesn’t take long for most to realize building a private cloud is much more complicated than paying for licensing and management software, said Ivan Pepelnjak, chief technology advisor at NIL Data Communications, at an Interop session.
"I can't tell you how wrong this approach is," Pepelnjak said. "And I see this happening with about 80% of the customers I see. The most important question is not 'OpenStack or vCloud Director?' The most important question is, 'What do you want to do?'"
In fact, most companies aren’t actually looking for a private cloud at all, Pepelnjak said. What they really want is the ability to automate some parts of VM provisioning. To be a true private cloud, IT must offer more.
"If your users have to open a support ticket to add a firewall code to the central firewall, it's not a cloud. It might be called 'cloud' so that someone gets his yearly bonus, but it is not a cloud. It is server virtualization," Pepelnjak said.
The correct way to approach a private cloud design is to first take a step back and decide what you need, and if you do need a private cloud, evaluating which platform will best fit your business. With that in mind, here are three common mistakes to avoid when designing a private cloud.
Building a private cloud for nobody
The first step in any potential private cloud deployment is identifying who will use it and what their needs are. All too often, IT will design a cloud infrastructure without consulting application owners, only to later find the cloud they've built doesn't fit their needs, Pepelnjak said.
Identifying your customers is also important to ensuring they'll know how to use the cloud once it's built.
"Look at how much Amazon is investing on documents about how to use their features. You have to do something similar for your customers or they will not use it," Pepelnjak said.
Consider taking the concept of identifying your customer even farther by choosing one person who's interested in or pushing for a cloud, and build a beta cloud tailored to fit him, Pepelnjak said. That customer will work with you through bugs and help you refine the offering. Once you have a cloud to be proud of, he'll become your chief salesman.
"He will tell other application teams how stupid they are for not using your cloud," Pepelnjak said.
Not trusting your customers
One of the biggest private cloud design mistakes IT teams make is failing to provide adequate self-service levels. Letting application owners configure firewall and load balancer settings can be scary, but a successful private cloud offloads as much of the decision making process as possible to customers.
In short, a cloud should give users the ability to provision and configure any service they would normally need. This will vary from one organization to another, or even among business units in an enterprise. Some customers may only need the ability to provision individual VMs, while others will need the ability to design and support complex application stacks with virtual networks and load balancing.
"Amazon would never succeed if you had to open a ticket for every firewall change,” Pepelnjak said. “The reason Amazon succeeded is that we can do everything we wish in their cloud. You have to copy that model.”
Building your own open source cloud
OpenStack has received a lot of buzz, but that doesn't mean it's right for you. Most companies deploying their first private cloud will be better off buying cloud software from a major vendor, like VMware, rather than building their own OpenStack-based cloud, Pepelnjak said.
"Building sounds cheaper, because you're not paying for licenses. But remember that, there's a reason this is free, and in the end it is not free. You can download OpenStack and you can play with it, and maybe two years later you will have a working cloud,” Pepelnjak said.
There are exceptions to this rule, including service providers that need to find every possible efficiency and organizations with staff already experienced with building OpenStack clouds.
This doesn't mean you should write off open source options altogether. In fact, Pepelnjak sees plenty of advantages to OpenStack clouds, and suggests organizations start experimenting with building their own alongside a production cloud from a major vendor, just so that they can learn how.
"So, in two or three years, when you have time to roll out 'MyPrivateCloud Version 2.0,' you will be ready. But if you are just starting today, you are not ready to build," he said.