Now that I am back from summer break, I want to continue to further the discussion on cloud and map out how medium and large enterprises can build their own private cloud. As we’ve discussed previously, software-as-a-service, engineered stacks and private cloud will be the biggest IT winners in the next five to ten years. Private clouds hold the most potential — in fact, early adopters such as JP Morgan Chase and Fidelity are seeing larger savings and greater benefits than initially anticipated.

While savings is a key reason to move to a private cloud, shorter development cycles and faster time to market are more significant. Organizations can test risky ideas more easily as small, low-cost projects, quickly dispensing with those projects that fail and accelerating those that show more promise.

While savings is a key driver to moving to private cloud, faster development cycles and better time to market are turning out to be both more significant and more valuable to early adopter firms than initially estimated. And it is not just a speed improvement but a qualitative improvement where smaller projects can trialled or riskier pilots can be executed with far greater speed and nominal costs. This allows a ‘fast fail’ approach on corporate innovation that greatly speeds the selection process, avoids extensive wasted investment in lengthier traditional pilots (that would have failed anyway) and greatly improves time to market on those ideas that are successful.

As for the larger savings, early implementations at scale are seeing savings well in excess of 50%. This is well beyond my estimate of 30% and is occurring in large part because of the vastly reduced labor requirements to build and administer a private cloud versus traditional infrastructure.

So with greater potential benefits, how should an IT department go about building a private cloud? The fundamental building blocks required for private cloud are a base of virtualized servers utilizing commodity servers and leveraging open systems. And of course you need the server engineering and administration expertise to support the platform. There’s also a strong early trend toward leveraging open source software for private clouds, from the Linux operating system to OpenNebula and Eucalyptus for infrastructure management. But just having a virtualized server platform does not result in private cloud. There are several additional elements required.

First, establish a set of standardized images that constitute most of the stack. Preferably, that stack will go from the hardware layer to the operating system to the application server layer, and it will include systems management, security, middleware and database. Ideally, go with a dozen or fewer server images and certainly no more than 20. Consider everything else to be custom and treated separately and differently from the cloud.

Once you have established your target set of private cloud images you should build a catalogue and ordering process that is easy, rapid, and transparent. The costs should be clear, and the server units should be processor-months or processor-weeks. You will need to couple the catalogue with highly automated provisioning and de-provisioning. Your objective should be to deliver servers quickly, certainly within hours, preferably within minutes (once the costs are authorized by the customer). And de-provisioning should be just as rapid and regular. In fact, you should offer automated ‘sunset’ servers in test and development environments (e.g., after 90 days the server(s) are allocated, they are automatically returned to the pool). I strongly recommend well-published and clear cost and allocation reporting to drive the right behaviors among your users. It will encourage quicker adoption, better and more efficient usage and rapid turn-in when no longer needed. With these 4 prerequisites in place (standard images, a catalogue and easy ordering process, clear costs and allocations, and automated provisioning and de-provisioning) you are ready to start your private cloud.

Look to build your private cloud in parallel to your traditional data center platforms. There should be both a development and test private cloud as well as a production private cloud. Seed the cloud with an initial investment of servers of each standard type. Then transition demand into the private as new projects initiate and proceed to grow it project by project.

You could begin by routing small and medium size projects to the private cloud environment and as it builds up scale and provisioning kinks are ironed out, migrate more and more server requests until nearly all requests are routed through your private cloud path. As you begin to achieve scale and you prove out your ordering and provisioning (and de-provisioning processes) you can begin to tighten the criteria for projects to proceed with traditional custom servers. Within 6 months, custom, traditional servers should be the rare exception and should be charged fully for the excess costs they will generate.

 Once the private cloud is established you can verify the costs savings and advantages. And there will be additional advantages such as improved time to market because of improvements in the speed of your development efforts given server deployment is no longer a long pole in the tent. Well-armed with this data, you can now circle back and tackle existing environments and legacy custom servers. While often the business case for a platform transition is not a good investment, a transition to private cloud during another event (e.g., major application release, server end-of-life migration) should easily become a winning investment. A few early adopters (such as JPMC or Fidelity) are seeing outsized benefits and strong developer push into these private cloud environments. So, if you build it well, you should be able to reap the same advantages.

How is your cloud journey proceeding? Are there other key steps necessary to be successful? I look forward to hearing your perspective.

Best, Jim Ditmore