You are an IT manager and it is late Friday afternoon. You are in the lift descending from the fifth floor and you can already hear the gentle buzz of your friendly local bar. But then the lift stops at level three and the director of human resources gets in.

“I’ve been meaning to talk to you,” she says.“Can you explain why I’m paying so much for my IT?”

It doesn’t have to be this way. In an ideal world, not only would the answer be available in a portal, with costs easily understandable and broken down, but she would be able to configure the IT services she receives, adjusting costs accordingly.

In an ideal world, your datacentre would be able to support all this too and you would still be able to make the pub in time for the first round.

Today’s datacentre is often inward-looking. Managers are concerned with keeping the lights on and the engines running. They view any alterations or additions with suspicion because they work to tight budgets and exacting operational constraints. That is why it can take weeks to provision new storage or server capacity.

This world looks much like the world of the 70s. The computers have changed but from the outside IT still looks like an ivory tower.

Most IT departments want this to change, but how?

It's all about service

IT has to become a service-centric culture, in which the needs of the users are considered first and foremost. This requires a reframing of the entire IT process and therefore of the datacentre.

“The key requirement for a service-centric approach to delivering IT is to define the services required by the various business groups and functions and their value to these groups and functions,” says Reid Purvis, a datacentre specialist at Microsoft.

“These services should then be delivered to the associated users – business application owners and groups, end-users and consumers, or other IT staff.”

A service catalogue with a description of the IT service, service level agreements, costs, and authorised service recipients, is just one part of the offering.

Customers will also want to know what they are paying for. A reporting service is a basic requirement, so that users understand how much they have used a service. Even more business friendly is a chargeback service so they can see what they are paying for.

In a service-led IT culture, the infrastructure must become more malleable so computing resources can be moulded to provide the necessary services within a short time.

Virtualisation provides the answer. Putting a layer of abstraction atop hardware resources enables an IT department to create virtualised resource pools which can be optimised for service delivery.

Anyone designing and building a virtualised datacentre wants to ensure that those resources are effectively managed, otherwise the same old problems will just be shunted up a layer. Badly managed resources are burdensome, whether or not they are sitting on a virtual software shim.

Automation is therefore an important part of tomorrow’s datacentre architecture. It will need orchestration software that handles tasks such as load balancing and workload portability so that virtual machines can fail over to alternate hardware without disrupting services.  

The various components of the infrastructure are often in disarray

Many of these features span different domains. A datacentre offering an in-house service may find itself overwhelmed by a spike in demand or suffering a catastrophic failure. This may require workloads to be moved to a public cloud provider offering an infrastructure as a service contract.

This orchestration layer needs to be service aware, too. It must understand the services offered in the catalogue and ensure that services are meeting the specifications given.

It is clear, then, that this future datacentre architecture will be far more vertically integrated than ever before.

In today’s datacentre, the various components of the infrastructure are often in disarray. Storage networks are separate from server networks. Storage, servers and applications are manually placed, and everything has to be tweaked separately to work with each other. And as for service management, forget it.

In the architecture of the future, the infrastructure is converged. The same network that supports the storage infrastructure is also used for the servers and applications. The application running atop this infrastructure is pre-certified, so that datacentre managers know it will work.

Stack it up

And all of this talks to service orchestration software which ensures the user-facing services are supported smoothly, with minimal disruption. Welcome to the world of the integrated stack.

Vendors have woken up to the need for a computing stack in which all the layers are tightly connected, and they are offering pre-certified integrated stacks. All of the components, from storage through to servers and applications, are provided out of the box, so they are easy to set up and guaranteed to work together.

Some companies have built integrated stacks on their own, including IBM, which has all of the components that it needs to put everything into one box. It can even layer the software on top.

Companies are also partnering with others to fill in the gaps in their portfolio, creating integrated stacks with all of the parts they need.

VMware, EMC and Cisco have  teamed up to create the vBlock offerings, for example, which offer certified storage, networking and computing capabilities in a pre-certified bundle. Cisco also works with NetApp on the competitive Flexpod offering.

Purvis says: “Standardising, abstracting and automating the IT components that business functions and end-users do not interact with directly means these components can be designed in a common, building-block manner.

"They can be highly optimised and made efficient, offering lower costs and better reliability and resiliency.”

In some ways, this takes us right back to the old days, creating a mainframe-like culture in which we trade infrastructural openness for a service-led focus.

Buying an integrated stack pre-configured from a single vendor is not exactly a reinvention of the mainframe, but it has many of the same characteristics and it risks locking people into a single supplier.

An alternative is to focus on the benefits of stack integration while leveraging open interfaces at each layer. Some standards initiatives are already moving in that direction.

Facebook’s Open Compute Project, for example, specifies a range of standard specifications in areas ranging from rack mounting (including high-speed backplanes for connecting servers and storage) through to hardware management tools.  

Don't fence me in

At the software level, the various products should allow user choice. Systems management software platforms should allow for the use of multiple hypervisors, for example, and should manage virtual machines, along with storage and networking, regardless.

“Any organisation considering cloud computing has to consider the vendor lock-in question,” says Rocky Heckman, a technical evangelist for Microsoft’s Azure cloud platform.

“That is why it is critical that they go with a vendor that supports industry standard technologies such as .NET, PHP, Java or Ruby, so that workloads can be transferred to other providers or back in-house.”

Purvis notes the rapid adoption of cloud computing and points to what he calls its key advantage: the overall reduction of vendor lock-in.

“In previous computing models – the mainframe era, the mid-range and Unix era, the large x86/x64 client-server era – hardware vendors exerted much more incumbency power over businesses and enterprise IT departments,” he says.

“With the advent of x86/x64 virtualisation capabilities, the abstraction and standardisation that cloud capabilities offer, these vendors are now forced to compete on the overall value, capabilities and price of their hardware items."

Smart tailoring

At the application level, both proprietary and open source software choices should be available to users who want to mix and match their own solutions, while creating a highly functional, service-aware stack. Customers interested in an integrated stacks do not have to get it all from a single vendor.

At both the hardware and software layers, they should be free to swap out key components in favour of third-party solutions and, so that they can configure the stack to their own precise requirements. After all, tailoring a compute solution to a specific workflow is what integrated stacks are supposed to be about.

Microsoft provides customers and partners with standard blueprints that can be used to assist in building and deploying private clouds, while supporting a variety of third-party products, according to Purvis.

“The Fast Track programme provides pre-integrated and tested solution designs from both Microsoft and various hardware partners that customers can use to implement solutions easily,” he says.

This idea of an open integrated stack has benefits for a service-oriented architecture. It enables you to mix and spread workloads across different sections of your architecture, without worrying about whether the Oracle/Sun stack you bought this year is going to work with that IBM Cloudburst stuff you forked out for the year before.

Thinking about these open interfaces upfront is an important component of datacentre architecture planning. Keeping options open will allow your datacentre to stay flexible.

Purvis offers a final thought: make sure you include more than just technology in your infrastucture plans.

“The IT organisation need to be updated to provide a more collaborative team dynamic, with each team member much more aware of the overall IT objectives and how their peers contribute to the delivery of the IT service,” he advises.

Cloud architecture is all about silo busting. Avoiding vendor lock-in is one form of it, and so is creating links between teams of IT specialists so they understand the impacts that other specialists make within the cloud infrastructure.

A truly sophisticated cloud architecture extends to the team, helping them to understand how they benefit each other in the delivery of IT services. ®