IT shops first turned to virtualization to consolidate hardware and reduce IT costs. But today, virtualized infrastructure has evolved well beyond these use cases.
Now IT departments can use virtualized environments to create IT as a Service (IaaS) models. They can share software and hardware resources, more flexibly and dynamically serve up applications to end users, and reduce management and maintenance headaches. And IT departments can potentially provide this new flexibility while still reducing costs.
This grand vision for computing— also known as a private cloud computing model—holds serious potential but also substantial caveats. While many technology vendors tout their products as all-encompassing cloud “suites,” many fall short of enabling a “true” cloud environment. Indeed, many lack key management features—from Web portals to template customization to visibility into physical resources— or rely on other providers to fill in feature gaps—but at the expense of ease of use. To make good on the promise of a private cloud infrastructure, the technology still needs to catch up.
But technology immaturity is only part of the problem. Behavioral shifts are required as well. As virtualization matures and moves “up the stack,” applications will be delivered in a new way. As a result, IT departments need to adjust to view end users as “end customers” for whom they provide efficient, effective service. Without accepting this service component of IT, end users will invariably turn to external sources to fulfill their needs and continue to buck corporate IT mandates.
In what follows, we outline providers’ current capabilities, some guidelines to weigh cloud providers’ offerings and some of the behavioral shifts that a private cloud model requires.
Evaluating cloud provider options
Several major vendors offer some version of this private cloud vision, including IBM, Microsoft, Vmware Inc. and open source providers. But gaining the benefits of a private cloud model hinges on the strength of a cloud management suite’s capabilities.
Indeed, not all these cloud offerings are created equal. Many vendors have some, but not all, the building blocks for a true private cloud solution. When a software vendor offers only some of the pieces required, customers are invariably forced to blend different providers’ offerings together. In other cases, vendors push back components of the cloud to a third party, which is often a euphemism for an underdeveloped area of a provider’s product. But in either case, this “patchwork approach” drives up complexity and creates compatibility problems. This invariably leads to vendor finger-pointing.
Many vendors also suggest that shops need only a core hypervisor rather than a hypervisor plus a management platform. But I doubt that over the long term a raw hypervisor approach will pay dividends. Yes, you can save money by not purchasing VMware’s management suite, vCenter, for example. But given that it is a basic requirement for all advanced VMware management features, few private clouds will be built on a hypervisor-only approach.
1. Self-service portals. When considering the merits of a private cloud vendor, the first necessary feature is a self-service Web portal.
End users need an easy way to spin up new virtual machines (VMs). While you might expect all vendors’ suites would include a basic Web front end, some providers do not.
Further, some vendors separate self-service portals from cloud administration management tools, which allow you to carve up an array of physical servers as standalone hypervisors or as systems that are part of a wider group of resources in a high-availability server cluster. This separation of administration tools from a portal can be cumbersome. But it may be a necessary evil to keep the two separate for increased security.
Most vendors also allow you to create a “service catalog” of preconfigured server definitions, or templates, from which to choose. Most piggyback off virtualization vendors’ “template” feature. But the template-only approach is too inflexible; most templates use a flat-file approach to deploying new servers, which doesn’t allow for sufficient customization. Customers want to be able to choose their options.
The best service catalog offerings use a flat-file template to build out the initial OS, but then give end users a field picker to include additional software and services as needed. In a Linux environment, for example, this allows you to add or remove services such as Apache or MySQL with a mere tick of a box.
A few providers allow you to create your own library of templates that is separate from those provided by virtualization vendors. And an even smaller number enable you to build a new server on demand, which gives far more granular design control than a template offers.
Additionally, vendors are still stronger in building out options for Linux-based rather than for Windows-based virtual machines. Linux lends itself to customization via scripted installations more than does Microsoft Windows. One cloud provider with which I spoke has no support for building out Windows-based VMs other than via templates and using Microsoft’s Sysprep tool, with much of the rest of the guest customization being left to the customer. But how is this low level of automation more than what you can achieve with base virtualization tools?
2. Workflow creation and management. Most companies impose a complex series of steps on new servers before they can go “live” on a network. Each step must be carried out in a specific order, and along the way, various members need to approve a server’s status before it can move on to the next step.
But few cloud vendors I evaluated had little more than an email system by which requests could be sent with a simple Approve or Deny button option. For many IT shops, this approach is not sophisticated enough to handle the complexity of their operational processes. The best vendors include workflow engines so a cloud administrator can see his processes in a flow chart. Despite their 1970s-era throwback feel, flow charts are widely used to map out complex if-then-else logic to govern each step.
Standout cloud workflow components are highly visual and built into a cloud offering from the ground up. Weaker offerings had no workflow component, or the component was sold as an add-on. These add-ons often had interfaces and menus that differed substantially from the main cloud administration tools. For end users, this lack of integration between interfaces can be a hurdle.
3. Physical resources. Cloud management tools have to provide visibility into physical and virtual resources. In fact, a virtualizationonly approach to building a private cloud would ignore most data centers’ reality today. Various studies indicate that, on average, virtual machines account for only 25% of instances on the network.
One of my customers, for example, has physical servers that run VMware ESX, Microsoft Hyper-V or Citrix Systems Inc.’s XenServer. Whether these physical resources run a hypervisor or OS directly installed, they must be part of our cloudy vision. To ignore them is to consign the cloud project to a narrow virtualization-driven cul-de-sac.
So while cloud and virtualization are closely bound together, physical servers continue to play an important role—even if they exist only to run virtual machines. Strong cloud providers understand this reality and offer physical and virtual machines from their portals.
4. Chargeback and showback. Another essential part of cloud management is the ability to do chargeback and showback (a lightweight, less formal version of chargeback).
Many private cloud vendors offer little guidance on the price of the various parts of a virtual machine. The vast majority recommend a vague (i.e., gold, silver and bronze) pricing system. While this approach is sensible, few cloud vendors are willing to outline the dollars attached to each management feature and how they combine to create each level of service.
As a result, many businesses will struggle to develop a pricing structure for VMs that reflects these servers’ true cost. Many shops have little experience in developing chargeback or showback models because they have been allowed to consume IT resources with little consideration for true cost. But despite unfamiliarity with chargeback and showback models, businesses need to be able to demonstrate that a project will be successful in reducing cost and also demonstrate the hypothetical cost if infrastructure were to run on a commercial basis.
The best cloud vendors handle chargeback and showback dynamically. As an end consumer defines a VM, he can see the cost difference between choosing a VM with four virtual CPUs (vCPUs) and one vCPU. Customers want to see what customization will cost before they make a purchase, not after. Sadly, many cloud vendors allow you to spec out the cost of a virtual machine only at the end of, not during, the process.
5. Networking. Most cloud providers are weakest in the network layer. The best cloud vendors create new network definitions, or pools, on demand as needed. But surprisingly most still depend on manual work from an administrator, who must define virtual local area networks (VLANs) at the physical switch and then map to them with virtual switches. This manual work makes provider claims to “secure multi-tenancy” somewhat hollow.
The best providers can create new VLAN definitions among an array of hypervisors on demand, and only a token few offer networking that allows for secure network translation. Without an easy way to separate one set of VMs from another, customers can be restricted in the cloud provider software they can adopt.
6. End-consumer paradigm. Increasingly corporate IT will have to accept that it is part of the marketplace for IT services. Once the cloud approach takes root, commercial vendors will compete with internal IT for these customers’ attention. So IT shops have to stop thinking in terms of end users, or application owners, and start thinking in terms of end consumers.
Cloud offerings should thus have the slick presentation that one would expect from a commercial public cloud provider. If corporate IT doesn’t step up and present its offering as an external commercial provider would, it runs the risk of end consumers taking their business elsewhere.
But despite all the talk about IT servicing business needs, many end consumers find corporate IT unreceptive. While corporate IT may not be ready for this sea change in thinking, it may nonetheless have a short time to react or, instead, suffer extinction.
New technologies and new behavioral models required
The success of the private cloud model will hinge in part on whether management technologies can catch up to business needs. Users want to dynamically spin up virtual machines from a Web portal, to create custom templates for VM creation, to view their physical and virtual resources together, and to price out the costs of individual components of a private cloud infrastructure. Cloud vendors should strive to offer these capabilities themselves or to better integrate third-party add-ons into their management products.
So of course, technology is only half the battle. Changing IT departments’ behavior is also part of the success of the “cloud equation” that will enable this new architecture to trickle down to average IT shops. Until IT comes to view its users as end consumers rather than nagging, high-maintenance users, the private cloud model will remain the province of a rarefied few.
About the Author
Mike Laverick is an IT instructor with 15 years of experience in technologies such as Novell,Windows and Citrix. Since 2003, he has been involved with the VMware community and is a Vmware forum moderator as well as a member of the London VMware User Group Steering Committee. He is the owner and author of the virtualization blog RTFM Education, where he publishes free guides and utilities for VMware users. He is also writing a book on building a cloud with VMware vSphere as the foundation.