As with most complicated projects, you're better off building in automation from the get-go; retrofitting is more expensive and less effective. So we were somewhat discouraged with the results of our InformationWeek 2012 Private Cloud Survey. The good news is, this technology has reached a tipping point: 51% of 414 respondents are either building private clouds (30%) or have them in place now (21%). But when we asked those in the construction stage about nine critical steps, orchestrating automation across multiple subsystems came in dead last.
Let's be clear: No automation, no cloud. How do we figure that? NIST defines cloud as having five essential characteristics: on-demand self-service, broad network access, resource pooling, rapid elasticity or expansion, and measured service. Virtualization and solid WAN engineering will provide resource pooling, elasticity, and broad network access, but measured service and--most importantly--on-demand self-service aren't part of standard virtualization management suites. For these, automation is required.
Self-service isn't the only benefit. More efficient use of data center resources, self-healing, improved application availability, better power management, and preplanned responses to various scenarios are among the other potential benefits of a solid automation deployment.
Unfortunately, there isn't a standard way to do cloud automation; in fact, there isn't even agreement on what, exactly, it entails. While virtualization vendors have invested a lot of effort in developing APIs that provide extensibility and control, automating those infrastructures is simply not a part of the core virtualization feature set. And yet, controlling a virtualized infrastructure is going to be a key point of any automation strategy, because virtualization is where your resource pools and elasticity live.
At the most basic level, cloud automation packages support runbooks that take preprogrammed actions when a trigger event occurs. But preprogrammed events are just the beginning; new and innovative products, like those we list on p. 25, take management to a new level by enabling policy-based automation. These products use multiple management engines to stay in touch with all aspects of the infrastructure and make policy decisions based on specific scenarios and self-service requests.
Vendors are evolving these systems from workload management suites used to automate diverse virtualization and infrastructure components through a central policy engine. However, because the enterprise private cloud automation market is relatively new, there's no stock feature set, so what you'll be able to do out of the box varies dramatically. For example, while most of these suites have powerful central execution engines that can read data and act on it, some, like Moab, incorporate enhanced resource management for virtual infrastructures and self-service Web provisioning portals as well.
Despite the immaturity of this market, it's worth evaluating these suites, because automating the cloud has huge potential to maximize your investment and slash operational and capital expenses--an important point, as 61% of survey respondents not using a private cloud cite reduced operational costs as a major reason to consider moving to the cloud, with capital expense savings (44%) and technical advantage (45%) as strong secondary factors.
Build On Your Accomplishments
Successful automation deployments sit on top of strong virtualization deployments that provide high availability, scalability, and a degree of fault tolerance, or at least fault recovery.
Still, the first step in preparing to automate is cleaning house. Automatic actions and self-service provisioning will exacerbate poorly configured virtual infrastructures. In addition, better engines help improve resource management, which is difficult to do if the infrastructure is already overloaded. If you don't have the spare capacity to maintain high availability, self-service provisioning may put core network services at risk.
What exactly does housecleaning involve? Consider resource provisioning first. If demand spikes jeopardize the performance of mission-critical business services by starving application servers, that can be a problem. You must segregate workloads into resource pools and assign priorities to them--as self-service requests come in, critical servers must be given priority access to underlying physical resources. When selecting an automation management system, look for the ability to manage resource load in cloud environments, but be aware that just throwing resource management at a badly configured infrastructure is likely to net you a lot of angry help desk calls.
In addition, make sure you have a method to track when a VM produced by automation becomes a mission-critical server. It's easy enough to increase its resource priority after the fact, but this is one area where you don't want to run blind. And don't spend on automation if the overall capacity of the infrastructure is lacking. If you're barely able to satisfy your current workload, the last thing you need is new machines being rolled out without human intervention.
This report includes 14 pages of action-oriented analysis. What you'll find:
- Ten IT success metrics and how well private clouds deliver
- Example goal sets and enabling processes and self-healing actions