Thursday, February 25, 2010

As mentioned, management’s already bought in to the virtualization dream and IT is committed to making it work. The justification used for funding virtualization often assumed historic workloads would remain the same and resource efficiency would improve. But while these projects have netted some real server consolidation, they’ve also seen an increase in virtual machine instances. And they haven’t always produced the improvements in resource utilization or efficiency that were possible. Managing dynamic, virtual resource pools has proven a challenge for operators, who often see administrative time requirements increase. Having a chance at effectively managing this new ecosystem requires a more granular understanding of the virtual server environment. Unfortunately, many organizations are still using management tools that were designed for the physical environment. They’re fighting this new virtual ‘war’ with weapons from the last, physical one.

More Servers and Complexity

Based on the ease of VM creation, the number of virtual server instances can exceed the number of physical servers before the project. In other words, the virtualization project itself can cause an increase in servers and a decrease in available resources, especially time. While we’ve virtualized servers and can rapidly create them, we haven’t found a way to do the same with administrators. Resource sharing is a cornerstone of the virtual environment that’s supposed to optimize resource usage, but also results in increased complexity. Managing VMs includes the process of allocating the ‘four food groups’ that feed servers - CPU cycles, Memory capacity, Disk I/O and Network I/O. These resources must be in balance; as you increase one group, you have to increase the others or risk waste. This dynamic process, a classic multi-variable equation of sorts, requires more than the static information available from existing tools. Isolated data on specific components is less effective in a virtual environment where change is a constant and tuning an ongoing activity.

More Expectations

The expectations of management and those of internal business unit customers can be a challenge that didn’t exist in the pre-virtual environment. Having been sold on the VM concept, internal user customers often expect the ability to create more server instances on short notice and have more resources available (due to expected efficiency improvements), all while their application service levels remain the same. They may also assume any problems that occur are due to the virtualization project and won’t hesitate to let IT know this.

Next generation virtual infrastructure management systems provide operations with ‘weapons’ to fight the current war against VM sprawl, resource mis-allocation and administrative overload. Solutions like Vizioncore’s vFoglight can help by providing a new level of information about the virtual environment, with relationships between data sources and an intelligent abstraction of this larger volume of information. They can help IT resolve the issues brought on by server virtualization projects.

Navigate Potential Problems

Virtualization’s abstraction layer hides problems by withholding details about server instances and resources. Also, the VM population increase can serve to raise ‘noise level’ as IT monitors more server instances. What’s needed are monitoring systems that help separate data about conditions that warrant immediate action from the noise. ‘Intelligent abstraction’ is needed to help filter available data and prioritize it by immediacy, presenting the highest level information with simple drill down methodology to manage information overload.

These VM management systems use visual alerts, graphics with colors, movement and logical screen arrangement to make it easy to spot trouble when it occurs. Custom dashboard layouts enable systems to be tailored to the preference of the operator and include the most critical VMs or resources in prominent positions. All this allows for faster uptake of important information with fewer errors.

Manage the Complexity

Once a condition is identified, the system must provide all the needed details to resolve the issues, quickly. This means the user must be able to drill down from an alert with the context and perspective needed to interpret these data effectively. For example, CPU utilization from the VM perspective means something different than CPU utilization of the ESX host. Both of these data points are important, as is the relationship between them, and this context is critical when trying to balance resources, essential in a virtual environment. In addition to VM and ESX host context, the Virtual Center, Data Center, Clusters, Resource Pools and Data Stores levels are important contexts to monitor.

Optimize the Environment

After putting out fires, IT must get back to the task of increasing efficiency and resource utilization and driving down costs. After all, these were some of the justifications for making the move to server virtualization to begin with. This often involves moving VMs around between physical hosts. These new virtual infrastructure management systems include tools to model the migration of VMs and determine the effects of moving resources within the environment, before the changes are executed.

Troubleshoot Effectively

These systems give you the ability to look at each metric in real time or in an historical mode. This timeline format presents the same metric as recorded at points in the past so that certain events of interest can be replayed. The granularity of these views are user configurable, so that less data is recorded and stored as time passes and the likelihood of review decreases.

Get data you can use

Data sets can be captured in a report and shared with management and others to help manage the expectations surrounding a virtualization project. ‘Bookmarks’ can be placed on a screen to simplify returning to that data point later, or sharing with others. Pertinent data objects can be dragged from other screens and saved as a new dashboard, allowing operators to keep track of ‘problem’ servers or applications. The four basic resources, CPU cycles, memory, disk I/O and network I/O are presented in visual, colored format on the dashboard, with drill downs for each. More sophisticated chargeback models are also common, where business units can be charged based upon different calculations of resource usage. The most common is a tiered, flat rate model, similar to a cable bill, where larger servers are charged more. Another method, similar to an electrical bill charges by the amount of each resource actually used in a period.

Conclusion

IT is committed to making server virtualization work. But while the theory of consolidating server instances in an effort to improve resource utilization and increase flexibility is sound, the reality of implementation is different. The sheer volume of data generated by an increasing number of VMs is part of the problem, as too much detail obscures critical conditions. The sharing of resources that are essential to each VM and the abstraction that virtualization employs brings a new level of complexity to the environment, something not seen in the physical environment it replaced. IT needs tools designed for this new virtual world, not the ones it used in the ‘old world’, if it hopes to meet the expectations of internal customers and management.

Eric Slack, Senior Analyst