Virtualization Track at Cloud Expo

Virtualization is the key technology for the cloud. Its ability to separate the OS and application from the hardware enable it to best deliver on-demand cloud services. Charles King, Principal Analyst at Pund-IT, said it best: "Without virtualization there is no cloud - that's what enabled the emergence of this new, sustainable industry." But, how can IT organizations leverage virtualization to create their own private cloud?

Virtualization Evolution: The Three Stages
Typically, organizations advance through three stages of virtualization as shown in Figure 1. The first stage is characterized by virtualizing the "low hanging fruit." Consolidation and disaster recovery strategies are initially focused on garnering the capital expenditure savings from virtualizing applications that have low business impact.

The second stage of virtualization is when things get tricky due in large part to the complexity created in stage one. This is where most enterprises are today. Many CIOs now have a "virtualization first "policy, requiring IT to virtualize for the cost benefits. Yet, IT is concerned about how to maintain high service levels. In this second stage, organizations may have started to use IT resources, including applications, servers, storage, and networks, as pools of resources that can be managed in aggregate rather than isolated silos. Thinking about IT resources as an entire system becomes important for fully leveraging the power of virtualization as it allows IT to move resources on demand to efficiently balance computing loads and use capacity more efficiently. The issue of performance requires attention at this stage as well. Organizations may have experienced unexpected poor performance in stage one and know they will need help to ensure performance of the business-critical applications (Exchange, SQL, SAP) that are virtualized in stage two and beyond.

Thus, it's critical in stage two that organizations obtain the right management tools beyond what comes "out of the box" with their virtualization software. Tools, such as infrastructure performance management solutions, that can help enterprises optimize while they virtualize are essential. These tools enable IT to get the visibility and control they need to troubleshoot and ensure performance, optimize performance and capacity, plan server and storage requirements, and manage service levels to advance from stage two to stage three.

Stage three is characterized by virtualizing mission-critical applications and a transformation in which virtualized resources shift to become a pool that is managed at the pool level. This allows workloads from different business units to become fluid. At this point, infrastructure is shared and IT must embrace a service management focus to deliver the private cloud in which physical and logical resources are made available through a virtual service layer across the enterprise.

Figure 1: The three stages of server virtualization evolution

Mastering Stage Two: Focus on Performance and Capacity Management
Let's take a closer look at what's required of infrastructure performance management solutions that can enable the move to stage three and the private cloud.

1. Troubleshooting Virtual Infrastructure
IT administrators should look for infrastructure performance management tools that provide cross-domain visibility and analysis for effective troubleshooting across their entire pool of resources. Effective root-cause analysis requires a comprehensive view of all elements within the infrastructure including automatically mapping each element and its interdependences from application to virtual and/or physical server to storage array disk group.

This category of tools must also have an Infrastructure Response Time (IRT) metric to troubleshoot performance issues (see Figure 2) Defined as the time it takes for the system to perform work submitted by an application, IRT directly shows how responsive the system is. IT infrastructure owners can use this metric to quickly demonstrate to application owners whether poor end-user performance is due to infrastructure issues, such as resource contention or hot spots, or something outside of the infrastructure. Problem resolution that used to take hours or days can be reduced to minutes.

Figure 2: Infrastructure Response Time is a metric used to quickly identify where infrastructure issues exist

For example, it's not uncommon for organizations to use "virtualization best practice" planning for consolidation, but be unable to fully assess the storage requirements and how they factor into performance problems. One organization that migrated their storage "as is" ended up having phantom slow-downs that resulted in finger pointing and multiple fire drills. After installing an infrastructure performance management solution that automatically mapped the topology and highlighted bottlenecks and configuration problems, they were able to show and prove that the data store was the bottleneck and that the IO was not being delayed at the array.

2. Optimizing the "System"
IT administrators need to match virtual machine (VM) densities to application needs, optimizing performance and utilization. True infrastructure optimization requires insight and analysis into both utilization and performance - virtualization makes this challenge more difficult. Focusing on any one of the two is incomplete, but a good infrastructure performance management tool provides the combination, allowing organizations to maximize infrastructure performance at the lowest possible cost. These tools use analytics to collect data from each element (VM, server, and storage array), perform modeling and analysis, and create unique data center level statistics. The resulting performance indicators score the balance between the application requirements and the infrastructure's ability to deliver. Performance indicators are produced from both VM and virtual host perspectives, providing the opportunity to maximize VM densities, appropriately size VMs, and accurately determine headroom for each application. They can also ensure optimal resources to groupings of applications.

Determining Optimal Sharing in Resource Pools
VMs may belong to an identified resource pool. Many virtual server implementations have not yet taken advantage of the benefits of defining finely grained resource pools. Dividing up a large server or server cluster into smaller resource pools enables ensuring optimal resources to groupings of applications. Each pool's member VMs effectively share the pool's resource entitlement. (A VM's entitlement settings are evaluated within the pool's entitlement.)

Resource pools make it easy for IT to allocate resources to specific business units that might run many virtual machines. Each business unit can then decide for themselves how to best share their resource pool among their own applications through VM-level settings. This method guarantees resources to each business unit and avoids forcing all the VMs in a large cluster to compete head-to-head. Resource pools can insulate business units from those that tend to "hog" resources or guarantee resources to departments that are funding them without having to dedicate the resources in isolation.

Other useful resource pool strategies may group applications by type, by function, or by other relevant attribute (e.g., user geography/time zone, administrative control, business priority, or service-level objective). Think of resource pools as a set of "school playground" rules for virtual machines, dividing up the playground into different spaces for each class to use as they see fit while allowing for common (or currently unused) areas to be shared according to some "fair" rules.

Infrastructure performance management solutions not only analyze and chart each resource pool's overall utilization by VM, but show the resource pool's utilization against the pool's entitlement (see Figure 3).

Figure 3: A good infrastructure performance management tool can ensure optimal resources to groupings of applications

If a resource pool is consistently operating above its entitlement, it may indicate a need to examine its member VM's actual requirements and entitlement settings. If a resource pool is operating significantly below its entitlement (and specifically its reserve), there may be opportunity to reallocate or "better share" underlying resources.

1. Planning Capacity
IT administrators must do capacity planning in order to accurately forecast appropriate capital purchases. Efficient capacity planning delivers IT resources to applications when they need it and without over-provisioning. An infrastructure performance management solution can provide datacenter level statistics to predict performance and capacity growth. Look for a tool that provides automated scorecards on VMs, servers and storage to easily see a consistent set of metrics for all IT resources. This type of tool can be leveraged as an independent advisor, enabling IT managers to only invest in capital equipment when they truly require it, not when an IT vendor says they do.

A typical example comes from an online automotive marketplace company. They needed to plan how many ESX servers per VM cluster they needed to support company consolidation and growth. The infrastructure performance management solution provided information on CPU, memory, and storage usage rate changes, high water marks, and trends. It showed the ESX servers that were over- and underutilized. As a result, the team was able to mix workloads to get higher VM densities, minimize vMotions to put less stress and background IO on the storage arrays, and quickly justify budgets to accelerate virtualization.

2. Ensuring Service Levels
IT administrators must manage service levels in order to meet business requirements. Business users demand service-level agreements but virtualization's abstraction of the physical infrastructure and underlying resource contention make this challenge more difficult. An infrastructure performance management solution that provides administrators with system-wide visibility and analysis is a must-have for ensuring performance of virtualized business-critical applications. As discussed earlier, IRT shows the performance delivered to an application by all of the resources assigned to it. It can be used to baseline and predict future service and even alert on service deviations.

A tool that provides chargeback reporting allows virtualized application costs to be accurately tracked by VM, server and storage, rather than limited to any single element. Look for a tool that provides automated reporting of system-wide configuration data and historical performance and utilization data rolled up into reports for administrators and their managers.

A local telephone company in the process of deploying two new virtual farms that were growing daily used an infrastructure performance management tool to baseline and compare performance before and after virtualization. With the tool, they were able to demonstration to their customers that they received the same (or better) level of service once their applications were virtualized.

Getting to Stage Three: Optimized Virtual Infrastructure that Is Ready for the Cloud
Virtualization has enabled IT organizations to cut costs and improve efficiency, leverage new IT investment, and improve service. As companies advance through the stages of server virtualization adoption, they need an infrastructure-wide performance management solution to provide visibility and control to troubleshoot and ensure performance, optimize performance and capacity, plan server and storage requirements, and manage service levels. Infrastructure performance management provides the predictability and confidence that enables IT organizations to meet business user demands and provides the tools to easily manage and report on those requirements. Organizations that do not implement such tools simply cannot effectively leverage the efficiency, agility, and scalability that can be achieved in private cloud computing.