OPTIMIZED SYSTEMS

Protecting Mission-Critical Workloads

John Jainschigg | Date: 09-13-11 | Comments

Before running mission-critical applications in a highly virtualized environment, organizations should use performance and cost-analysis tools to evaluate whether they are proper candidates for virtualization.

To ensure mission-critical workloads are available in a virtualized data center demands meticulous attention to application performance requirements, resource utilization and other technical characteristics. Such information is needed in order to properly identify workload candidates for virtualization, manage their moves effectively, ensure resilience and preserve continuity.

As should be clear, candidate evaluation should begin even before you select and implement final solutions for compute, storage and other functions, since evaluation will reveal much about required solution scale and configuration. Evaluation will also help you manage stakeholder expectations about what can be virtualized safely and cost efficiently. Hitting the right mark here can help make certain that you don�t unexpectedly run up against virtualization limits or endanger mission-critical applications. This step is also helpful to ensure that your implementation achieves more general cost, resiliency and performance goals.

The basic process is to list all instances of all applications, then evaluate them for resource utilization and requirements in fine detail--including memory, CPU and I/O, as well as security and resiliency requirements. It�s vital to perform these tests systematically and repeatedly over time to determine peak demands and identify potential sources of contention. Luckily, many tools are available to make this job easier and that help document the process. IBM, for example, offers server- and application-discovery, evaluation and constraint-documentation tools, as do major hypervisor vendors (e.g., VMware), as well as third parties like Xcedex.

Most software solutions reduce their computations to an index value, where servers indexing at or below a certain threshold are prime candidates for moving to virtual machines. Those workloads showing higher values are flagged as more problematic. Being flagged doesn�t necessarily mean that a workload can�t be virtualized; it only means that additional evaluation is needed.

Some workloads--among them, perhaps the most mission-critical--may be determined in this way to be non-virtualizable. This is not a best-case scenario, but it is fairly common. Applications with very high I/O, memory or CPU requirements may best be kept on their own servers to avoid their edge-cases disrupting what would otherwise be a fairly "flat" and cost-effective virtualization solution.