As server virtualization technology continues to become mainstream, more and more server workloads will become candidates for virtualization. In some enterprise environments 100% of workloads may be good candidates. However determining whether a workload should be virtualized requires careful workload sizing and analysis. There is no hard-and-fast rule for which workloads should be virtualized but there are some common patterns for determining which workloads should be excluded or at least analyzed more closely. The three main patterns for exclusion are described below.
A. Large Workloads
Workloads that require excessive amount of processor, memory, disk and/or network resources may be candidates for exclusion because:
1. These workloads will consume a disproportionate amount of resources on virtual machine hosts. This means that fewer virtual machines would be able to run on each host on average and as a result the cost per VM would increase significantly. In other words, the cost of hardware, virtualization software and management applications would be amortized among fewer virtual machines.
2. Mobility of large virtual machines is much more limited compared to that of small VMs. For example, if you are planning to use VMotion or XenMotion, finding a server with sufficient headroom to host a large virtual machine is non-trivial. Furthermore, migrating larger VMs results in higher overhead and may be disruptive to other workloads.
B. Highly Variable Workloads
Workloads that have high resource consumption variability, especially for processor and memory, may not be good candidates for virtualization because:
1. These workloads may starve peer workloads and affect service levels of business applications that depend on these workloads.
2. If resource optimization technology such as VMware DRS is used, then highly variable workloads may cause frequent virtual machine migrations among hosts to compensate for variability. This in turn will result in higher overhead for virtual machine hosts and may be disruptive to other workloads.
C. Rapidly Growing Workloads
Rapidly growing workloads may quickly outgrow virtual machine environments and cause similar issues as larger workloads. Instead of looking at the absolute size of each workload, it is essential to consider how much each workload will grow over 6-12 months to decide whether it is a good candidate for virtualization long-term.
Recommendations
To determine which workloads should be virtualized, it is essential to start every virtualization project with measurement and analysis of existing workloads. In absence of such an assessment, the cost of a virtual infrastructure and deterioration of service levels may dramatically reduce its benefits.
