Gartner briefing: Know when and where to reach your virtualisation ceiling

Virtualisation enhances flexibility and agility by detaching workloads and data from the functional side of physical infrastructure. Although attractive possibilities abound for networks, storage, servers and desktops, virtualisation is not always the right option.


Virtualisation enhances flexibility and agility by detaching workloads and data from the functional side of physical infrastructure. Although attractive possibilities abound for networks, storage, servers and desktops, virtualisation is not always the right option.

IT managers and planners need to know where its implementation could create problems.

Gartner generally advises against implementing virtualisation in situations where this technology could create problems such as performance bottlenecks, inability to meet mandatory performance and security baselines, and commercial support and licensing issues. In these cases, Gartner’s advice is to avoid virtualising the workload or application, or to wait for and review improvements to hardware and hypervisor technology.

Gartner advises avoiding virtualisation in circumstances that feature:

Process/physical constraints

Some applications require physical hardware events to occur and to be recorded with time stamps. Hypervisors and virtualisation technologies cannot guarantee these event triggers and time stamps, which can impair the accuracy of measurement and the billing of resources.

The need for physical commitment will restrict the number of process-constrained applications that can be virtualised, with virtualisation being unsuited to, for example, Integrated Lights Out (iLO) technology, branch applications in the retail and manufacturing sectors, and physical devices such as USB equipment, including copy protection dongles. Real time applications, time stamps and abstraction still don't mix.

Transactional constraints

Networking and storage throughput remains a virtualisation challenge. In these cases, more hardware assistance is required initially, with direct-path input/output (I/O) (hypervisor bypass) and even multipath I/O (for concurrency). Overall, high usage and/or the lack of ability of I/O-intensive workloads to be virtualised drive transactional applications. As the capabilities of hardware assistance for virtualisation improve and become more widely deployed, more I/O-bound workloads and applications will become virtualised and I/O-intensive.

Commercial and support constraints

This class of applications is restricted by commercial agreements already bound up with the application portfolio. Although many of these commercial issues could probably be overcome by the efforts of independent software vendors (ISVs), the issues continue to limit the number of applications that can be virtualised in production and to require more investment in the test/development phase of virtualisation projects. Licensing, support and break/fix procedures require parallel environments for production and testing/development, so they should only be virtualised with caution.

Organisations that lack mature processes for performance monitoring and dynamic provisioning of virtualised workloads, and that have fast-growing applications, generally make poor virtualisation candidates. Without the processes and tools to recognise when increases in demand for resources by a specific workload exceed preset parameters, that workload could quickly become performance-constrained.

Other factors to consider when deciding whether to virtualise workloads are:

Lack of support

This remains a problem for many applications because software vendors often will not support applications when they are installed in virtual machines (VMs). Gartner clients report that some vendors still require them to recreate problems on physical hardware before they will provide support. This represents an additional administration overhead that may be unacceptable, and this becomes a concern especially when consolidating large multicore x86 hardware deployments.

Licensing constraints

Licensing restrictions can inhibit virtualisation. Applications in which licensing or pricing is based on measures of physical capacity — such as the number of cores or hardware platforms — can present administrative and cost penalties. Such applications are often a poorly suited to virtualisation management tools that permit live migrations of VMs, as they require every physical resource they touch to be licensed. However, some vendors, such as IBM, license by processor value units and support live migration through sub-capacity metering.

Mission and safety-critical applications

These applications are often poor candidates for virtualisation because the potential cost/efficiency benefits of this approach can easily be outweighed by the risks — additional layers of software mean there is more to go wrong generally, while consolidation with other VMs means there are more things to go wrong on the platform. However, some disaster recovery and high availability extensions, such as VMware VMotion, and live migration on top of virtualisation, can mitigate these problems.

Lack of internal skills

Organisations that have yet to develop expertise in managing virtualised resources should regard applications with highly sensitive or business-critical data as poorly suited to virtualisation. Without appropriate virtualisation management skills, processes and tools, trusted and untrusted workloads could end up combined on the same server, so making sensitive data more vulnerable to exposure than if not virtualised. This problem might be overcome by using virtualisation technologies hosted and/or managed by an external service provider in the cloud.

Whichever constraints you are checking for, it’s important to remember that virtualisation involves substantial upfront investment and is therefore easier to do during the test/development phase than in the production phase, as long as test/development budgets and costs are factored into the decision. Remember — the more you virtualise, the more the remaining physical workloads become the problem.

"Recommended For You"

An application-centric approach to virtualisation Next-generation server virtualisation