To properly plan for any migration to virtualization, or virtual infrastructure startup, proper capacity and resource planning is paramount. The hardware upon which the virtual platform is to be installed must exceed the processing power, memory capacity, and disk I/O required for the virtual machines as tasked.  Although the ability to dynamically allocate most resources “on the fly” is a benefit of virtualization, a proper measurement for peak usage must be established. Once this is established the need for additional headroom past this point has to be added to hardware requirements thus allowing for future increased demand. Some might say that if the time comes when present hardware resources are no longer able to sustain the workload of the virtual machine(s) new hardware should be purchased to either:

1.     Create similarly tasked virtual machines to share the load.

2.     Migrate the existing virtual machines over to.

Arguments can be made for either scenario but this impasse should never be reached due to poor planning as the negative economic and production impact could result in repercussions that affect business continuity or sustainability.

The problem with industry standard server benchmarking tools is that they don’t provide the information necessary for virtual machines tasked with a single application or consider the virtual machine capacity of the host machines hardware. Virtualization benchmarking applications need to be specifically written for these tasks and need to be cross platform (accommodate various hypervisors and hosts) capable.

The Standard Performance Evaluation corporation (SPEC) has recognized the need for standardized virtualization benchmarking in establishing the SPEC Virtualization Committee with the following goals:

 ·      Delivering a benchmark that will model server consolidation of commonly virtualized systems such as mail servers, database servers, application servers, web servers, and file servers.
·      Providing a means to fairly compare server performance while running a number of virtual machines (VMs).
·      Producing a benchmark designed to scale across a wide range of systems.
·      Supporting hardware virtualization, operating system virtualization, and hardware partitioning schemes for server consolidation scenarios.

To this date a benchmark has not been developed by the committee. However, one of its members (VMware) has.

 

VMware VMmark is a freely available benchmark that was created with the goal of “open standards & platform neutrality”. VMmark is designed to deliver accurate performance valuations using testing methods that allow for real word benchmarks of virtual machines tasked with running specific applications common in the industry. Some of the typical workloads evaluated are virtual machines tasked as web servers, database servers, and java servers.  VMmark can properly evaluate server capacity by using a “tile based implementation” to benchmark a group of virtual machines tasked for various purposes as would commonly be present in the typical data center.  Results of the benchmarks can tell with reasonable certainty the capacity (in number of virtual machines) of a particular hardware platform as configured.

The virtual infrastructure is capable of being very dynamic in resource allocation by shifting resources where and when they are needed. Although the hypervisor provides these functions, many resources are finite due to the hardware configuration of the host machine. To plan a virtualization strategy we need to draw from benchmarks that can accurately gauge the capability and capacity of hardware to maintain mission critical applications in the data center.