The concept of a virtual hardware platform is crucial to virtualization. It frees a workload from specific physical hardware devices so that the workload can run on any physical host with the proper computing resources available. It also allows key virtualization features, such as live migration, which moves workloads between physical servers with no downtime.
Benefits of a virtual hardware platform
Perhaps the most interesting attribute of a virtual hardware platform is its versatility, because an administrator can adjust the levels of each resource -- adding more memory and additional processor cores, allocating another VHD or assigning more network adapter ports, for example. Increasing resource levels will normally boost that workload's performance or responsiveness -- especially on older physical servers -- or allow the VM to support more users.
Conversely, an administrator can also remove virtual hardware from a VM. For example, an application may not be able to utilize a full 2 GB of memory or two processor cores. Removing excess resources will free those resources for allocation elsewhere, improving the performance of other busy workloads or increasing utilization on the physical server.
Generally speaking, an administrator will need to power down a VM before adding or removing resources.
Other virtual hardware platform uses
It is sometimes possible to allocate more memory to a virtual hardware platform than what is actually available from the physical host. For example, it is possible to configure a VM with 16 GB of memory on a physical server with only 8 GB of RAM. This process, called memory overcommit, allows more VMs to reside on a server, because most VMs don't utilize all of their allocated memory space. When physical memory runs short, the virtualization platform implements a virtual memory swap file on the virtual disk.
The practice of memory overcommit is well accepted, but there is still a serious performance penalty when using a swap file. Therefore, it's best to ensure adequate memory on the server.
This was first published in January 2010