But with Virtualization, these lines are getting blurred.

A modern high-performance multi-core server today can easily host up to dozens of Virtual Machines (VMs) on the hosting hypervisor.  Within the hypervisor is a ‘virtual switch’, which allows the ability to network the hosted VMs as if they were separate hardware instances.

Although the flexibility and power that this grants the Virtualization Administrator is welcome, now the ownership of the last segment of the network is potentially ambiguous.

There are reason Virtualization(/System) administrators welcome this capability: They are closest to the applications, and they frequently leverage this awareness to better optimize the environment.

However, Network Administrators may not welcome this meddling in their back yard: When something goes wrong with the network, users are likely to blame ‘the network folks’ – even if the ‘network folks’ are innocent.

The answer isn’t to make the system less flexible and attempt to impose old, pre-virtualization paradigms on the new world order by encouraging the Networking folks to ‘take back control’ of the entire network all the way through the ‘last mile’ of the ‘network’ that only exists virtually (inside the virtual switch within the hypervisor.  Doing this only remove flexibility that is one of the key value drivers for virtualization.

A better approach is to encourage cooperation and education – both for the Network as well as the Systems teams.

A key valuable capability of Virtualization Administration is the ability to optimize the Virtual Environment based on information gleaned from the monitoring tools at your disposal. For example: If a Virtualization Admin detects there is a network imbalance between two of their VM hosts – one is in need of (‘starved for’) additional network bandwidth while another VM host enjoys a surplus – the Admin generally must negotiate with the Network administrator to secure additional bandwidth for the ‘starved’ server – which will likely require shutting down at least one of the servers to add an additional network card.

If the VM host is configured with 10GbE (becoming more common, and practically necessary for state-of-the-art high-performance multi-core servers hosting dozens to hundreds of VMs), the ‘unplanned’ request for an additional scarce 10GbE port may very likely be denied, owing to the high expense of 10GbE today,

Hence, the Virtualization Administrator is generally limited to managing and optimizing their I/O within a single VM host, even if they see an opportunity to optimize across multiple VM hosts.

There is a better way: Leverage I/O virtualization: Wire the servers once and manage and optimize the I/O across groups of servers.