Architecturally, hypervisor virtualization will likely work similar to the appliance-based storage virtualization that I discussed in the last entry, with the primary capability to make disparate storage devices be managed as one. Typically what will happen is that you will purchase a storage device, allocate all of it to the hypervisor, and then let the hypervisor manage the device from that point forward. While this is not the fine-grained virtualization approach that storage system-based virtualization products take, this method--especially in heavy virtualized environments--does deserve consideration.
Each device in this configuration is essentially it's own island of storage, but the hypervisor acts as a sophisticated bridge to those islands. You are essentially allowing the hypervisor to assemble a loosely coupled storage cluster, which allows the discrete components of that cluster to be different storage systems from different manufacturers.
Compared to a scale-out storage cluster where virtual machine images are written across the components or nodes of the cluster, the granular-ness of a hypervisor-controlled virtualization system is per node. In other words, an entire virtual machine image is contained on one particular node (and potentially mirrored to another). This means that performance of a hypervisor-controlled storage node is limited to the performance of the node it is on.
The "virtual-ness" of the system comes from this interconnection between the storage tiers that the hypervisor provides. Today this can be as simple as the ability to transparently migrate an active virtual machine from one tier of storage to another. In the future, expect hypervisors to understand the performance capabilities of the storage devices it controls and move virtual machine stores automatically between different storage devices. This automatically balances storage performance in the same way hypervisors can automatically balance processor utilization.
To embrace this type of storage virtualization through these systems, storage administrators are going to expect enterprise-class features like snapshots, thin provisioning, and cloning to work without performance degradation. While most hypervisors have these capabilities today, they do so at a significant performance loss. As we discuss in our recent article "The VDI Storage Trade Off - Efficiency vs. Performance," third-party applications are being made available to add these capabilities to current hypervisor technology.
The value in this type of storage virtualization is that point storage purchases are now manageable. You can buy a specific device to solve a specific problem without concern for its expandability. Typical mid-range and high-end systems that can scale capacity and performance have an added cost that customers pay dearly for to be able to deliver that scaling.
For example you could buy a low-capacity but high-performance solid state storage device for active, performance-sensitive virtual machines and a very low cost but high-capacity device for less active virtual machines. Then, leveraging the hypervisor, move live virtual machines between storage devices as needed. With the third-party storage application described above, you wouldn't even have to be concerned about feature set on the storage hardware, you could focus on cost, reliability, and performance.
The cost of storage is a major inhibitor to many server virtualization rollouts. In a virtualized environment, this strategy could dramatically reduce the cost of storage to support that environment while providing a great deal of flexibility to address performance demands as they occur.
Follow Storage Switzerland on Twitter