Storage For Virtual Environments

The prospect of live virtual machines scooting among physical hardware--even across various data center locations--is unsettling to storage pros, who typically design around static resources. But we need to get over our fixation on fixed assets--server virtualization is, without a doubt, the most impactful infrastructure trend for today's enterprise data center architects. The fact that server virtualization also poses challenges to longstanding architectural assumptions means our assumptions need to change.

In the case of storage, that means demanding greater flexibility, performance, and integration capabilities.

Fortunately, storage and data networking vendors are responding to the demands of virtual environments. Storage arrays are getting smarter, and a new generation of network convergence protocols and products is based on Ethernet, especially the current 10-Gbps version, which will outperform 8-Gbps Fibre Channel immediately.

Storage Virtualization's Role

Enhancements to the presentation of storage capacity, as well as the I/O channels that carry it, demand advanced capabilities within enterprise storage arrays. Features like thin provisioning, automated tiering, snapshots, and replication are in demand by virtual server architects, and it is these capabilities that will keep enterprise arrays attractive to buyers.

The high level of performance required by consolidated, virtualized environments depends on the use of high-end storage area networks and network-attached storage systems. Though they also boast advanced features, these cost more on a per-capacity basis than the direct-attached storage often used for nonvirtualized servers. Many IT pros have argued that server virtualization drives excessive growth in the use of storage capacity--the ease with which new virtual disks can be provisioned, copied, and snapshotted often leads to virtual disk sprawl, with primary storage used for unnecessary copies. This combination of expensive capacity and high usage can drive the cost of storage unacceptably high.

The answer is technologies that add capacity efficiency--for example, thin provisioning, which is widely available on the high-end storage systems typically used to support virtual server environments. By allocating capacity on demand (and deallocating it when it's no longer used), systems using thin provisioning boast far better capacity utilization. This reduces the effective cost of storage.

Data protection is another critical feature for supporting virtual environments. Both Microsoft's VSS and VMware's VADP allow for off-loading the creation of snapshots to the array. These frameworks also support native software-based snapshot creation if array support isn't available, but there's a massive difference in performance between these options. Many enterprise arrays use "delta" or "copy on write" technology to store only differences between a snapshot and a primary data volume. That allows these storage systems to create snapshots in moments and to store them for long periods without much performance or capacity impact.

Storage arrays also boast native data replication capabilities that can be leveraged by virtual servers. In particular, the ability of virtual machine images to run on dissimilar hardware without changes is extremely powerful when architecting disaster recovery systems. Since these images are self-contained and stored on disk, array-based replication is a prime technology enabler for data recovery.

Another key storage trend for virtual environments is the development of virtual storage appliances, or VSAs--storage arrays that exist entirely in software. Often leveraging open source operating systems and storage stacks, VSAs offer a quick and effective route to implement shared storage for virtual machine clustering and similar applications. They have proved popular in development and test environments, and offerings suitable for production deployment, such as those from FalconStor, HP LeftHand, and Nasuni, are rapidly gaining attention in the industry since they can be configured to "follow" virtual machines as they move throughout a virtual data center or as part of a disaster recovery scenario.

Ethernet's The Future

Ethernet's road map means it's a natural for virtualized, converged networks. Future enhancements will boost performance by an order of magnitude, leaving all competing protocols in the dust. And standardizing on Ethernet brings an economy of scale that will reduce cost as well.

While iSCSI was designed to leverage TCP/IP for reliability, moving Fibre Channel to Ethernet requires significant upgrades. This was the motivation behind the creation of Data Center Bridging extensions to add guaranteed and predictable delivery of Ethernet packets. Networking vendors quickly added support for priority flow control (802.1Qbb), bandwidth management (802.1Qaz), and a Data Center Bridging Exchange protocol to communicate capabilities. Further enhancements will bring widespread use of congestion management technology as well as new network topologies for enhanced performance and flexibility.

For now, though, combining data and storage networking over Ethernet requires special converged network adapters, and major vendors have responded to this need quickly. The latest models from Brocade, Emulex, and QLogic offer solid DCB support for server hypervisors from Citrix, Microsoft, and VMware. Meanwhile, Intel is set to release a software stack for its 10-Gbps Ethernet adapters that promises broad DCB and Fibre Channel over Ethernet support in the future.

Flexible I/O requires more than convergence, however. Networks must be able to rapidly adapt when systems move, and this requires virtualization of hardware interface identifiers.

Fibre Channel flexibility is enhanced with N_Port ID virtualization, which "names" a virtual machine rather than a physical interface. A similar technology inside servers, Virtual Machine Device Queues, correctly allocates network traffic to virtual machines. Much development in network virtualization remains to be done, but exciting technologies like OpenFlow promise fully programmable, flexible Ethernet networks.

No Going Back

The growth of server virtualization is driving network storage deployments, no question. But it's also an opportunity to advance the state of the enterprise storage art. Never before has an application requiring such extreme levels of flexibility, integration, and performance seen such widespread use.

As enterprise storage professionals embrace virtualization and convergence, they must also become actively involved in developing advanced architectures. Get a seat at the table when virtual data center plans are developed, because new technologies such as virtual desktop infrastructures and consolidated or converged networks affect the storage infrastructure.

Storage is both a requirement for and a key enabler of highly virtualized server architectures, and much attention and development effort are being focused on storage devices tailored for these environments. These next-generation arrays must present storage in a manner appropriate to the demands of the hypervisor, address convergence trends, and offer new features like thin provisioning and replication, so storage pros must also stay on top of what vendors are doing. The world of enterprise storage will never be the same--and that's a good thing.

Very Virtual

Storage for Highly Converged Networks
Our full report on storage for highly converged networks is free with registration.

This report includes 17 pages of action-oriented analysis and 8 charts from our IT surveys. What you'll find:

Get This And All Our Reports

Stephen Foskett is a consultant and writer. Write to us at [email protected].