This content is part of the Essential Guide: The ultimate network-attached storage guide

Essential Guide

Browse Sections

Using NAS devices for data center virtualization storage

Network attached storage (NAS) devices are finding a role in data center virtualization storage, whether it be for quick swap space or for a central library of virtual machines.

If there is a common theme that runs through all virtualization environments, it is the need for data center virtualization storage space. Despite the move toward Storage Area Network (SAN) technology, Network Attached Storage (NAS) meets this need for space in many an enterprise data center.

As the number of virtual machine images multiplies in a data center, hundreds of gigabytes of space are consumed. Likewise, as virtual machines move between physical servers in order to maximize the efficiency of the entire environment, a shared network medium between those servers allows for a quick handoff and transition.

In a virtualized environment, NAS appliances can serve as swap space to move virtual machines between servers, act as a backup medium, or play the role of central repository for all virtual disk images. In any of these scenarios, data centers and network administrators need to understand the role of NAS appliances and their implications to the network.

Why NAS over SAN for data center virtualization storage?

There are two major approaches to network storage: NAS and SAN. The two approaches vary in both network architecture and how each presents itself to the network client. NAS devices leverage the existing IP network and deliver file-layer access, abstracting the physical disks available to it and presenting a compatible file system in the form of a network share to the end client machine using protocols such as CIFS or NFS. NAS appliances are optimized for sharing files across the network because they are nearly identical to a file server.

Conversely, SAN technologies, including Fibre Channel (FC) and iSCSI, deliver block-layer access, forgoing the file system abstractions and appearing to the client as essentially an unformatted hard disk. FC, the most prevalent SAN technology deployed today, operates on a dedicated network, requiring its own FC switch and host bus adapters (HBA) in each server. A complimentary emerging standard, Fibre Channel over Ethernet (FCoE), collapses the storage and IP network onto a single converged switch, but still requires a specialized converged networking adapter (CNA) in each server. Another block level technology, iSCSI, encapsulates SCSI commands within standard IP traffic and can use existing Ethernet network interface adapters, but is typically complimented with a TCP/IP offload Engine (TOE) to optimize performance. SAN solutions have a traditional advantage over NAS devices in terms of performance, but at a cost of some contention issues. A partitioned area in a SAN array can be shared between two machines, but both machines treat the space as their own, creating the risk of contention between the two machines. While there are solutions available to address the contention issue, the fix adds an additional layer of abstraction -- something that is already built into NAS solutions.

Both types of storage have a role in the virtualized data center. Microsoft's Exchange server, for example, only supports network storage of its data through block level access. SANs typically offer higher performance than NAS solutions, so enterprises will utilize storage networks for high I/O applications, such as databases. For the purposes of actually managing the library of virtual machine images, however, file level access is most likely the simplest approach to deploy in many environments. Depending on the organization, data center managers might simply have to request file level storage from the storage team or actually install a dedicated NAS appliance for their data center virtualization storage needs.

Network design considerations for NAS in virtualization

Fortunately for networks administrators, NAS devices rely on the same file sharing and transfer protocols that most file servers utilize, so the design approach to deploying NAS is very similar. Given the nature of the virtualization environment and the large file sizes involved, however, particular attention needs to be paid to delivering those virtual machine images to their intended server target as quickly as possible.

Network administrators should approach a NAS deployment by ensuring that all connections involved are at least using Gigabit Ethernet. They must also verify that the network switches connecting the virtualized environment are at adequate subscription levels for the number of ports in use,giving enough bandwidth to the storage appliance. Administrators should also reassess any Quality of Service (QoS) policies that may impact the links between the servers and storage and adjust if necessary to give additional priority to the increased storage traffic. Setting network baselines both before and after the NAS deployment will help ensure a clear view of the impact of adding a NAS device to the data center network. If performance becomes a problem, segregating the storage traffic on the network, either through VLANs or dedicated switches, is also available as an option.

Network administrators should also work closely with data center managers to performance test and compare each of the network protocols available on the NAS device, typically NFS and CIFS. The implementation of these two standards can vary greatly among network devices, operating systems and virtualization solutions. Given these variances among platforms, as well as the organization's unique needs, performance testing of each available option will help determine the best solution for your particular environment.

Alternatives to NAS for virtualized environments

While many consider NAS as the cleanest approach to support virtualized environments, there are a number of alternatives available. For organizations that already have a SAN installation, hybrid NAS-SAN solutions are available that build the file system abstraction on top of space available from the SAN. A hybrid solution can create the abstraction layer that data center managers prefer, while maintaining the performance and manageability of a centralized storage network. Smaller organizations could also consider sharing space on the physical server direct attached storage, using one of the servers as a central repository, with the rest grabbing the VM images from there. While this solution may function in the smallest of environments, it could quickly turn in an I/O bottleneck as it grows in a larger one.

Moving forward

Finally, network administrators and data centers managers need to be concerned when deploying a NAS device in the data center about potential growth that is too large for the technology. The role of network storage can quickly expand, opening the door to I/O bottlenecks and other constraints. A quick fix for the virtualized environment could rapidly snowball into a piecemeal enterprise storage solution, particularly in smaller organizations. It should be an imperative that even a small NAS deployment is followed up with a discussion about a more encompassing network storage strategy for the enterprise. Depending on the enterprises needs, network storage could grow organically with larger scale NAS solutions or migrate to a SAN solution if higher throughput is required. Any network, whether data or storage, can easily become an unmanageable collection of point solutions if no consideration is given to the overall design.

Dig Deeper on Content delivery networks (CDNs)