Storage area networks and network attached storage are not necessarily rival technologies, and may even complement each other or merge into a single technology in the future.

Storage area networks (SANs) and network attached storage (NAS) are often positioned as rival technologies, with one seen as outselling or overcoming the other. The truth is that the two are a lot more alike than they are dissimilar.

In particular, both aim to do the same thing Ã,­ to split data storage from the servers and move it out onto a network. The difference is in how they achieve this and in the complexity of the resulting setup.

NAS uses storage appliancesÃ,­ complete file servers that can be attached to a network and made available for other servers to access, using standard protocols such as network file system (NFS). These normally share the same LAN as other servers, so the network needs to be at least Fast Ethernet and preferably Gigabit Ethernet.

Like other file servers, NAS appliances are file-based, which makes them particularly suitable for tasks that are heavily file-oriented such as Web cacheing and multimedia content distribution.

Compare and contrast
By contrast, the key feature of a SAN is that it uses a separate network dedicated to storage traffic. This is typically 1Gbit/s Fibre Channel, but it need not be, as a SAN could equally well be built on Ethernet or tunnelled over a WAN for long-distance data backup or mirroring. SANs are normally block-based rather than file-based, so servers can use the storage as they please.

Many storage experts feel that the need for a second network means SANs are suited only to centralised applications, whereas NAS would probably be better for a distributed business with a number of branch offices, especially as the branches are unlikely to have the IT staff needed to support the more complex SAN option. Even with a single site, size is also an issue, with a rule of thumb suggesting that only once an organisation needs to store 500GB of data or more will a SAN become cost-effective.

The other thing that distinguishes SANs today is that the storage devices-Ã,­ whether tape drives, disk drives or disk arrays-Ã,­ are directly attached to the network. Fibre Channel allows each device to be assigned to a specific server, with zoning added if necessary to protect devices from over-acquisitive systems such as NT4, which will attempt to seize all the devices they can locate.

The difference between the two technologies can also be seen as a tactical and strategic answer to the same question. NAS fixes the problems of storage being in short supply, for instance, and also consolidates it in one place when it can be protected properly. With the increasing speed of LANs, NAS can even provide networked storage that is faster than locally attached storage, according to Network Appliance's northern European marketing manager, Chris Gale.

What NAS does not do is relieve the LAN of that extra traffic, nor does it address the storage pooling issue. And because it works at file level, its performance may not match that of a block-based alternative.

Blurring the line between NAS and SAN

But it is still true that SANs and NAS are blurring. For example, it would be feasible to build a dedicated storage network on Ethernet, with storage capacity provided by NAS boxes. Would this be a SAN? In some respects it would, but its reliance on file-based access might suggest otherwise.

Similarly, file servers sold by Network Appliance attach to Ethernet LANs but also use SAN technology for backup. They use Fibre Channel disk drives internally and can be inter-linked, so that several NAS boxes can share a single tape library.

Gale said that it is a question of what you are trying to achieve. 'We use Fibre Channel on the back end Ã,­ it's a good way of connecting storage,' he argued. 'But it's not good at multihost connectivity. Are all hosts going to share the same volume? If so, who's going to write the file system?'

His view is that because almost every system supports NFS or one of a handful of other file-server protocols, this is the best way to serve and share files. The alternative would be to have the same file system on every server, allowing them all to share storage devices at the more efficient block level.

Several developments are under way in this area, including the directly addressed file system (Dafs), which will also allow applications to access memory without going through the operating system.

There are also concepts that extend beyond both SANs and NAS, such as the creation of a storage utility. This is an architecture where all management and design tasks are hidden as much as possible, so it delivers content on demand without the need to worry about where it is physically stored or the identity of the storage supplier.

This will demand much more sophisticated storage management software than is available today, however, to build storage utilities, make up for the skills shortage that exists as data volumes grow and counteract the growing dearth of experienced staff.

'Human resource issues are driving backup, storage consolidation, storage resource management and virtualisation,' said Tom Clark, director of technical marketing at Fibre Channel switch developer Vixel. 'If I had enough slaves to do it, I would have everyone doing everything by hand. The pressure is to reduce cost and exposure to risk, and with administrative talent in such short supply it's an amazing driver.'

Wayne Rickard, chief technology officer of Gadzoox, who was recently made the chairman of the Storage Networking Industry Association's (SNIA) technical council, agrees. 'We recognise as an industry that there is a pressing need for management tools to be improved. We hear IT managers saying that it would be nice to run them on Ethernet because they know how to manage that. They say that with server-attached storage there was nothing to manage. Now IT managers need to be able to take blocks of storage and link them in an interesting way to their servers, making sure the switches don't become new problems.'

The difficulty for users is that it will all cost money, according to IDC storage analyst Claus Egge. 'Users don't appreciate how much money they will have to stump up for new storage management solutions,' he warned. 'But the development money has to come from somewhere, and considering the shortage of storage managers, it's not unreasonable.'

Much simpler option

So have the NAS vendors been vindicated? It certainly seems that way Ã,­ the market is running in their direction and even the strongest supporters of SANs now acknowledge that NAS has a role in the enterprise.

As companies such as Network Appliance point out, NAS is by far the simpler option. It does not involve the installation of new networks, or many changes in the way that existing networks operate. By comparison, SANs look expensive and over-complex, requiring wholesale redesign of the systems infrastructure.

'NAS starts at the low end with a device like the Quantum Snap. It's really just a network-enabled hard disk, so it doesn't need a PC server Ã,­ it's self-managed and it's cheap,' said Egge. 'Conversely, a SAN means lengthy discussion of how to implement it, and what hardware and software to buy.'

One result is that SAN vendors have become increasingly conciliatory towards NAS, to the extent that the SNIA includes both sides among its membership. They are working towards a way of unifying the two Ã,­ or at least defining a spectrum in which they can both fit.

'The SNIA technical council took on the job of creating a layered model for storage that would take in SANs, NAS and virtual storage,' said Rickard. 'At the end of the day we agreed on more than we disagreed. It was just a question of agreeing on language.'

The SAN zoning concept is now being expanded to enable storage virtualisation, where storage is pooled and made available to the servers as pure SCSI blocks. This does away with the need to allocate physical devices to specific servers and enables dissimilar devices to be mirrored.

Virtual storage relies on there being a control device somewhere in the SAN. This is the only SAN node that sees both the storage devices and the servers as they really are Ã,­ everything else merely sees what it is permitted to see.

In symmetrical storage virtualisation, the control device sits astride the data path and all data requests must pass through it. But there can be multiple devices, both for redundancy and to provide improved performance. Symmetrical virtualisation is typified by devices running Datacore's SanSymphony software, such as the Gadzoox Axxess box.

Asymmetrical virtualisation places the control device alongside the data path. This allocates storage blocks to the servers, but the servers must then keep track of these allocations themselves, using modified host-bus adapters (HBAs) capable of storing the necessary block map.

The best example here is perhaps Compaq's VersaStor technology, which has attracted support from a number of other key SAN companies, including HBA developers such as Emulex, JNI and QLogic.

The symmetrical scheme has the advantage that it can work with today's SAN hardware, whereas VersaStor must await the development of new hardware. However, asymmetrical virtualisation could offer higher reliability and performance.

Copyright © 2009 CBS Interactive, a CBS Company. All Rights Reserved.
ZDNET is a registered service mark of CBS Interactive. ZDNET Logo is a service mark of CBS Interactive.