Storage area networks are redefining the way data is managed within today’s businesses. Find out how SAN will change the way you work.
Although most people seem to agree the industry is suffering through a slowdown in IT spending, somebody forgot to tell the storage vendors.
Driven by soaring customer demand and the rapid maturation of their solutions, storage companies have been busily building the products and inventing the standards that are fundamentally redefining the way data is stored and managed within the enterprise.
That way is through storage area networks (SANs)—dedicated networks that aggregate hard disk space, tape libraries, high-speed interconnects, and redundant design to produce a scalable storage infrastructure that’s redefining the way data is managed within today’s businesses.
The perils of proliferation
Although it’s been clear that today’s companies would be lost without their data, storage technologies had taken some time to recognise that fact. Until recently, most companies relied on DAS (Direct Attached Storage)—the hard drives installed in a server. If you needed more storage, you added new hard drives to the server. Server full? Just add another.
That approach may have worked fine in the past, but today’s storage growth rates simply don’t allow it anymore. IDC, for one, reported that Asia-Pacific disk capacity increased 93.6 percent, or 14,325 terabytes (TB), between the end of 2000 and the end of 2001.
Contemplating that kind of growth is enough to give any storage architect the shakes: DAS becomes a management nightmare as environments gain dozens or hundreds of servers, each handling different applications. Data becomes distributed amongst the servers, with no way of knowing what data is where. Empty space on one machine stays that way, even if another server is choking on its data load and could desperately use a bit more room to move. Backup is difficult, with each server typically backed up using its own tape drive.
Recognising the futility of the DAS approach in the long term, the past four years or so have seen a steady growth in the use of NAS (Network Attached Storage). Pioneered by still-dominant vendor Network Appliance (NetApp), NAS devices are large storage arrays that hook into a network on their own right. They include an operating system kernel—often based on proprietary operating systems or on Linux, but increasingly on Windows 2000—that finds and mounts any hard drives inserted into the NAS chassis.
Using the operating system’s built-in file system, those drives are configured to look like any other server. But because they’re built for storage, NAS devices typically accommodate hundreds of gigabytes of fault-tolerant storage. Need more space? Just add another drive, and it will automatically be given its own drive letter and made available to the network.
While it’s an improvement on DAS, NAS has presented some of the same problems to users as its predecessor. Its file system-based design makes for easy integration into enterprise networks, but the NAS architecture offers few other advantages over DAS. Data management is still difficult, particularly as NAS devices proliferate across the business and require their own backup and management strategies. Here again, tape is the inevitable method for backing up. And while you might end up with fewer tapes to manage with NAS than DAS, you still have to manage them. Sheer volume of data is also a problem, since in a backup all NAS data runs across the network—from NAS device to backup server to tape drive. That means NAS backup can put a major drain on network bandwidth—slowing the performance of enterprise applications and forcing administrators to keep backups outside of normal operating hours.
The SAN solution
Although they provided better scalability than DAS, NAS devices offered little that was revolutionary in terms of storage architectures. That revolution came a little later, as customers’ insatiable demand for storage drove the market towards an architecture where storage and backup tasks were handled on a completely separate network.
Structured like a conventional local area network, SANs use high-speed connections to link dedicated storage arrays. SANs can also include a range of related data archiving equipment—for example, large tape libraries providing automatic management of backup tapes. If more storage capacity is needed, additional SAN boxes can be easily added.
Because SANs aggregate all data in one place, they offer a far easier data management paradigm. Backup software doesn’t have to chase data all over the corporate network, and it’s relatively easy to see exactly how much space is being used. And instead of having their own storage space (as with DAS and NAS), applications share a common storage repository.
“The SAN is almost like a mainframe: the data is there to stay, and you just point your application there and let it process against that data,” says David Solsky, director of sales and marketing with SecureData Group, which recently helped the National Library of Australia set up a 2TB SAN to support DOSS (Digital Object Storage System), a major project involving the digitisation of the library’s more than five million books and 500,000 photographs.
The DOSS SAN, which combines an EMC CLARiiON FC 4700 storage array and Brocade Fibre Channel switch, has been designed to quickly scale up to an estimated 20TB; SecureData is already in the process of adding another 3TB of space.
“If they’d used DAS every time they wanted to do an upgrade, it would have meant downing the server,” says Solsky, adding that SAN projects account for half the company’s work these days. “[With a SAN] if servers reach the end of their lives, customers can plug a new server into the SAN and point the server at the data set without having to move data physically between the boxes.”
The flexibility of SANs has forced vendors to make a number of major architectural changes compared with conventional networks. First of all, putting all a company’s data in one place puts a tremendous strain on the network; conventional 10Mbps and 100Mbps Ethernet networks would simply fall over if they were trying to service dozens or hundreds of data requests at once.
To remedy this problem, SANs are typically based on Fibre Channel, a new data management protocol that controls the flow of data between devices on a SAN. Those devices are linked using fibre-optic cable, which offers ample room to support data transfers into the hundreds of megabits per second. Fibre Channel typically encapsulates the commands systems use to control disk arrays—including SCSI connecting RAID storage arrays at up to 320MB/sec; 17MB/sec ESCON (Enterprise Systems CONnection) for mainframe-attached storage, and IBM’s 80MB/sec SSA (Serial Storage Architecture) for midrange systems. While this fibre-optic cable was initially run in a ring between devices connected in serial—an architecture known as Fibre Channel Arbitrated Loop (FC-AL)—this approach has rapidly given way to a fully switched fabric, where each SAN device is connected directly into a Fibre Channel switch, or multiple switches for redundancy.
These switches—most commonly from vendors such as Brocade, McData, and Hewlett-Packard (via its acquisition of Compaq’s StorageWorks SAN products)—provide the gateway between a SAN and the rest of the corporate network.
The switches recognise the difference between outbound traffic and that which is meant for devices on the SAN; internal traffic (for example, data running from a SAN to a backup device or to a server) is switched directly to the destination device, whilst outbound traffic (for example, a client-side application’s request for data from the SAN) is directed off of the SAN onto the network. That way, the only data passing over the corporate network is that specifically requested by users.
The use of fibre in SANs provides another important characteristic: distance. By leasing fibre-based, high-speed wide area network (WAN) and metropolitan area network (MAN) connections, a company can build a SAN combining equipment in two geographically distant sites. Because they’re using a direct fibre connection, the high-speed interconnect frees the devices from the latency normally experienced over WAN connections. It also makes disaster recovery far easier than before: simply set up similar SANs in two locations, link them using fibre, and continually synchronise the data between the two SANs. Should one facility go offline, the other one can pick up the job instantly.
Better disk utilisation
In the past, storage administrators typically planned storage purchases by talking with business managers to find out how much data they expected to use, then quadrupling that number. This guesstimation meant storage allocation was an extremely imprecise science—and an expensive one, when disk space that had been purchased for a particular volume was left unused because actual demand didn't meet projections.
Because they present themselves as a single, massive storage array, SANs take a markedly different approach to file management. Whereas NAS devices are individual volumes formatted to conform with a single operating system's structure—for example, NTFS in Windows NT environments—SANs are built for operating system independence. That allows data on a SAN to be accessed equally by mainframe, Unix, Windows, and other servers.
To make this possible, SANs have to act like chameleons—presenting each attached system with a file system that it knows and loves. This happens through use of a completely different low-level structure in which storage space is organised around operating system-agnostic blocks—chunks of data logically partitioned from the rest of the SAN’s storage arrays.
Using this approach, SAN blocks can be aggregated into any number of logical partitions, each with its own identity. This process, known as virtualisation, allows multiple operating systems to each access the part of the SAN they want to see, mindless of the fact that there is completely different data residing on the same SAN storage array.
Because data on a SAN is organised into clusters of blocks, and not logically inflexible volumes, virtualised storage can be dynamically grown or shrunk as application needs demand. Such dynamic sizing—accomplished through use of SAN management software such as Veritas SANpoint Control or Legato NetWorker—therefore allows companies to increase their utilisation of the storage they have.
“Customers have gone ahead and installed SANs, and they’re not trying to make more efficient use of them,” says Simon Elisha, senior systems engineer with Veritas. “The vision of centralised management is that at any given time I can look at my SAN, run any reports instantaneously, and know that ongoing provision of storage is automated. It is getting easier.”
It’s also getting more flexible. Using HSM (Hierarchical Storage Management), storage administrators can also set up an environment where only frequently-used data is kept on expensive storage arrays; lesser-used data can be automatically shuttled onto near-line storage—tape jukeboxes or inexpensive arrays of IDE or slow SCSI drives that keep data to hand but reduce consumption of premium disk space.
Those efficiencies can quickly add up. Suncorp Metway, the country’s sixth-largest bank, found out just how much after it recently upgraded from its diverse DAS-based storage environment to a SAN incorporating nine Brocade SilkWorm 2800 switches, an Hitachi Data Systems 9963 storage array, and StorageTek 9840 tape libraries.
Compaq Windows NT servers and Hewlett-Packard HP-UX servers running key applications access the SAN with equal ease.
Implementing the SAN allowed Suncorp Metway to reduce its number of Windows NT servers from 150 to 30, and to reduce the number of tape backup devices from 70 to just 14 in two SAN-attached tape libraries. Since backups are run over the SAN and don't have to fight general network traffic, Suncorp Metway has observed a five-fold increase in backup throughput.
And most important to the bottom line, storage utilisation has increased from around 60 percent to nearly 90 percent; that’s translated into more than $2 million in savings thanks to a reduction in the amount of unused disk space.
Such benefits commonly flow as customers come to make better use of virtualisation, says Wayne Glynne, ANZ storage business unit executive with IBM Australia. “Virtualisation takes storage to a higher layer by allowing it to be managed as a single set of resources,” he says. “Customers can implement things like a single disaster recovery policy, whereas today they may have separate implementations by platform. The single point of management is a key, since customers can basically move data around the environment to minimise planned outages.”
