Up the SAN-Box - storage area networks take off - Statistical Data Included

Emedia Professional, July, 1999 by David Doering

fibre channel and new directions in network storage

1998 is the year network storage turned sexy.

Now that's meant as a PC concept, not as a concept that's PC. Prior to 1998, talking about storage held all the fascination of a speech at a plumbers' convention. While the pipe-and-solder people might discuss copper versus plastic or 1.5-gallon versus 3-gallon toilets, on the computer side, it was copper versus fiber or 100Mbps versus 1Gbps. But the concept's the same, and until recently, the unvarnished truth about network storage has hardly been the stuff of gripping drama.

So what happened in 1998? The storage area network (SAN) concept took off.

That year, more vendors at the enterprise level announced SAN storage solutions than ever before. SAN has become all the rage. Today, almost every major player in the enterprise storage arena has announced or delivered SAN product.

Never mind that others pioneered these tools years before (just ask data-protection vendor Vinca, which launched the whole SAN movement), today it's EMC, Network Appliance, and Comdisco. In fact, SANs provide a superior level of storage access for networks. Compared with host-attached devices, SAN-attached storage systems are faster, more reliable, and easier to manage. So is an optical-drive-equipped SAN in your future?

Maybe.

A DATE WITH DENSITY

To understand why SAN's promise of high-density storage may or may not be in your future, we'll need to take a quick look at how SANs work, where storage bottlenecks are, and what solutions there are to fix these bottlenecks.

Compare, for instance, traditional host-attached storage with network-attached. Both provide convenient access with a simple setup. If there's a fault with the server, the storage becomes inaccessible. There's also significant overhead when using the standard Ethernet protocol. Whether it's 10BaseT, 100BaseT, or gigabit Ethernet, there's always the Ethernet overhead to contend with--which averages a minimum of 10 percent and could rise much higher depending on the types of devices connected and the amount of traffic each generates.

So how do we make a storage area network out of a variety of storage devices? Each device connects directly to the network cable, as would occur with the familiar network-attached storage (NAS). However, this network cable is accessed only by a set of network servers (these can be database, application, or Web servers). The loss of one or more servers does not prevent the remaining servers from accessing the SAN-attached drives. In fact, the Fibre Channel loop itself can be broken and access can continue utilizing the self-healing capabilities of the protocol.

In a SAN, the administrator can add or remove storage devices as required without downing a single network server. This provides a robust and reliable storage service. It also means that the storage service can more closely match actual need, rather than building in a lot of extra capacity to meet expected future requirements. With traditional servers, adding drives is so complex it is best to do it only when necessary. So over-capacity is a method for compensating for this complexity.

The SAN architecture does not demand a specific protocol. In fact, in the past, SANs were constructed using SCSI and could still be today using its multiple initiator capabilities. However, it is universally understood that SANs as we have come to define them use the Fibre Channel protocol and wiring.

One way to think of Fibre Channel is as a super-SCSI, built upon all the amassed development understanding gained over the last 15 years in storage since SCSI first appeared. Fibre Channel has more throughput than UltraSCSI (100MB/sec versus 80MB/sec) and supports more devices. While significant discussions persist over the more universal presence of SCSI devices versus the less-common Fibre Channel, in one key respect, they are completely different--distance. The Fibre Channel specification supports connections over single-mode fiber of up to 10KM, while SCSI remains limited to distances closer to 10 feet.

Like SCSI, however, Fibre Channel needs devices constructed with this type of connection in mind. Currently, a variety of RAID and tape subsystems are offered with built-in Fibre Channel support.

Fibre Channel also features low-overhead compared with a traditional Ethernet connection. For that reason, a Fibre Channel SAN is superior to NAS for accessing large amounts of data. So why isn't Fibre Channel the preferred solution? As a destination, 120mm optical drives on a SAN have various limitations from capacity to speed to limitations inherent in the Fibre Channel SAN itself.

THE DENSITY BIT

One challenge to incorporating DVD/ CD-R/RW/RAM media on a SAN, such as for backup and archiving purposes, is the relatively small density of the individual discs involved. From DLT to AIT, tape delivers four to five times the capacity of a single DVD-RAM disc, with 20GB a common capacity for high-end tape devices. How can writable DVD's 2.6GB to 4.7GB--let alone CD-R's 650MB--hold up?