Defining the Technologies

To understand how and when to use technologies like NAS or SAN it is important to understand what they are and what they offer. The technologies differ in how they are used and what advantages they provide. Many administrators assume that they need SAN when often a NAS will suffice. Because IT budgets are far from limitless, it is to your advantage to know that you aren't overbuying for your solution.

What is a SAN?

A Storage Area Network is a high-speed special-purpose network or subnetwork that connects various data storage devices with associated data servers on behalf of a larger network of users. Typically, a SAN is but part of an overall network of computing resources for an enterprise. A SAN is usually located in relative proximity to other computing resources, such as databases and file servers, but might also extend to remote locations for backup and archival storage. These remote locations are traditionally connected via wide area network carrier technologies such as asynchronous transfer mode (ATM) or Synchronous Optical Networks (SONET).

It is very important to understand that the SAN is more than just the chassis that contains the disks. It includes the RAID controllers for the disks, the Fiber Channel switching fabric, and the Host Bus Adapters that reside in the data servers. SANs are traditionally connected to hosts via Fiber Channel. Although it can be fairly easy to support dual arbitrated fiber loops in a corporate environment, keep in mind that one of the primary benefits of SAN is the ability to do block-level mirroring to another SAN. If this SAN is located remotely, up to 1,000km away with current Fiber technology, a company needs to have fiber between the two locations. A fiber connection across those kinds of distances can be quite expensive.

SAN technologies excel in the area of disk performance. Fiber channel networks regularly push 2GB/sec of throughput. Although SCSI technologies can move data at up to 320MB/sec, they are limited to less than 25 feet of distance. SAN, not unlike SCSI, is seen by the host system as RAW disk. This is also referred to as a block-level technology. In the past, database applications required block-level access to the disk as well as the "near 0 latency" offered by SAN.

Zero Latency

Although most SAN manufacturers refer to the performance of their products as having zero latency it is important not to misinterpret this. Zero latency refers to the fact that Fiber Channel has extremely low overhead and doesn't add additional latency. The laws of physics, on the other hand, are still in effect. A 1,000km fiber run will still take 7 milliseconds roundtrip.

What is NAS?

Network attached storage (NAS) is a hard disk storage technology that uses its own network address rather than being attached directly to the host computer that is serving applications or data to a network's users. By removing storage access and its management from the host server, both application programming and files can be served faster because they are not competing for the same processor time. The network-attached storage device is attached to a local area network via Ethernet and given an IP address. File requests are mapped by the host server to the NAS device.

Network-attached storage consists of hard disk storage, including multidisk RAID systems, and software for configuring and mapping file locations to the network-attached device. NAS software can usually handle a number of network protocols, including Microsoft's Internetwork Packet Exchange, Common Internet File System and NetBEUI, Novell Netware Internetwork Packet Exchange, and Sun Microsystems Network File System. Configuration, including the setting of user access priorities, is usually possible using a Web browser though many NAS offerings require command line configuration. Most NAS manufacturers include specialized software for allowing specific applications such as SQL or Exchange to take advantage of special functions provided by the NAS. These functions include things like mirroring, failover, automated recovery, and snapshotting.

A Single MAC Address Conversation Cannot Span Multiple Network Interfaces

When considering network architecture to support a NAS device, remember that when bonding multiple Ethernet interfaces, a single MAC address conversation cannot span multiple network interfaces. If the NAS is on Gigabit Ethernet and a server has a quad fast Ethernet card, it will only talk to the NAS device at 100MB/sec even though it has 400MB/sec of potential throughput.

NAS has the advantage of using existing Ethernet technologies that are much less expensive than Fiber technologies. With the availability of 10GB Ethernet, NAS is able to compete with Fiber Channel “based technologies even with the added overhead of Ethernet over Fiber Channel.

What is DAS?

Direct Attached Storage (DAS) is the traditional "local disk" that most administrators are accustomed to. Technologies like IDE and SCSI are used to connect drives or RAIDs to the host computer. Direct Attached Storage has many advantages in the area of performance and cost. With SCSI technologies like Ultra 320, which can transfer 320MB/sec of data across the bus and controllers with large amounts of cache memory, DAS is well situated to provide fast and stable disk performance. The disadvantage of DAS is that the management load of each of these DAS subsystems increased linearly with the number of servers. This can eventually become unwieldy as there is a lack of centralized management.

With the advent of Serial ATA and Serial ATA RAID controllers, IDE devices are becoming more and more viable for systems that need fast access to large amounts of data that won't see huge numbers of transactions. Systemlike backup servers that will cache to disk before spooling to tape can benefit from the speed and low cost of Serial ATA devices. IDE drives almost always lead SCSI drives in the area of size and are always less expensive. IDE RAID controllers offer excellent performance at much lower costs than their SCSI equivalents. IDE-based RAID controllers don't handle large numbers of requests as well as SCSI-based RAID controllers. IDE-based RAIDs are still limited to four devices. This isn't always an issue as SATA drives of 250GB or more are commonly available. A 1 Terabyte SATA RAID can be built for well under $1,500. With throughput of over 180MB/sec, Serial ATA RAID can be a very tempting DAS technology.

IDE Technologies Can Compare Favorably to SCSI Technologies

Although IDE technologies can compare favorably to SCSI technologies in the area of seek time and bus bandwidth, they can't compete in the area of IO transactions per second. Most applications that require large amounts of disk are IO bound, not bandwidth bound. Be sure to understand the IO requirements of an application before choosing disk technologies.