Higher penetration and dependence on technology by businesses in the past decade has ensured that processing power and storage requirements have grown exponentially. The offshoot has been the increasing importance of linking servers and the storage devices in order to tap into the data and convert into something of a competitive advantage or a facilitation tool. Storage-area networks (SANs) is the answer to this corporate need and these have proven to boost performance, reliability, and scalability of the critical link between servers and storage devices.

SANs in short are gigabit-rate networks that rely on Fibre Channel for higher throughput, greater distances, and more connectivity options between servers and storage devices. They can be built as switched- or shared-access networks, and either way they offer better scalability, fault recovery, and diagnostic information than current approaches. And the initial outlay for SAN infrastructure could pay for itself in terms of reduced management costs over time. A SAN makes sense whenever data integrity and availability are stringent requirements. Small workgroups may benefit if there’s lots of mission-critical data involved.

SANs generally comprise servers (hosts), storage devices (tapes and disk arrays), and bridges and multiplexers, all connected to Fibre Channel switches. As with LANs or WANs, the switches furnish the backbone for all connected devices, with one or more switches acting as a Fibre Channel switching fabric. SAN switch fabrics allow attachment of thousands of nodes. Besides Fibre Channel SANs help take the load off servers, which have historically been given the added burden of transferring data to storage devices and to the LAN. Now, servers can offload data transfer to the SAN—and return to their original processing role.

As for setting up the SAN, the job is relatively easy. Fibre Channel SANs can be designed as both shared-media and switched-access networks. In shared-media SANs, all devices share the same gigabit loop. Trouble is, as more devices are added, throughput goes down which can be addressed by having a backbone, based on Fibre Channel switches thus increasing a SAN’s aggregate throughput.

Another key issue for devices attached to the SAN is the ability to discover all the devices in the switching fabric. Fibre Channel defines a discovery mechanism—SNS (simple name service)—that learns the address, type, and symbolic name of each device in the switching fabric. SNS information resides in Fibre Channel switches, and NICs and storage controllers request SNS data from the switches. Thus, when deciding on a vendor IT decision-makers should look for Fibre Channel NICs and storage controllers that back SNS which also help in error recovery and fault isolation through an optional feature called RSCN (registered state change notification). Another plus of Fibre Channel is that it can integrate diverse protocols and helps IS managers build SANs that accommodate even the largest data center’s storage needs.

As for management, the administrators should be able to use all the tools and systems they use for LANs and WANs. That means looking for SAN devices that can be managed via SNMP or through the Web, via HTTP (hypertext transfer protocol). Devices also should support telnet (for remote diagnostics or servicing). All of these should furnish detailed information on device status, performance levels, configuration and topology changes, and historical data. Key status and performance information would include throughput and latency metrics. Future requirements will include a commitment to delivering tuning and optimization tools.

To see how SAN concepts can be put to work, consider a medium-sized company looking to upgrade its storage capabilities. Its infrastructure consists of four buildings, two of which contain two Windows NT servers each, and the other two of which house four Unix server farms each. All the servers use redundant RAID arrays connected via parallel SCSI links.

One of the first things the company should address is the sharing of I/O and file systems. Windows NT is a so-called shared-nothing environment relative to I/O and file systems, in that multiple
NT servers can’t share these resources. Therefore, IS managers will need to implement a third-party distributed lock manager to share files and storage. Many versions of Unix have distributed lock capabilities for I/O alnd file systems. With these capabilities in place, it is then possible to begin building the SAN.

As for the switched/shared decision, management requirements may tip the balance toward a switched SAN in this case. With multiple sites and servers, troubleshooting is difficult. The relatively small cost difference between a shared design with arbitrated loops and a switched fabric is more than compensated for by the management information available from the switched setup. A switched fabric is also a better choice if high availability is a requirement. Using distributed lock management or a zoning mechanism, IS managers can design the SAN to isolate workgroups (like those in marketing or engineering) or operating system environments (like NT and Unix), all the while using the same switch. For backing up the storage, this network design can include tape devices attached directly to the switch fabric. IS managers can then set up temporary time-of-day zones to do backup for both Unix and NT environments.

For disaster tolerance, this design would benefit by placing the backup system 10 km away from the switched SAN. Additionally, the redundant RAID devices can be separated from the server farms and instead directly attached to a Fibre Channel switch. This allows RAID arrays to be located in different buildings, further enhancing redundancy.

At a glance for the IT decision-maker

1. SANs allow servers to do what they are meant for-processing and boost performance, reliability, and scalability of the critical link between servers and storage devices

2. SAN is relevant for companies in businesses where data integrity and availability are stringent requirements or organizations which have small and spread workgroups involving sharing of mission-critical data

(The article was written by the businessgyan editorial team with inputs from various industry sources and experts)

Issue BG8 Nov01

Related Items:

---

Be first to comment this article

Only registered users can write comments.
Please login or register.