Although companies such as Paramount and Unisys are already reaping the benefits of FC-AL, it will be some time before they'll be able to fully realize the benefit of SANs. Several limitations of Fibre Channel technology are holding back the development of automated SANs.
Instead of relying on the loop architecture, full Fibre Channel implementations use switches to route traffic (see the sidebar "Inrange Releases 128-Port Fibre Channel Fabric Switch")this requires a major software overhaul. Four years into Unisys's deployment, the company is still working on rewriting the software code in order to make switching work. It's expensive and laborious and requires a lot of expertise.
Nevertheless, it is possible to build a small SAN with FC-AL (loop), people use "full" or switched FC for larger-scale networks or for future scalability.
Inrange Releases 128-Port Fibre Channel Fabric Switch
Banking on the accelerated adoption of Fibre Channel SANs in the enterprise marketplace, Inrange Technologies has released the 128-port model of its IN-VSN FC/9000 Fibre Channel Director, a switching infrastructure that enables users to centrally manage and add storage resources in SANs.
A Fibre Channel director provides a central point of connectivity and manageability for SAN infrastructures. According to Inrange, based in Lumberton, New Jersey, the director is the "traffic cop" that servers and storage systems connect to in order to speak to one another. For scale, it is generally accepted that a director must support at least 32 concurrent devices.
Inrange's customers and the market don't want to rely upon dozens or hundreds of discrete little devices, all stitched together with cables. It's too tough to manage, too tough to plan, and very high-risk.
Because directors are one logical switching element that can accommodate 128 devices, they do not require cables between little switches (called interswitch links, or ISLs). Directors provide a way to aggregate and scale SAN infrastructure and improve manageability. Directors fundamentally simplify the deployment and management of infrastructure.
The FC/9000-128 has no single point of failure protecting the network from outage danger due to a failure in the switching element. With little switches stitched together, there is a real risk of downtime if one or more of the switches that are cabled together have a problem.
Directors have what is known as a nonblocking architecture, which means that all devices in the SAN can speak through the network at full speed, without having their traffic blocked due to network traffic jams. Networks of smaller switches normally are prone to oversubscription, a bad thing that can be associated with overbooking in the airline industry. In an oversubscription environment, there are more devices attached to the network than there is internal network bandwidth, which means that someone is going to just have to wait for network resources being freed up.
Inrange introduced the industry's first 64-port Fibre Channel director in early 2000, in response to this emerging market need for users to build SANs that could expand easily and maintain performance levels as ports were added, that were simpler to manage than weaving together a mesh of smaller switches, and that offered bulletproof levels of redundancy to protect businesses with no tolerance for downtime.
The 128-port FC Director is available immediately for factory orders. Users of the 64-port FC/9000 systems can migrate to 128-port models through simple field upgrades.
According to Inrange, more than 98% of the traffic directed by its FC/9000 technology is for UNIX, NT, Linux, and other open systems platforms. The technology has no server or storage platform bias. The design allows SAN switching to scale from 24 to 256 ports, where the performance per port does not degrade.
With its technology partner QLogic, Inrange will provide 2GB and 256 port director technologies later in 2002. In the Fibre Channel space, Inrange competes with the likes of Brocade and McData, but only McData is producing another director-class product. List pricing for the FC/9000-128 range from about $200,000 for a 48-port unit to just over $500,000 for a fully configured 128 port unit.
The key problem that Unisys has encountered is with the drivers. There are big software changes as soon as you start spreading out availability of different storage devices.
A switched network might include as many as 16 million disk drives.
Moreover, while Fibre Channel provides more centralized management, servers don't automatically route data where it belongs. The instructions must still come from the multiple host adapters, and the management software to automate those instructions needs to be improved.
Interoperability, reliable chip sets and compatibility with Windows NT are other thorny issues that need to be addressed before switched Fibre Channel is ready for prime time. One vendor's device doesn't necessarily work with another's, and you can't ignore incompatibility with a platform such as NT. Today's SANs support UNIX.
Not surprisingly, vendors are singing a slightly different tune. Most host bus manufacturers are developing the enabling drivers for Fibre Channelready switches.
NT compatibility will exist soon, and companies such as Compaq and StorageTek are integrating drivers into Fibre Channel network devices such as hubs and switches. Even though the driver issue isn't solved, switched Fibre Channel is so superior to FC-AL that it's worth the implementation hassles. The real issue of switches versus the hub loop is that the loop is not reliable. When the loop fails, the whole thing goes down. With the switch approach, there is fault tolerance.
Nevertheless, hot-pluggable drives allow administrators to replace a failed drive without bringing the whole looped network down. Besides, vendors such as Ancor, Brocade, and Vixel are pricing their Fibre Channel switches at $19,000 and up. When you're spending that kind of money, the programming costs associated with switched Fibre Channel won't look so expensive.
Nothing is easy, and Fibre Channel is no exception. It is, however, a viable mechanism, and despite its glitches, it's the only candidate out there at present.
Companies with extensive testing capabilities have the best chance of success with Fibre Channel switching rollouts. If you're thinking of becoming an early adopter, consider buying all your Fibre Channel products from the same vendor because the technology is hard to debug and troubleshoot. People tend to forget that new technologies just take time to work themselves out. Until then, the expense and labor associated with true SANs will go with the territory of being an early adopter.