Special Report: Enterprise Storage<br>Managing Enterprise SANs: Reality Checks and Best Practices

In a conventional storage model, the task of management often breaks down along platform lines: A storage subsystem is physically connected to a Windows NT or Solaris server and mapped on a 1:1 basis. The storage area network (SAN) paradigm throws this model out of whack, allowing many servers to transparently access any number of storage devices at the same time. Because of its complexity, however, the SAN approach can cause a number of management headaches.

The task of managing SANs is especially acute in Windows NT environments, says Brian Reed, senior director of business development at SAN switch and hub manufacturer Vixel Corp. (www.vixel.com). He says part of the problem is that at the time Microsoft was developing Windows NT 4.0, it assumed a 1:1 relationship between the networking operating system and storage devices.

"SANs have broken that paradigm, and the whole SAN topology says you can have many servers to many storage devices," Reed explains. "So operating systems over time have been updated and made more robust, and NT 4.0 is in need of an overhaul."

Microsoft Corp.’s much-ballyhooed overhaul to Windows NT 4.0 -- Windows 2000 -- includes native storage management and enhanced clustering capabilities that will do much to increase the Windows platform’s viability in SAN environments.

But Said Rahmani, senior vice president at SAN specialist Pathlight Technology (www.pathlight.com), says IT managers must be made aware of the fact that almost all major operating systems must be overhauled to address the task of management in a world of networked storage.

"Frankly, there is a lot more that operating systems vendors in general could do to be more heterogeneous and provide better resource sharing across SANs," Rahmani contends. "As it took some time for traditional networks to become part of mainstream operating systems, SANs must go through a similar revolution, as well."

Some of the biggest problems underscored by vendors and analysts are the lack of comprehensive manageability tools (see sidebar on page ??) and the difficulties posed by a lack of overall interoperability among devices.

In the absence of completely interoperable SAN devices, most vendors caution customers to deploy only solutions that have been tested for interoperability with one another.

Companies such as MTI Technology Corp. (www.mti.com), Exabyte Corp. (www.exabyte.com), and EMC Corp. (www.emc.com), have compiled compatibility lists of SAN components from other manufacturers that they’ve tested and certified for interoperability with their product offerings.

"The customers that we talk with absolutely want to move to SAN architectures, but they recognize that it’s not a plug-and-play technology and you can’t go buy an off-the-shelf, build-it-yourself kit yet," says Kevin Liebl, product manager at MTI.

More vendors are likely to follow the lead of MTI, Exabyte, and EMC, predicts Carolyn Dicenzo, principal analyst with GartnerGroup's Dataquest (www.dataquest.com) unit.

"Open standards are a nice mantra, but it takes so long to get to those. Getting everyone to agree on what is really needed takes a lot of time, so what’s happening is that these types of proprietary, packaged solutions are emerging first -- and that’s normal," Dicenzo says.

According to Tim Werts, a feature engineer for storage and adapter management at NCR Corp. (www.ncr.com), packaged solutions are really the only safe bet at the present time, given the potential for chaos that exists among devices that simply can’t be guaranteed to interoperate reliably with one another.

"We say that we want everything to be open and not proprietary, and that’s great. But the reality of it is that if you’re looking to deploy a SAN, you’d better start small, and then you’d better get a solution that has been fully integrated and tested and that you know is going to work, because the interoperability is just not there yet," Werts says.

Another problem, notes Dale Quisenberry, vice president at SAN multiprotocol router specialist Crossroads Systems Inc. (www.crossroads.com), is that many potential customers mistakenly believe they have to upgrade to Fibre Channel-only storage infrastructures if they want to implement a SAN.

"There is absolutely a huge installed legacy base of SCSI devices, and we just don’t see that changing very quickly. SCSI-based solutions still compose a majority of the storage devices being sold today," Quisenberry says. "We absolutely believe that Fibre Channel is the storage medium and SAN the storage network of the future, but customers still buying SCSI today will need a connectivity solution to bring those SCSI devices onto a SAN."

Crossroads Systems provides a product line of multiprotocol SCSI-to-Fibre Channel routers that enable customers to protect their existing investments in SCSI, Quisenberry says.

The fact that most operating systems are designed around conventional storage models and 1:1 mappings of storage devices to server hosts, presents another problem. Most SAN environments employ a technique called logical unit number (LUN) masquing to circumvent this difficulty and provide transparent many-to-one or many-to-many access to storage subsystems.

"LUN masquing is very important in a cross platform, heterogeneous SAN," says MTI’s Liebl. "You need to be able to partition out the individual storage to make sure that the servers don’t step on one another, but because of the lack of interoperability among devices, this can be very difficult."

In the final analysis, most IT organizations will probably want to deploy SAN implementations based on the maturing Fibre Channel standard.

Elisa Wade, program manager for software engineering at Exabyte, says although it is theoretically possible to deploy a SAN over Ethernet and Fast Ethernet networks, IT organizations should look to the performance and the proven standard of Fibre Channel.

"I know people who’ve deployed a SAN subnet using NetWare 3.12, and I can develop a SAN based on 10-BaseT or 100-BaseT networks," Wade says. "[IT organizations] can simply create a storage subnet now, but they’d be better off taking advantage of the speed and incredible capacity of Fibre Channel."