By Lisa Coleman
Analysts predict that over the next few years, more and more users will integrate their network-attached storage (NAS) and storage area network (SAN) architectures. In a recent InfoStor reader poll, the majority of users who have both NAS and SAN said they have a need for products that merge the two environments (see chart). But analysts say that need is not translating into demand. In many cases, users are not asking for NAS-SAN convergence products because they do not know they exist.
Vendors are responding to the challenge of uniting storage technologies primarily by connecting NAS heads to SAN storage arrays or by using "hybrid" devices that combine the two architectures in one box.
While vendors are hawking NAS-SAN convergence products, the IT community is not that specific when it asks for new functionality, according to Arun Taneja, an analyst with the Enterprise Storage Group. "Users may want to manage a certain amount of terabytes per administrator, have higher availability, or better manageability, but they're not necessarily asking for convergence of NAS and SAN," he says.
In addition, the term "NAS-SAN convergence" has many meanings. "The term has been hijacked by marketing people," says Randy Kerns, senior partner at The Evaluator Group. "Depending on which vendor you talk to, it means different things."
Kerns believes that NAS-SAN convergence holds little meaning for users because they still have to make a decision about how to handle their storage I/O. Users will either go with file-access protocols via NAS or block-access protocols with SAN or direct-attached storage (DAS). "SAN-NAS convergence doesn't really change that decision," explains Kerns.
Bill Pinkerton, director of the storage solutions business unit at Pioneer-Standard, a distributor/reseller, agrees that most customers have not thought of converging NAS and SAN because they are buying two different solutions. "The hardware can converge, but the functions can never converge because basically SAN is block-level I/O and NAS is file-level I/O," says Pinkerton.
Several types of technologies provide a kind of NAS-SAN convergence, according to Kerns. These include metadata servers such as IBM/Tivoli's SANergy or EMC's HighRoad software; SAN-based file systems; symmetrical pooling solutions that allow IP access with virtualized back-end storage; NAS gateways such as EMC's Celerra (which attaches to EMC's Symmetrix arrays); and special-purpose NAS-SAN hybrid boxes.
However, this article will focus primarily on two of these technologies: NAS-SAN gateways, which allow a NAS head to hook into a back-end SAN array, and hybrid devices that attach to existing storage subsystems for data transport over IP.
Jamie Gruener, an analyst with The Yankee Group, says that NAS heads attached to back-end SAN arrays will be the dominant architecture for converging NAS into SAN over the next few years. He also sees hybrid devices as potential contenders.
"I think a lot of customers will embrace NAS heads if they have excess capacity on their back-end [SAN arrays] that they want to leverage. But for flexibility, having a box that does both should be the endgame," says Gruener.
Electric wholesale market operator PJM began using IBM's 300G NAS gateway last year to help simplify and centralize storage from 18 servers onto an IBM Shark array in a SAN environment.
According to Rich Brenton, senior technical architect at PJM, the company was looking for a NAS device that could handle its new heavily file-based, distributed computing application. "[IBM's gateway] was a way to simplify storage rather than having DAS on all the servers," Brenton explains.
The 300G NAS gateway sits between the disk array and the IP network, routing file-level I/O over the IP LAN and block-level I/O through the SAN fabric. Brenton believes other users can benefit from a combination NAS-SAN environment. "NAS makes it very easy to share data between Windows and Unix environments. Or, if you have large amounts of data and a SAN, and you want to leverage the storage and the management of the SAN, NAS is a good extension."
Another company using NAS heads is Nth Generation, which has multiple HP StorageWorks E7000 NAS devices in its lab facility and production environment. Hooked up to the company's back-end SAN array, the E7000s allow NAS and SAN integration. Nth Generation needed NAS to keep up with enormous file generation that numbered in the tens of thousands.
The E7000 NAS server plugs into a Fibre Channel switch for storage connectivity. "Once you allocate the storage and present it to the E7000, it can then allocate it to all the NAS users," says Rich Baldwin, president and CEO of Nth Generation.
Because the E7000 is based on the Windows kernel, it offers a number of other advantages in the areas of backup and management, according to Baldwin. For example, the company can leverage its administrators' knowledge of Windows, as well as Windows applications and backup software.
Baldwin says that NAS heads are an effective way to use extra SAN disk capacity and to save money. He also expects the technology to gain market traction because there is virtually no connection cost compared to a SAN that requires Fibre Channel host bus adapters (HBAs) and expensive switch ports.
"It could cost as much as $10,000 per machine to connect to a SAN if you have redundancy with multiple HBAs and Fibre Channel switches," he explains. "[With a NAS head,] the network connection is almost free because you're going over your existing Ethernet."
A potential downside to NAS gateways is that they usually work only with the same vendor's SAN devices. One exception may be Auspex's NSc3000 controller gateway, which the company claims supports any SAN equipment. So far, Auspex has tested the controllers and found that they are compatible with equipment from EMC, Hewlett-Packard, Hitachi Data Systems, LSI, and StorageTek equipment.
StorageTek has been using the Auspex gateway internally for about six months. The controller sits between a StorageTek 9176 disk array and the IP network. The gateway routes file-level data over the IP network, but does not control the movement of block-level data, which is the SAN's function. Since the controller routes file-level data over the IP network using SAN storage, it eliminates the need for a separate NAS server with dedicated disks, according to Todd Stafford, a systems administrator at StorageTek.
Bill Ubelacker, director of IT at Burlington Coat Factory, says his company has merged NAS and SAN, but he still sees NAS and SAN as two separate and distinct technologies. He says that even though they may work together, the technologies are still not really converged.
"Depending on who you talk to, you get a different definition of what SAN is and what NAS is," says Ubelacker. "In my opinion, a SAN is something that's dedicated to storage only, not communications."
Burlington Coat Factory is using Hewlett-Packard's SANlink Plus, a SAN appliance that sits in the data path and provides connectivity to heterogeneous storage. In Burlington's case, it is connected to Hitachi Data Systems disk arrays, which are connected to a SAN through Brocade switches. The NAS heads are Linux PCs that are attached to the SAN. Burlington's total storage network capacity is approximately 20TB, with about 2TB used as NAS.
Jeff Pelot, chief technology officer at the Denver Health Hospital, began looking into NAS technology about two years ago to help store about 14TB of patient records, such as radiology images and pharmacy information. Because NAS is file-based, it did not meet Denver Health's needs because the hospital has huge database requirements. Pelot believed the hospital needed a SAN.
Starting with EMC Clariion arrays, Pelot later added four Network Storage Module (NSM) devices from LeftHand Networks. An EMC SAN handles applications that must operate 24x7 with 99.999% availability. LeftHand's NSM 100s handle another set of applications that are more business-oriented and do not require 24x7 availability.
The Clariion array and NSM modules work in conjunction with each other. For example, Pelot can copy data from the NSMs and mirror it to the SAN. Medical records are stored in the SAN, and business data resides on the NSMs.
When the NSMs are plugged into the network, they are aware of each other, explains Pelot, and administrators can build LUNs across the units and perform RAID striping across the boxes to mitigate single points-of-failure. Even though Denver Health uses the NSMs in file mode, the modules can also be used in an IP SAN. In this configuration, the modules operate in block mode and appear as local disks.
BlueArc continues NAS-SAN convergence
BlueArc recently extended its SiliconServer line with three new servers—Si8300, Si8700, and Si8900—that combine a NAS front-end and Fibre Channel SAN back-end in a single box, extending its reach into the enterprise NAS space. The company's goal is to deliver SAN performance and scalability to the NAS market.
The servers, which scale from 7TB to 250TB (ranging from $50,000 to $125,000), can simultaneously read and write data at multi-gigabit rates.
BlueArc recently added virtual volumes, snapshot capability, Ethernet tape library connectivity, and support for NDMP copy to its NAS systems.