Network file servers deliver speedy data access

The network file server, a specialized system for users who need fast access to large stores of data, may still be a niche market, but the product itself is evolving rapidly.

In its basic form, the network file server involves basic server hardware, software and storage systems that have been tailored to send data very quickly across the network, supplementing an organization's existing network server.

However, the market has diversified over the last several years, expanding from high-cost, customized-hardware-intensive systems to more inexpensive options based on standard components. Also, although the product has its roots in the Unix world, vendors are developing solutions that serve both Unix and Microsoft Corp.'s Windows NT.

Mike O'Donnell, the federal region manager for Network Appliance Inc., Santa Clara, Calif., compares the evolution of specialized file servers to the development of the network router.

"Originally, customers used a [Digital Equipment Corp.] VAX or [IBM Corp.] System/38 running routing software and with a network interface card installed," O'Donnell said. "Cisco and Synoptics recognized that that was not productive and came out with a box that was designed to be a router. Performance went through the roof, system administration went way down, and reliability went way up.

"Network Appliance sees the same problems occurring with file servers," he continued. "People are handling two and three times more data each year, but how we manage that data hasn't changed."

Streamlined Operations

Network file server vendors have addressed this problem by streamlining file server configurations and operations, stripping away everything that is not needed to store and distribute data.

Typically, a general file server would involve a reduced instruction-set computing (RISC) chip-based server for Unix systems or an Intel Corp. PC-based server for Windows NT networks, with a disk drive storage array attached to the server.

"You can have a Sun [Microsystems Inc.] or Hewlett-Packard [Co.] workstation with a nice graphics card, but it is not optimized to be a file server," said Phil Kao, product marketing manager for Articon Inc. "It is not the best use of material." Articon's customers include NASA's Cape Canaveral, where they archive launch trajectory data, and the Army, where they store terrain data collected by roving Humvees.

Much of the computing power of general servers is wasted in file server operations, which makes it a very poor investment, vendors said. Furthermore, workstation- and PC-based servers lack the remote administration and high-availability features customers desire in a server.

"Being a file server has everything to do with the [input/output] data path, not computing power," O'Donnell said.

The general-purpose flexibility of such servers extends all the way down to the operating system, said Greg Garry, server analyst for Dataquest, a market research unit of Gartner Group Inc. A modern multitasking operating system can have 6 million lines of code, and it provides many functions that are not needed for a file server. A stripped-down file-serving-specific program is a fraction of the size and runs much faster. "That cuts down on the overhead and the price as well," Garry said.

"[A full-service OS] works, but you've got an operating environment that is cumbersome to maintain because the tools were [used] to manage file servers," said O'Donnell. "All of our tools were written around file serving. It eliminates 80 percent of the work."

The network file server was originally a device designed to handle Sun Microsystems' Network File System files exclusively on Unix-based networks. Now, these servers are still widely referred to as NFS servers, although they are quickly adding support for Microsoft's Windows NT file system as well.

Keeping files on a server in a format that is accessible by different users on different types of computers lets agency users share data and integrate various types of computers on a network.

"Everything you can think of is on our system," said Lyn Kendrick, a system administrator at Los Alamos National Laboratory. "Sun, SGI, HP, DEC, PCs, Macs -- you name it. For a PC or a Mac to access a Unix file system is pretty good," she said.

Users in the lab depend on the ability to use shared data, she said, so the lab's Auspex Systems Inc.'s file server provides that data sharing at high speeds. Still, this data must be translated for those users with non-Unix computers, so the industry is moving rapidly toward NFS file servers that treat NFS and Windows NT file systems identically, providing equally fast access to data regardless of the source of the data or the operating system accessing it.

Unix is still far more common, but federal customers know that they need to be ready to use either operating system, and vendors are scrambling to accommodate that need. Network Appliance and Articon are shipping products that support both file systems, and Auspex said it is preparing those products to be ready early this year.

"Even the systems that were previously only NFS servers are adding Windows support," said Robert Gray, research manager for storage systems at International Data Corp., Framingham, Mass. "People are looking for that as a checklist item. More and more of the market is going to be [Windows] NT networks and heterogeneous networks."

The flexibility appeals to Gregory Powell, manager for the Freedom of Information Act Information Processing System at the Immigration and Naturalization Service. "You could get rid of the [Unix] server, come back with a Windows NT server, and there is your data, ready to use," Powell said.

A Growing Market

The network file server market started off as a fairly small niche but is expanding rapidly, both in the number of buyers and sellers, according to industry analysts. This emergence has attracted vendors from other, more general market segments, such as Sun Microsystems from the general-purpose server market and EMC Corp. from the storage market.

"Auspex and Network Appliance have shown that this is a very viable market, and it is growing very fast," said Gil Press, manager of network storage marketing for EMC, Hopkinton, Mass.

The company found it very attractive to move from its established base in channel-attached storage, which connects to existing servers, to network-attached storage that has its own specialized server, Press said. "In this case, we have provided a specialized server that focuses on network file serving." EMC's customers include members of the intelligence community who deal in patterns, images and huge collections of records, he said.

With the growing number of vendors, customers can choose from an increasingly wide range of price and performance. At the high end, Auspex, one of the original vendors in the market, offers customized hardware and software solutions for users with very demanding requirements.

"Our boxes are designed to go in environments where customers understand the value of dedicated systems," said Kuljeet Kalkat, senior director of marketing for Auspex, Santa Clara, Calif. Sandia National Laboratories and Los Alamos National Laboratory apparently fit that description because they use the company's products to store their vast amounts of data. "Department-level products were not our focus," he said.

However, Auspex is developing less-expensive products that will address departmental customers soon, Kalkat said, so the company will not leave that market open to its more affordable competitors.

EMC and Network Appliance target the affordable end of the market by using standard hardware components as their building blocks.

INS' Powell said that the Network Appliance servers he bought were priced about the same as the IBM disk arrays he was considering for the six-processor IBM RISC System/6000 server at INS.

"I was looking for more disks for the RS/6000," he said. "I looked into Network Appliance and didn't really believe their claims."

In response, the company gave him a Network Appliance on a trial basis. "They said they could have it set up in 30 minutes," Powell said. "I've never done anything on the RS/6000 that didn't take two or three days."

Touting Performance

One benefit that all the file server vendors tout is improved performance. They promise not simply better performance than using general-purpose machines in the server role but better performance than storing the data on local hard drives.

Powell was skeptical of this claim. "There is no way you can tell me it can access files quicker on the network than in the box itself," he said. "That is hard to believe." But he said he tested his system and found it was true.

The reason for faster-than-local file access is the speed of modern Ethernet connections, which blast data at up to 100 megabits/sec. That is faster than the data rate of even the speediest SCSI disk controllers in computers, so the network is not a bottleneck. The file server software and hardware are further optimized to provide top performance, vendors said.

Indeed, some customers may use the network file server as a supplement to local disk drives.

"The Network Appliance is our C drive," said John Marshall, division officer of enterprise services for the Office of Naval Intelligence. "We even load [Microsoft] Office 97 applications on our server." The office has more than 800 such users connected to a Network Appliance as their home directories. "The performance is outstanding," Marshall said.

Other users want huge amounts of storage that is unrealistic to provide locally. "Some people want 32 [gigabyte] file systems," Los Alamos' Kendrick said. "To just slap that into a workstation is a problem." Instead, she lets them create huge file systems on the Auspex file server.

"I think the performance is good," Kendrick said. "Occasionally I have had people complain, but typically I've found that the cause was a network problem that had nothing to do with the Auspex."

Measuring Availability

But more important than performance is availability, vendors say, which is an area that is subject to hard-to-verify claims or using circumstances that can not be fairly compared.

"There are established benchmarks for performance, but not [for] availability," EMC's Press said. "We have found that even more important than performance is the availability of the file server."

The reason is that users are completely dependent on the file server if they are using it as a substitute for local hard disk drives. "Even two hours of downtime can cost an incredible amount of money because you have all these expensive people sitting idle," Press said. "Users complain a lot about the server being down." EMC systems can reboot, if necessary, within 100 seconds, according to Press.

Network Appliance and Auspex both claim annual downtime of a few minutes for their customers. They support Redundant Array of Independent Disk (RAID) 4 configurations, which use multiple disks to back up data.

Because of how network file servers are configured, RAID or disk mirroring does not hurt overall performance, as it does on traditional servers, Powell said. "I was talking to [a vendor] about features like mirroring, and the phrase 'slight performance loss' kept coming up," he said. "I've been doing this long enough to know what 'slight' really means."

Carney is a free-lance writer based in Herndon, Va.