Distributed file system on tap

With corporate storage pools springing up all over, one of the most tantalizing new management ideas is the distributed file system. A DFS manages multiple, geographically dispersed storage devices, including Fibre Channel and network-attached storage servers. A variety of start-ups and traditional vendors are developing products, pitching IT managers on the ability to scale storage capacity and performance at will.

For example, start-up Zambeel offers Aztera, a distributed NAS file system that will let users consolidate multiple users, departments or projects within a secure storage infrastructure. And Acirro, another start-up, offers Acumula, software that lets users combine data across storage devices, networks and formats.

Other DFSs for storage contenders include 1Vision, ADIC, DataCore Software, Scale Eight, Z-Force, and traditional vendors such as IBM Tivoli, LSI Logic, Silicon Graphics, Sun and Veritas Software.

It seems such product development comes none too soon. Users have been waiting for DFS for the past five or more years, chiefly on the promise of unlimited scalability and more balanced storage workloads, says Arun Taneja, a senior analyst with the Enterprise Storage Group. "Managing different computers, each with its own file system instance, and its own storage is a lot harder than if there was a solid, interlocking file system that ran on the cluster. This is particularly true for a single application, such as e-mail," he says.

If you run out of capacity in a traditional direct-attached storage environment, you have to bring the server down and add more, Taneja says. If you run out of compute power, you either buy a larger server or add another one and split your application and data. "Either way, you have to bring your systems down," he says. "What a pain from a user [perspective] and from a storage management perspective."

Although DFS players take slightly different approaches in their products, each presents one file system image to the application server or client seeking file services, Taneja says. Instead of being serviced via Network File System from three NAS servers today - and thus having at least three file systems mounted to your NFS client software - you would only have one mount point to the DFS. "This alone would make your IT administrator's life significantly easier," he notes.

For DFS to interplay effectively with the various file systems that come with NAS devices (NFS for Unix boxes and Common Information File System [CIFS] for Microsoft servers), vendors tweak the technology in myriad ways for maximum storage performance, says Mike Kahn, chairman of The Clipper Group, a technology acquisition consultant. "In one instance, a singular file system takes over for many servers. In others, a metafile server 'floats' overhead," he says. "Others use an installable file system or stub to replace the operating system's native file system."

Whatever the technique, many users find promise in the idea of centralizing storage resources in a single pool. In a recent survey of 96 network professionals attending Network World's summer seminar tour, "Storage Town Meeting: Ensuring Business Continuity," two-thirds of respondents said they do not yet have geographically dispersed NAS and storage-area network (SAN) environments. But of those who do, the desire to manage them as a single storage pool was nearly unanimous (see graphic).

DFS education

Such is the case at University of Hawaii (UH), in Honolulu, which has been using Acirro's Acumula DFS software in a pilot program launched in April. The pilot enables sharing of educational materials among 10 UH campuses, across four islands, letting students participate in classes regardless of their physical location, says Brian Chee, a systems programmer and associate director in the university's Advanced Network Computing Laboratory. The program also enables information sharing among UH and Hawaii's K-12 schools.

The Acumula smart replication and storage management software is based on Acirro's Global Distributed File System technology. The software lets users combine data across storage devices, networks and formats. They can aggregate diverse NAS and SAN devices regardless of physical locale, and get one view of storage files.

"Server aggregation seems to be a pretty good driving force," Chee says, pointing out that his department has collected a variety of stand-alone servers, ranging from Windows 2000 running Microsoft's CIFS, Unix boxes running NFS and "dumb" NAS boxes.

"Going to a system that will present all of our data on a single login is a good thing," he says. "This type of system also allows me to further hide the NFS boxes behind firewalls and get quite a bit more aggressive on firewall rules."

UH hasn't yet decided to roll out DFS beyond the pilot program, but has seen operational improvements. For example, Chee says the DFS has reduced the amount of human intervention traditionally necessary to synchronize local Web and FTP storage of classroom support materials.

Switching files One technologist shares his thoughts on the future of distributed file systems for storage. Distributed file systems for storage may be just the beginning. The future of DFS really lies in file switching, says Barry Lynn, management consultant, venture capitalist and former CIO. "As storage technology prices have, exponentially, over the years, become lower and lower, the justification for putting more and more online has materialized," Lynn says. Click here for more.

"Some materials, such as video [and] compiled Java libraries, can become quite large," he says. "Instead of forcing the students to wait for a long download over the WAN link between campuses, DFS provides a way for faculty to drop their material onto locally mounted devices. They can expect the material to show up on outer-island campuses without human intervention," he explains.

Chee adds that DFS for storage works well with the file systems that come with NAS devices, including NFS for Unix boxes or CIFS for Microsoft servers. "The promise is tremendous," he says. "I will eventually have a mix of both Active Directory and NFS/NIS authentication. This system will allow me to front-end both environments and allow me to integrate both together for the first time. It is going to make it possible to consolidate servers logically so that my less-sophisticated users can stay in their native environment instead of having to eat the learning curve for something new."

Other DFS tools also are geared toward heterogeneous environments. ADIC's CentraVision 2.0 software, for example, is intended to let heterogeneous clients share files across high-speed storage network connections. ADIC says with CentraVision, users can share storage resources among Linux, Windows and Unix servers so all storage looks native to all hosts. The file system "owns" the storage and replaces individually managed volumes and NAS devices. For example, data on a Fibre Channel disk is shared through servers to many hosts without separate controllers.

But Clipper Group's Kahn cautions enterprise users to take care when deploying DFS in multivendor, multiplatform environments. "DFS products tend to work best in homogeneous environments; for heterogeneous environments, they tend to be immature and not high-performing," Kahn notes.

The politics of DFS

Network professionals who want to try DFS storage software must navigate more than the waters of new technology. DFS, because it layers a file system over NAS and SAN devices, can stir up choppy intercompany political waves over ownership.

UH's Chee can see the potential difficulty. "At this moment, management of my lab servers [Active Directory] and my Unix machines [Lightweight Directory Access Protocol] is not run by the same person," he says. "While such a move is possible - integrating two relatively independent groups will be a challenge."

Even so, Chee is eager to tap into the DFS storage pool. He predicts that even if there's a DFS collision of sorts between his more "backroom" Active Directory folks and the Unix folks who like to mix more with the UH faculty, everyone will eventually learn to swim without drowning. "I can still see such integration as being possible and good for the department in the long run," he says.

Ryan-Garcia is a freelance journalist in Coram, N.Y., specializing in the storage industry. She can be reached at freshcontent@aol.com.

Topics: Storage
Stay on top of the latest storage news, reviews, opinions and more.

Buyer's Guide: Storage-area networks
See how SAN switches compare and how they can be used for business continuity. Also, download our Interactive Buyer's Guide to compare the offerings of six of the top storage-area networking vendors.

Storage newsletter
Get the latest news, information and analysis on storage and application management so you can make the best decisions.

SAN audio primer
In this 5-minute primer, you'll learn how a SAN works and its benefits over traditional storage systems.