In-Depth
Enterprise Storage: Make DAFS, not SANs
A few weeks ago, I was quoted in a press release from Network Appliance, stating some favorable things about the Direct Access File System (DAFS) protocol. If you haven’t heard about DAFS, I am willing to bet that you soon will.
Network Appliance, the company that first added substance to the idea of a network-attached storage (NAS) array, is currently the dominant player in the NAS market it helped to create. Among its many accomplishments was the creation in 1995 of the Write Anywhere File Layout (WAFL) filing system that enabled data to be stored in a multiprotocol-accessible format.
In other words, data stored in WAFL (pronounced "waffle") could be shared among UNIX servers and workstations via the Network File System (NFS), or between Windows servers and workstations via Microsoft’s Common Internet File System (CIFS), or among Web servers and browser-equipped workstations via Hypertext Transfer Protocol (HTTP). To paraphrase the company’s white papers and marketing literature, "Multiprotocol filing liberates the data infrastructure, largely freeing it from the constraints of operating system preference or legacy investments."
The WAFL idea was a solid one, as Network Appliance’s record of 70 percent annual revenue growth since 1995 suggests. Today, it is rare to survey the data center of a large Web-hosting firm without seeing a NetApp Filer system. Plus, the vendor’s client list within the brick-and-mortar world reads like a Who’s Who of modern business.
DAFS is the latest innovation spearheaded by the company. And, in my humble opinion, it promises to continue the revenue trend lines for the company high and to the right for the foreseeable future.
What’s more, DAFS may prove to be the breakaway technology that will finally move Storage Area Networks (SANs) out of the backwaters of proprietary vendor offerings and into the realm of a viable business IT infrastructure.
Where DAFS Fits
Explained simply, DAFS is another network-borne protocol for sharing access to data stored on NAS arrays. One might wonder why we need another file sharing protocol given the relative prevalence and stability of NFS, CIFS and HTTP. Good question.
Currently, network file system protocols solve a specific set of issues. Fundamentally, they provide "elbow room" for enlarging the space for storing files and enable the sharing of storage volumes among many servers and users. Shared volumes accessible via NFS, CIFS or HTTP may be installed within, or tethered to, servers, or they may be part of a NAS array.
Most NAS arrays support only one of the popular network file system protocols -- NFS, CIFS or HTTP. Network Appliance filers, armed with WAFL, support all three. WAFL is a universal file layout that enables requestors to access files stored on the NAS using whatever protocol they require. Upon request, the files are formatted and shipped via the requested network file system protocol.
Neat as this all sounds, traditional NAS only addresses part of the storage equation. Not all storage requests are handled through file system requests. For example, transaction-oriented applications and databases often read and write data from and to storage media using physical addresses of storage "blocks," rather than file names. Filename-based storage, in fact, is an abstraction: When you request a specific file by its name, the actual retrieval of the requested data is handled on a "block" basis by the operating system. The process is concealed from the end user intentionally -- to reduce fear-inducing complexity and anxiety-related hair loss -- in other words, to enhance user friendliness.
Over the past few years, there has been a fairly clear bifurcation between file system and block-based storage technologies. While any block-based storage array, including an EMC Symmetrix, could be used for file system storage, these high-priced products were designed to appeal to customers seeking high-speed block-level file access.
On the other hand, NAS has evolved over time as the penultimate file system storage technology. In most file system-based applications, the number of simultaneous input/output requests is significantly lower than in a transaction system. The emphasis with NAS has been to provide unlimited room for growth for stored files, to facilitate access by authorized machines and persons, and -- above all -- to make implementation as user friendly as the file system itself: plug and play. Plus, the cost of most NAS devices was significantly lower than their cousin block-level storage arrays.
Until now, if a block-level array vendor was asked about the possibility of using a NAS array to do transaction system storage, he would smile confidently across the table, while chewing a bite of his rare, two-inch-thick, aged-and-marbled steak, put down his fork, dab his chin with his napkin, take a sip of his $600-per-glass red wine, then look you square in the eye and say, "A NAS array can’t do block level storage." He would go on to say that network file system protocols are too inefficient to support block level operations, especially in high transaction environments. The protocols themselves consume too much overhead to execute with sufficient speed. "For industrial-strength block-level storage, you need an Big Iron block storage product."
He would probably have been correct … until DAFS.
The initial DAFS specification is on-line at www.dafscollaborative.org. The DAFS spec anticipates direct memory-to-memory file access and leverages the forthcoming Virtual Interface (VI) architecture as its underlying transport mechanism. VI allows bulk data transfer directly to or from application buffers and permits applications to access VI-capable hardware directly without operating system intervention. In short, according to backers, including Intel Corporation, Seagate Technology and Network Appliance, DAFS will enable heterogeneous, block-level, data sharing at sufficient speeds and feeds to support high-volume, high-performance transaction environments.
Work to Be Done
Obviously, DAFS will not appear overnight. There is significant work to be done with the specification before it is ready for submission to a standards body. What is exciting to me about DAFS is that it potentially enables the deployment of a SAN-like solution with the plug-and-play simplicity of NAS.
DAFS is a meaningful initiative that lends credence to ideas that have been floating around the industry for the last year that NAS devices would eventually become "portals" to back-end SANs. While many NAS devices consist of captive storage disks inside a thin, network-attached server, Network Appliance and others are thinking outside of the box and considering ways to attach a network of storage products behind the appliance itself. Such a strategy would enable the SAN to be deployed -- and managed -- readily using the NAS thin server OS.
In late July, Werner Glinka was tapped to serve as Executive Director of the DAFS Collaborative. Glinka, an affable fellow with 24 years in the storage technology business, knows his way around industry collaborations and standards bodies. He will have a struggle ahead of him getting all of the storage vendors who are participating in the collaborative to work and play well together, if the experience of the iSCSI initiative within the Internet Engineering Task Force is any indication.
Personally, I wish him the best, and encourage all readers to learn more about DAFS, and to support this important effort to make practical and open SANs a reality.
About the Author: Jon William Toigo is an independent consultant and author of The Holy Grail of Data Storage Management. He can be reached via e-mail at jtoigo@intnet.net.