More storage, now. And make it faster! The explosion of data gathering in today's online business environment means that your storage requirements keep going up, and the amount of data that you need to process and analyze also grows. Unfortunately, the tools we've developed and implemented so far are having a hard time keeping up.

The answer may be the Direct Access File System (DAFS), a protocol that recognizes the speed and reliability of today's Gigabit Ethernet and InfiniBand protocols, provides for remote direct memory access (RDMA) between applications running on separate machines in a cluster or LAN, and is built around files, not blocks.

To understand DAFS better, let's review the differences between a storage-area network (SAN) and network-attached storage (NAS). A SAN is a complete subnetwork dedicated to storage and is connected to one or more servers. Connections to a SAN are made over a high-speed protocol such as Fibre Channel or iSCSI. SAN storage is accessible from all servers, so users can access any storage device on the SAN, regardless of the physical location of the storage or users. SANs were designed to help manage and speed storage by simplifying the data path and taking hard-wired servers out of the loop, but they move data in low-level blocks, thus necessitating a translation to files to use the data.

NAS is simpler than a SAN; basically it's shared, hard-disk storage that's given an integral, dedicated server and its own network address. A NAS appliance attaches to a network without powering it down and requires no changes to the existing file servers. NAS was designed to make it easy and inexpensive to add storage, and data is handled in files, the way the user expects.

Unfortunately, both systems incur considerable processing overhead, which slows data transfers.

DAFS uses the Virtual Interface (VI) architecture, which was designed in 1996 by Microsoft Corp., Compaq Computer Corp. and Intel Corp. as a data transport mechanism. VI is a high-speed, low latency (i.e., low waiting time) network protocol that lets applications on different machines in a LAN or cluster read and write to memory addresses used by other applications in the network. This memory-to-memory interconnect is called RDMA.

DAFS also reads and writes using a file-based protocol, so it preserves information that would otherwise have to be rebuilt every time it's fetched. File requests to open, update, append, lock or close are exchanged between users' storage clients and storage servers.

Thus DAFS combines the speed of a high-speed SAN and the simplicity of file-based NAS, and does it with lower overhead than either.

DAFS's file-based storage can be managed by several different types of products and also promises to remove some of the internal processing that current block-based databases must perform.

History

Early work on DAFS was done by Network Appliance Inc. in Sunnyvale, Calif. DAFS is now spearheaded by an 80-plus member industry association called the DAFS Collaborative. The collaborative has completed Version 1.0 of the DAFS specification and turned it over to the Internet Engineering Task Force standards group and to a newly created DAFS Implementors Forum, a subgroup of the Storage Networking Industry Association.

DAFS was first demonstrated in action with an Oracle database in December 2001, but no products have yet reached the market, making it unclear whether DAFS will succeed. Some NAS and subsystem companies don't believe that DAFS is really an open standard, but others believe that it's a genuine step in the evolution of NAS technology that could deliver significant benefits.