Of File Systems and Databases: WAFL, Network Appliance, and Spinnaker Networks
WAFL enabled a new type of appliance-based storage hardware and set the bar for all other NAS vendors—which makes Network Appliance’s purchase of Spinnaker Networks all the more puzzling.
Having just returned from a visit to Silicon Valley, it is interesting to note how many storage companies are working on scalable file systems. Network Appliance’s recent purchase of Spinnaker Networks (for $300 million), a purveyor of scalable file system-based network-attached storage, underscored this phenomenon, and has prompted IT consumers to focus increased attention on the technology. It's also the focus of this article, the first of a two-part discussion.
First, a few definitions. A file system consists of a standardized way for identifying blocks of data via a system of file names. Essentially, metadata is added to the blocks to allow their retrieval simply and expeditiously by using a file name descriptor. Any way you slice it, a file system is a method for virtualizing block storage so that data can be retrieved for use in a more convenient way.
File systems are typically controlled and restricted by operating system parameters. For example, most operating systems limit the maximum number of file names that can be handled within their file system. Some also limit the aggregate block size of files that can be managed under a file system.
Scalable file systems seek to surmount these boundaries. Typically, they replace or augment existing, or OS-native, file systems and provide the means to overcome the constraints of OS-native file systems by introducing metadata servers or other technologies to improve file storage and request handling.
A significant innovation in file system design was Network Appliance’s Write Anywhere File Layout (WAFL), introduced in 1995. WAFL was conceived initially as a UNIX-compatible file system optimized for network-based file access using the Network File System. WAFL was similar, in many ways, to the Berkeley Fast File System (FFS) and TransArc's Episode file system. The vendor’s white paper, online at http://www.netapp.com/tech_library/3002.html#I31, provides a complete overview of WAFL, which is summarized this way: WAFL was a block-based file system that used inodes to describe files, and wrote its data in 4 KB blocks with no fragments. Each WAFL inode contained 16 block pointers to indicate which blocks belonged to the file. Unlike FFS, all the block pointers in a WAFL inode referred to blocks at the same level of the inode tree.
Thus, inodes for files smaller than 64 KB used the 16 block pointers to point to data blocks. Inodes for files larger than 64 MB pointed to indirect blocks, which, in turn, pointed to actual file data. Inodes for larger files pointed to doubly indirect blocks. For very small files, data was stored in the inode itself, in place of the block pointers.
This file system innovation enabled the development of the network-attached storage “appliance”—accessed via Network File System (and later via the Common Internet File System or CIFS) to provide robust storage of files from both UNIX and Windows hosts. The new file system helped to surmount the “write penalty” that usually accrued to storing files to an NFS-mounted volume (particularly if that volume was itself RAIDed). Moreover, Network Appliance also built in a nifty snapshot feature that made data stored to appliances eminently fault tolerant, and made the whole system more recoverable in the event of a “dirty shutdown” (loss of power, equipment fault, etc.).
Basically, WAFL enabled a new type of appliance-based storage hardware and set the bar for all other NAS vendors—which makes Network Appliance’s purchase of Spinnaker Networks all the more puzzling. Spinnaker’s approach to file system design is, in many ways, the antithesis of WAFL, which sought to enable vertical scaling—the addition of more and more drive capacity behind a single NAS head—combined with continued high performance. Spinnaker’s play was to use clustering technology and a scalable file system that could be extended across multiple heads to enable horizontal, rather than vertical, scaling.
Network Appliance calls the two approaches “complimentary” and has given itself roughly three years to prove it. One has to wonder whether this is, in effect, the death knell for WAFL, or perhaps the introduction of a completely new product family at Network Appliance, or just another instance of a smaller upstart being taken out of the market by an incumbent before it can become a real threat.
Next week I'll look at clustered file systems, their niche fit, and their future outlook.
Jon William Toigo is chairman of The Data Management Institute, the CEO of data management consulting and research firm Toigo Partners International, as well as a contributing editor to Enterprise Systems and its Storage Strategies columnist. Mr. Toigo is the author of 14 books, including Disaster Recovery Planning, 3rd Edition, and The Holy Grail of Network Storage Management, both from Prentice Hall.