This article can also be found in the Premium Editorial Download "Storage magazine: Expanding SANs: How to scale today's storage networks."
Download it now to read this article plus other related content.
|SAN file system at a glance|
IBM's SAN FS roadmap
IBM's grand plan is to make SAN FS a heterogeneous distributed file system that supports multiple Unix and Windows clients and storage arrays. In this first release, however, operating system support includes IBM's own AIX, AIX HACMP, Windows 2000 and Windows 2000 Advanced Server operating systems. As far as storage arrays are concerned, SAN FS supports IBM's Enterprise Storage Server (Shark), as well as storage virtualized behind the SAN Volume Controller (SVC), the block-based virtualization appliance that IBM announced last spring.
In 2004, IBM will extend operating system support to Linux, Solaris, Solaris clusters and Windows 2000 clusters. Last spring, IBM also released source code to SAN FS, such that third parties could write agents for additional operating systems. Next year, storage array support will move beyond IBM storage with the help of SVC, which was recently enhanced to support arrays from Hitachi Data Systems (Thunder 9200) and Hewlett-Packard (MA 8000, EMA 12000 and EMA 16000), according to Bruce Hillsberg, IBM director of storage software strategy.
SAN FS is built upon four major storage technologies. They are:
- Global namespace. All application servers running SAN FS share the same files, which are accessed and named in a single, common way. The main benefit of a global namespace is that the environment typically requires less capacity, because, as Truskowski puts it: "The data is the data." In other words, multiple applications can share a single instance of a file, rather than relying on copies. That in turn can reduce data movement, simplifying application workflow, and improving performance.
- Storage pools. SAN FS also incorporates the notion of storage pools, or the ability to create groups, or classes, of storage. For example, under SAN FS, an administrator can specify a high-end class of storage that provides mirrored failover, a midtier class for reference data, and a low-end tier comprised of low-cost JBOD for storing temporary files. Storage pools enable another major SAN FS feature: policy-based file space provisioning. Working from a central SAN FS administrative console, administrators can set policies across all application servers in the SAN FS environment, which can be set according to "a variety of variables" says IBM's Hillsberg, including file extensions, application server, user name and access time, for example. A nice extension of policy-based file management is the ability to establish soft or hard quotas across your storage pools, he says.
- File-based snapshot. Then, there's the SAN FS feature called File-based FlashCopy Image, a snapshot implementation. But unlike existing snapshot products on the market, which are block-based, FlashCopy Image doesn't just snap an entire volume, but can snap all the way down to a specific file or directory. "This brings the concept of point-in-time copy to a much more granular level," says Hillsberg, with numerous implications for data protection and backup.
- Data migration. SAN FS introduces the notion of Automated Volume Drain for Volume Migration. The translation: easier data migration. When it comes time to migrate data to a different storage platform, an administrator can specify which volumes to remove from a pool. SAN FS performs the data movement and keeps track of where a file resides at any given time. In other words, an application server requesting a file can still retrieve the data, even if it is in the process of being migrated.
That's in contrast to how clustered file systems have fared thus far. To date, most users of these file systems have been heavy consumers of storage working in say, the oil and gas industry, or in video production. What these industries have in common is their need to handle extremely large numbers of large files, as well as an inherent need to share them between users and application servers.
It's not that IBM will ignore the core technical and creative market--initial customer references include Johns Hopkins University and CERN's particle physics research lab--but IBM's Truskowski says SAN FS was also designed to provide "good performance for business applications."
This was first published in November 2003