IBM delivers SAN file system, falls short

Support to be delivered in phases

By Heidi Biggar

After several years in development and a number of false starts, IBM this month began shipping the first iteration of its TotalStorage SAN File System, more commonly known as "StorageTank."

The TotalStorage SAN File System, a by-product of the company's Almaden Research division, is designed to facilitate high-performance heterogeneous file sharing and access in a storage area network (SAN) environment.

The first release enables heterogeneous file sharing and fast data access but only in IBM disk array and AIX/Windows environments. With additional support (reportedly for Hitachi Data Systems and Hewlett-Packard disk arrays) loosely slated for 2004, critics say IBM has left the door open for the possibility of another series of technology miscues next year.

Despite the criticism, StorageTank has received a lot of positive attention from analysts and even some competing vendors.

"This is the announcement we've been waiting for, for a very long time," says Steve Kenniston, a technology analyst at the Enterprise Storage Group (ESG). "People may give IBM a hard time for taking so long to deliver it, but this type of technology isn't something that's developed in a few months and then thrown out into the market," he says.

The complexity of this type of technology is something that Gabriel Broner, senior vice president and general manager in SGI's storage solutions and software group, is well aware of: "Its lengthy development time shows that creating such a file system is not easy."

Broner says it took SGI several years to "stabilize" its CXFS shared file system to the point where it could support heterogeneous operating systems and enable concurrent file sharing. SGI's file system currently supports AIX, Irix, Linux, Solaris, and Windows (see "SGI stages storage comeback," InfoStor, October 2003, p. 16).

SGI is among a growing list of companies, including ADIC, Sistina, Sun, and Veritas, that offer some type of shared or distributed file system for storage network environments.

"Customers want to be able to deal with files, servers, and storage of different types, and they want to be able to share files among these different environments," says Clod Barrera, a technology strategist at IBM. "They want a file system that can scale very large and deliver very high performance."

"IBM's announcement is a validation of the type of solution ADIC has been delivering for several years," says Bill Yaman, vice president of software at ADIC. "While IBM has the right to announce this technology, the challenge it faces is in delivering on the [complete] vision."

ADIC's StorNext File System has been available for three years and supports AIX, Irix, Linux, Solaris, and Windows, as well as a variety of heterogeneous storage devices. It is one of two components in the company's StorNext Management Suite; the second component—StorNext Storage Manager—provides data life-cycle management capabilities.

But what IBM lacks in server and array support today, it will probably make up for in sheer power and scalability down the road, according to some consultants. "StorageTank is meant to be a real SAN operating system," says Richard Lee, president and CEO of The Storage Consulting Group. "This is much different than a storage-optimized file system like SGI's or ADIC's, which are host-based and can neither scale nor support the scope and depth that StorageTank will, at least in principle," he adds.

IBM claims that its SAN File System will be able to manage and process petabytes of information. For example, the European Organization for Nuclear Research (CERN)—one of IBM's early users—is expected to use StorageTank to manage 10TB to 20TB of data this year and as much as one petabyte in 2005, according to IBM officials. However, IBM's Barrera says that he expects the file system to be implemented in environments with tens of servers as well as environments with hundreds or thousands of servers.

ESG's Kenniston expects clustered file systems to initially have greatest appeal among users with huge file sets that need to be shared, and in environments running parallel database applications. "There aren't a lot of companies that need this technology today, but the requirements being put on IT will slowly cause this technology to become more widely adopted. The good news is that this will give IBM time to address any issues going forward," says Kenniston.

StorageTank is built on top of a virtualization layer, which is intended to give the storage/server environment a single look and feel, simplify management, and make it easier to perform services (e.g., provisioning and backup) across a distributed SAN environment. IBM accomplishes this by creating a "virtual file system" that has a single global namespace to facilitate data sharing, among other things.

According to IBM officials, StorageTank works by tying together servers in multiple locations over an IP network and then allowing the distributed storage network to look and behave like a local file system, no matter where or on what operating system the data resides.

Also, unlike shared file systems that store metadata in the storage devices themselves, StorageTank spreads this information out across the servers (a lightweight client is installed on each participating server). Doing so, according to IBM, enables the file system to manage and process high volumes of data and to scale into the petabyte range.

StorageTank supports open standards and includes a Common Information Management (CIM) agent. The file system also allows users, via policy-based automation capabilities, to establish file pools that can be automatically allocated to supported disk depending on storage requirements.

Separately, IBM also announced several enhancements to its TotalStorage SAN Volume Controller (SVC) software, including support for Hewlett-Packard and Hitachi Data Systems disk arrays, as well as an SVC blade that will be embedded into Cisco's MDS 9000 Series Caching Services Modules. IBM is also reportedly working on integrating its two virtualization platforms and will introduce a common management tool (under the Tivoli brand name) for the two platforms in the first quarter.

This article was originally published on November 01, 2003