IBM R&D Aimed at Storage

Despite the ivory tower connotation of its name, IBM Corp.'s (NYSE: IBM) Almaden Research Center has practical applications in mind for some of its imminent storage developments.

IBM is targeting areas such as file-system extensibility and disk reliability -- not the sexiest of topics in storage circles, but among the more pressing concerns of storage managers today.

Researchers at Almaden, the company's facility in San Jose, Calif., discussed their progress during a visit by Byte and Switch last week. Almaden is one of eight IBM labs and home to much of the company's storage research.

Among the promising developments was the long-running Storage Tank project. "Storage Tank has turned out to be a great incubator for many different, second-generation projects," says Linda Duyanovich, manager of storage software and storage systems at IBM Research. "It's lent itself to a variety of different extensions."

Started seven years ago, the Storage Tank work emerged a year ago in the form of IBM's SAN File System (SAN FS) (see IBM's Big Fat File System). The product virtualizes multiple file systems on a SAN and presents them through a single global namespace, providing file sharing across heterogeneous systems.Though SAN FS was a few years late to market and short on expected functionality -- much of which has since been remedied -- the projects it spawned seem to be on a faster track.

The first, code-named Hydra, creates a cluster of existing file systems across multiple data centers. Supporting NFS and CIFS file protocols, Hydra load-balances file systems over heterogeneous servers in its cluster. If a server goes down, the system redirects access requests to another node.

Duyanovich compares Hydra to the clustered NAS approach of Network Appliance Inc.'s (Nasdaq: NTAP) Spinnaker product. As an eventual product, Hydra will span hundreds of file servers (as the Spinnaker product also claims to do) and be integrated with and administered from the SAN FS console.

Another offshoot of Storage Tank, dubbed Federated File System, tackles distributed file access. Here, the system relies on policy-driven replication and caching techniques to tie together different file systems over multiple locations and maintain a single namespace. The so-called federated file systems could include SAN FS as well as other local file systems, though only those supporting NFS.

On the hardware front, IBM researchers are eyeing cheaper disk drives like ATA that are making their way into enterprise systems. The problem with these drives is their diminished reliability compared with more expensive drives such as Fibre Channel. So the researchers focus on techniques, including RAID, that improve reliability.Almaden researchers have developed a "Distance 4" scheme that they say improves the reliability of RAID 6 implementations by more than 1,000 times. RAID 6 describes data-striping techniques that can tolerate two disk failures, as opposed to one disk failure with other types of RAID.

IBM's scheme also improves on storage efficiency compared to other RAID methods, says K.K. Rao, distinguished engineer for storage systems at IBM. Rao declined to provide specifics of the IBM research, but noted that it will primarily benefit applications with demanding performance requirements.

The project, he says, could take a couple years before it emerges in products.

Brett Mendel, Senior Analyst, Byte and Switch Insider