IBM upgrades brain-bending file system

GPFS - it's about policy

By Austin Modine in Mountain View • Get more from this author

Posted in Storage, 2nd October 2007 22:49 GMT

Free whitepaper – Reliability analysis of the APC Symmetra MW Power System

IBM on Friday is releasing a new version of its General Parallel File System (GPFS) for serious data crunchers.

The update, now version 3.2, features improvements to GPFS's policy-based file management system and offers speedier searches.

GPFS caters to hardcore users trying to run a file system across numerous systems. It provides shared access to the files regardless of what box in particular they are sitting on. IBM says GPFS can support access speeds of 130+GB/sec to a single file on a 2PB file system.

The update includes a revamped version of the existing, built-in policy manager that should make tweaking rules for storing and shifting data easier on customers. This type of technology comes in handy when tiering data across systems that have billions of files to manage, said Todd Neville, IBM development offering manager.

The file system can, for instance, let users specify that a certain type of file will be stored on higher performing disks. Users could also move all specific file types to lower-end systems if they have not been accessed for a certain amount of time, said Neville.

IBM said a pre-release version of the fresh GPFS was able to scan one billion files in less than three hours in an internal benchmark. They say with further improvement of the policy performance through parallelization techniques, they'll whittle the number down even further.

Other improvements include an accelerated file identification process for managing tiered storage, and support for pools of storage on tape. IBM has also added clustered management features.

Version 3.2 supports IBM System p systems, including the Power6-based IBM System p 570 server and machines based on Intel or AMD processors such as the IBM System x boxes. Supported operating systems include AIX Version 5.3 and some versions of Red Hat and SuSE Linux.

IBM's GPFS competes against homegrown software from the likes of SGI and Sun Microsystems along with open source and start-up offerings. ®

Free whitepaper – Selecting an Industry-Standard Metric for Data Center Efficiency