Analysis

posted on 31 March 2008 10:02

IBM's Scale-out File Services (SOFS) layers a clustered network-attached storage (NAS) head on top of IBM-only storage arrays to provide a highly scalable file storage resource. It is not a product though.

The end-result is a storage resource able to cope with enormous file storage capacity needs and high numbers of servers needing file access in parallel, just the kind of thing provided by Isilon, Panasas and newcomer Atrato as well as being promised by IBM's XIV technology and EMC's Hulk/Maui combo.

It can't be provided by NetApp as ONTAP GX isn't yet generally available with the kind of scale-out capacities needed, in which nodes can be added as needed with transparent inclusion into a global file name space.

IBM's is a services-led approach, making it expensive. It is based on its GPFS (General Parallel File System) which runs in pre-configured hardware nodes. IBM is coy about the hardware and software in these nodes with IBM blogging spokesperson Tony Pearson praising their performance, saying the nodes can run CIFS at 577MB/sec and NFS access at 880MB/sec over a 10GbitE pipe.

The node software includes GPFS plus a global namespace and it also includes ILM (information lifecycle management) capability with multiple disk tiers - FC, SAS, FATA, SATA - included running the performance to capacity gamut, plus a tape tier for archiving less-needed files.

The nodes can be clustered, over Ethernet presumably although IBM's website SOFS resources don't say.

The back-end disk storage can be IBM storage such as the DS3200, DCS9550, DS8300 Turbo and, interestingly the SVC (SAN Volume Controller) with potentially heterogeneous storage behind a Fibre Channel fabric switch to which it is connected. It can be any NetApp N-Series or any other NAS product. The back-end storage can also include a tape library with software included to move files to and from the tape.

This is reminiscent of the Quantum StorNext Management Suite running on a server attached to apple's Xsan which does the same sort of thing.

The newer scale-out file storage products - Panasas, Atrato, and XIV and Hulk/Maui seem to be predominantly designed to provide fast parallel access to masses and masses of files. Tape doesn't figure and nor, as a prime design focus, does ILM. With IBM's SOFS the ILM aspect seems to have been designed in from the start, and of equivalent importance to the file capacity and access scaling.

It may be that SOFS is a first generation scale-out file services product with the XIV technology being developed to provide the second generation product.

^{[Chris Mellor.]}