IBM SONAS Scales Out Storage with Parallel File System

IBM has announced its Scale Out Network Attached Storage (SONAS) system, an integrated storage solution based on IBM’s General Parallel File System (GPFS) technology. Although SONAS is a new offering, GPFS has been available for more than a decade as the storage component of the IBM Cluster1600 and Cluster1350 solutions, and as the Scale-out File Services (SoFS) service offering. Now, with integrated solutions becoming increasingly fashionable in the market, IBM has taken the next step of coupling GPFS with storage hardware in an appliance-like enclosure. SONAS leverages other clustered storage technologies besides GPFS, including Samba CTDB, which enables parallel accessing and high-availability (HA) for Samba. The result is a new class of NAS platform that is extremely scalable, robust, and high performing.
The modular SONAS design is optimized for adding I/O bandwidth and storage capacity independently of each other. Other designs adhere to a fixed ratio of I/O bandwidth and capacity, which can lead to underutilized resources. With SONAS, systems can be balanced for unique customer needs.

SONAS employs IBM System x servers (x86-based) as Interface Nodes and Storage Nodes, and the high-density EXP5060 enclosures (60 disks in 4U) as the storage subsystem. SONAS supports up to 30 Interface Nodes and 30 Storage Nodes. Each storage node supports up to 240 x 1 TB disks, for a total of up to 7.2 PB of raw storage per SONAS system (the capacity will double in April 2010 when 2 TB drives become available). SONAS targets environments with extraordinary performance and capacity requirements for file serving operations – environments that cannot be adequately served by IBM N series appliances. The N series appliances, which are rebranded NetApp filers, are based on a dual-controller architecture and the top model supports up to about 1.2 PB of raw storage using 1 TB SATA drives.

There are many general-purpose parallel file systems that originated in the high-performance computing cluster (HPCC) space, such as Lustre, Parallel Virtual File System (PVFS), GlusterFS, IBRIX Fusion, and GPFS. However, most of these file system technologies are available as software and require a high-level of technical expertise to implement. Besides the IBM GPFS-based SONAS, only IBRIX Fusion is available as integrated appliances from HP. The HP X9000 products, based on IBRIX Fusion, provide three models including an integrated model (X9320), a gateway model (X9300), and a blade-based integrated model (X9720). HP claims that multiple X9000 systems can join a single namespace, but it does not specify the upper limit. (The IBM SONAS scalability is an artificial limit; GPFS natively supports thousands of nodes.)

Both the HP X9000 and IBM SONAS solutions provide storage tiering (partitioning) and information lifecycle management (ILM) functions. However, SONAS ILM is further integrated with IBM Tivoli Storage Manager (TSM) for hierarchical storage management (HSM) for moving data across different platforms including tape (a feature planned for release in the near future). The X9000 and SONAS both use Samba for SMB/CIFS file serving, but the X9000 does not integrate with Samba CTDB (CTDB considerably enhances Samba performance). While both SONAS and the X9000 support snapshot, the snapshot function is only available on one of the three X9000 models (the X9320). However, X9000 supports a Continuous Remote Replication feature that can be used for extended distances, while SONAS currently supports only short-distance replication (long-distance remote replication is planned for later this year)

Parallel file systems were traditionally deployed primarily in HPCC environments, supporting technical workloads on large server farms. However, they clearly have the potential to help increase the scalability and performance of commercial workloads as well. To replace typical commercial NAS systems, though, solutions based on parallel file systems need to reduce implementation complexity and offer equivalent or better data protection and management features. The IBM SONAS greatly streamlines the GPFS deployment by pre-integrating best-of-breed components into an appliance form factor that is simple to implement. Upon the delivery of planned features, SONAS will also present a feature-rich NAS platform that is competitive with other leading commercial NAS platforms. With limited competition (only the HP X9000 is a close match), SONAS is likely to open many new doors for IBM to further penetrate the enterprise NAS market.

Comments

[Joseph Zhou]

Thanks Barbara and Rob. I never thought PanFS and OneFS were in the same class as Lustre, GPFS, and IBRIX, since PanFS and OneFS only work with proprietary hardware from Panasas and Isilon. But you are right, although GPFS and IBRIX support standard hardware, as integrated offerings, SONAS and X9000 do compete with Panasas and Isilon solutions. To complete the list, I guess we also need to add NetApp Data ONTAP 8.0 Cluster Mode – only supports NetApp hardware as well.

[Rob Anderson]

You also failed to mention Isilon, which has been shipping a clustered filesystem (OneFS) on hardware since 2001. GPFS and IBRIX have been software-only products for most of their lifetimes. Isilon has been a combined hardware/software solution since day 1.

[Barbara Murphy]

Hi Joseph,
You neglected to mention Panasas when you spoke about alternative solutions to GPFS. For the record, Panasas has been shipping an end-to-end hardware and software solution with its own highly competitive parallel file system since 2004, and has major installations in all the major HPC verticals.

[StorageBuddhist]

Joseph, one correction. SONAS does not use the LSI-based EXP5060 disk drawers. The backend storage technology used by SONAS is the DDN 6620 controllers and drawers.

[Jordan True Flight]

Thankyou for a very entertaining and enlightening piece. It definitly opened my eyes to allot of things I had not thought of before.