Critical Capabilities for Scale-Out File System Storage

Key Findings

• Most leading, scale-out file system storage vendors, which focus primarily on high-performance computing (HPC) environments, are adding capabilities to make products more ready for varied enterprise workloads, but some gaps remain.
• While all products rated performed reasonably well in capacity scaling, wide disparities exist for most other critical capabilities. Storage efficiency and manageability need substantial improvement.
• All major storage vendors have a presence in scale-out file system storage, primarily through acquisitions, alleviating customer concerns for the products' long-term viability.
• Customers still face skills shortages, uneven support and diverse management consoles.

Recommendations

• Take a top-down design approach that focuses on workload characteristics so that applications can optimize the scale-out nature of the underlying storage.
• Include an adequate training budget in your procurement, as managing scale-out file system storage differs from storage area network (SAN) management, and storage administrators may need more training.
• Examine thoroughly the scale-out file system storage product's interoperability capabilities to ensure support for the dominant independent software vendor (ISV) software and virtualization platforms in your IT environment.
• Develop a two-year road map for the critical capabilities, and match them to the vendor's road map.

The growing strategic importance of storing and analyzing large-scale, unstructured data is bringing scale-out storage architectures to the forefront of IT infrastructure planning. Scale-out storage often is capable of nearly linear horizontal scaling, and delivers highly aggregated performance through parallelism. Most scale-out storage vendors tend to use standard x86 hardware, thus reducing the hardware acquisition price, and embed the storage intelligence in the software layer. The major target market segment of scale-out storage vendors has traditionally been in academic and commercial HPC environments for workloads such as genomic sequencing, financial modeling, 3D animation, weather forecasting and seismic analysis. As a result, scalable capacity, raw compute power and aggregated bandwidth have been the major focus of the products, with data protection, security and efficiency as secondary considerations. However, the growing enterprise demand for large-scale storage to efficiently store and protect unstructured data is compelling vendors to bridge the chasm, and to address customer demands for better security, manageability, data protection and ISV interoperability. While many products still rate only modestly as general-purpose storage arrays, the tide is definitely turning in that direction.

IT organizations must carefully select a scale-out storage vendor through a rigorous planning process that involves thoroughly evaluating the vendors' critical capabilities. Vendors continue to fine-tune their products to focus on specific use cases, although the leading vendors in this research cater to a wide variety of use cases in enterprise environments. However, awareness of scale-out storage and global name spaces isn't common in enterprise IT, so training should be an important component of the budgetary allocation.

This research is designed to compare three common use cases — commercial HPC, large home directories, and backup and archiving — against nine critical capabilities.

Introduction

The growth in quantity of unstructured data is clearly outpacing that of structured data. Enterprises and service providers are demanding scalable and resilient storage infrastructure at affordable costs to address big data challenges, and to build the foundation for cloud computing. Scale-out storage is rapidly emerging as a viable alternative to address a wide variety of enterprise use cases, because it allows companies to add storage on a pay-as-you-grow basis for incremental capacity and performance. Gartner believes that unrelenting cost pressures — due to the explosion of unstructured data, the need for providing differentiated services, and the availability of better functionalities and professional support — will drive the demand for scale-out file system storage in the near future.

This research provides an overview of the leading scale-out file system storage products, and evaluates their effectiveness in addressing organizational needs for a range of use cases.

Product Class Definition

Scale-out disk storage systems are built using modules or nodes connected together in a system. The term "scale out" does not strictly refer to a cluster file system, which can span multiple nodes for a single image. Rather, the term is used to include a loose cluster that is aggregated by a global namespace. Depending on the system's design, there can be one or more node types in the system — for example, interface and storage nodes. This modular approach enables users to start small and grow their system without high startup costs or serious concerns about outgrowing the system before the end of its planned service life.

Scale-out storage architectures exhibit the following characteristics:

• Capacity, performance, throughput and connectivity scale with the number of nodes in the system.
• Scalability often is limited by storage hardware and networking architectural constraints, not by software designs.

Critical Capabilities Definition

Scale-out file system storage needs a multidimensional focus. It must be extensible, but must focus also on cost, aggregated bandwidth, input/output operations per second (IOPS), storage efficiency, data protection, compatibility, ease of management and durability — all on a massive scale. This research examines nine critical capabilities enterprises should consider when deploying scale-out file storage architectures. Enterprises can work toward these goals by evaluating scale-out file system storage platforms in all capability areas:

• Capacity: This refers to the ability of the platform to support growth in capacity in a nearly linear manner. It examines file system capacity scalability limitations in theory and in real-world practice, such as maximum capacity, the number of files, and the number of nodes and disk drives supported by a file system, volume or a namespace.
• Efficiency: This refers to the ability of the platform to support storage efficiency technologies, such as compression, deduplication, thin provisioning and automated tiering to reduce total cost of ownership.
• Interoperability: This refers to the ability of the platform to support third-party ISV applications, public cloud APIs and multivendor hypervisors.
• Manageability: This refers to the automation, management, monitoring and reporting tools and programs supported by the platform. These tools and programs can include single-pane management consoles, monitoring and reporting tools designed to help support personnel to seamlessly manage systems, monitor system usage and efficiencies, and anticipate and correct system alarms and fault conditions before or soon after they occur.
• Performance: This refers to the aggregated IOPS and bandwidth that can be delivered by the cluster functioning at maximum specifications, and observed in real-world configurations.
• Resiliency: This refers to the options and capabilities offered in the platform for provisioning a high level of system availability and uptime. Options offered can include high tolerance for simultaneous disk and/or node failures, fault isolation techniques, built-in protection against data corruption and other techniques (such as snapshots and replication) to meet customers' recovery point objectives (RPOs) and recovery time objectives (RTOs).
• Security: This refers to the native security features embedded in the platform that provide granular access control, allow customers to encrypt information, protect against malware and offer data immutability.
• Applicability for Infrequently Accessed Data: This is based on feature functions that are important for longtime archiving and backup, as well as how often the product is used for backup and archiving. It helps accentuate the file system storage design differences for different use cases.
• Applicability for Production Data: This is based on feature functions that are important for frequently accessed production data, as well as how extensively the product is used to store production data. It helps accentuate the file system storage design differences for different use cases.

Use Cases

This report evaluates the capabilities of scale-out file system storage vendors to participate in three different use cases (see Table 1):

• Commercial HPC: This is the most prominent use case most scale-out file system storage products are built to address. Commercial HPC environments are characterized by the need for high throughput and parallel read and write access to large volumes of data. Performance, capacity, applicability for production data, resiliency and manageability are the most important considerations in the choice of products, and are highly weighted.
• Large Home Directories: This is the classic network-attached storage (NAS) use case, but on a much larger scale. IT professionals keen to consolidate file server or NAS filer sprawl should consider using scale-out file system storage products that offer operational simplicity and nearly linear scalability. In environments characterized by file server sprawl, storage management is simplified by eliminating physical, client-to-server mappings through global name spaces, making it an ideal platform to perform tasks such as automated storage tiering and user transparent data migration. Resiliency, storage efficiency, applicability for production data and manageability are critical selection considerations, and are highly weighted.
• Backup and Archiving: File system storage has been used as a backup/archiving target for years; scale-out file system storage provides added scalability for large backup and archive datasets to meet increasing demands for disk-based backup and archiving. While embedded storage efficiency techniques are particularly important for backup targets, features such as integrated storage tiering and write once, read many (WORM; see Note 1) capabilities increase the attractiveness of scale-out file system storage as an archiving target in terms of simplicity, affordability and data immutability. Applicability for infrequently accessed data, capacity, storage efficiency and resiliency are important selection considerations, and are highly weighted.

Source: Gartner (January 2013)

Inclusion Criteria

The products covered in this research include scalable file system storage offerings with a sizeable footprint in the market. Panasas and Intel (Lustre) are not included in this evaluation, since both vendors focus only on the HPC use case.

The scalability of the file systems in the storage system is defined as:

• 100TB minimum per file system
• 1PB minimum per namespace, which can span more than two nodes

Other product inclusion criteria are:

• At least 10 production customers must have deployed 300TB-plus of storage based on the platform.
• Support must exist for horizontal scaling of drive capacity and throughput in a cluster mode, or in independent node additions with a global name space.
• The product must have been deployed for all three use cases.
• The vendor must provide one or more customer references.
• The product must be installed in at least two major global geographies.

Critical Capabilities Rating

Each product or service that meets our inclusion criteria has been evaluated on several critical capabilities (see Table 2 and Figure 1), on a scale from 1.0 (lowest ranking) to 5.0 (highest ranking).

Source: Gartner (January 2013)

Source: Gartner (January 2013)

To determine an overall score for each product in the use cases, the ratings in Table 2 are multiplied by the weightings shown in Table 1. These scores are shown in Table 3.

Source: Gartner (January 2013)

Product viability is distinct from the critical capability scores for each product. It is our assessment of the vendor's strategy, and of the vendor's ability to enhance and support a product throughout its expected life cycle; it is not an evaluation of the vendor as a whole. Four major areas are considered: strategy, support, execution and investment. Strategy includes how a vendor's strategy for a particular product fits in relation to the vendor's other product lines, market direction and overall business. Support includes the quality of technical and account support, as well as customer experiences with the product. Execution considers a vendor's structure and processes for sales, marketing, pricing and deal management. Investment considers the vendor's financial health, and the likelihood of the individual business unit responsible for a product to continue investing in it. Each product is rated on a five-point scale, from poor to outstanding, for each of the four areas, and is assigned an overall product viability rating.

Table 4 shows the product viability assessment.

Source: Gartner (January 2013)

The weighted capabilities scores for all use cases are displayed as components of the overall score.

Figure 2 shows the overall use case.

Source: Gartner (January 2013)

Figure 3 shows the commercial HPC use case.

Source: Gartner (January 2013)

Figure 4 shows the large home directories use case.

Source: Gartner (January 2013)

Figure 5 shows the backup and archiving use case.

Source: Gartner (January 2013)

Vendors

Dell Fluid File System (FS)

Dell's Fluid FS is based on the Exanet technology asset Dell acquired at the end of 2009. Fluid FS supports different Dell storage arrays in the back end, including PowerVault, EqualLogic and, more recently, Compellent. Many of the evaluated capabilities, such as capacity, performance and resiliency, vary depending on Fluid FS's back-end storage arrays. While the PowerVault solution aims at low-cost archive and backup use cases, the Compellent solution targets high-performance deployments. Fluid FS has a scale-out architecture based on high-availability pairs, and stripes metadata and data across all nodes in the cluster for performance. The largest production capacity deployed today is less than 1PB. Compared with its peers, Dell's Fluid FS often is deployed on a smaller scale because of Dell's strong presence in small and midsize enterprises. Fluid FS does not support deduplication/compression, multitenancy or WORM, and only supports Network File System (NFS) v.3 and Server Message Block (SMB) v.1 NAS protocols.

EMC Isilon

EMC acquired Isilon Systems at year-end 2010. Since then, EMC has grown the Isilon business from $200 million to $500 million, based on Gartner's estimates, expanding beyond Isilon's traditional turf, such as media/entertainment and life science, into enterprise data centers to tackle new workloads, such as large home directories and VMware virtual servers. EMC Isilon's NL-Series is increasingly being adopted for backup and archiving use cases. Isilon is based on a scale-out architecture with a full-stack file system that manages volumes, and provides built-in data protection, with up to four node parities to tolerate multiple node failures within a cluster. Isilon's latest OS refresh (OneFS v.7.0), launched in November 2012, added more enterprise-focused features, such as file-level writable clones, authentication zones, VMware vStorage APIs for Array Integration (VAAI) and VMware vStorage APIs for Storage Awareness (VASA) support, and enhanced WORM functionality. However, these new functions must be proved in the field. Among the products evaluated, Isilon stands out in capacity, performance, manageability and resiliency. However, it still lags behind in efficiency due to its lack of deduplication and compression, which are especially important for the backup use case.

Hitachi Data Systems (HDS) Hitachi NAS (HNAS) Platform

After a long-standing OEM relationship that started in 2006, Hitachi Data Systems acquired BlueArc in 3Q11 to capitalize on the growth in unstructured data disk storage. Hitachi NAS (HNAS) Platform is positioned primarily as a scalable, high-performance storage array for business-critical enterprise applications, and for vertical industries, such as life sciences, media and entertainment. HNAS uses hardware acceleration to improve parallel input/output (I/O), and recently launched performance accelerator software, an optional license key-based feature that can enhance overall performance. HNAS's namespace can span a maximum of eight nodes, and has competitive features for storage tiering. Through its use of enterprise virtual servers (EVSs), it can offer a resilient and secure environment that maximizes availability. Integration with Hitachi Content Platform (Hitachi's object storage) and native WORM features are extending the HNAS use case to include archiving environments. However, HNAS still has a smaller footprint relative to some competitors in the scale-out file system storage segment and currently lacks efficiency features, such as deduplication and compression. Deduplication is scheduled to be introduced in the first quarter of 2013.

HP StoreAll Storage

In 2009, HP acquired Ibrix and repackaged its parallel file system with ProLiant servers to form the X9000 series of scale-out storage systems, targeting high-throughput environments, such as HPC and Web-based archiving. In December 2012, HP introduced a new platform, HP StoreAll Storage, which leverages the Ibrix scale-out engine, and with new features, such as StoreAll Express Query, which can perform extremely fast search of massive content repositories. HP also has integrated Express Query with HP Autonomy Intelligent Data Operating Layer (IDOL) to deliver near real-time results on big data queries. HP StoreAll supports up to 16PB and over 1,024 nodes within a single namespace. Automated, policy-based data tiering is a standard feature, and manageability has been made easier by integrating with tools such as HP Systems Insight Manager (SIM) and Storage Essentials. The StoreAll series has native WORM features and broad ISV support, making it an attractive product for PB-scale archiving. HP also packages all hardware and software components in a single unified pricing scheme. However, the StoreAll series has modest efficiency features, relying on back-end storage arrays for thin provisioning, and lacking deduplication and compression. HP has a different deduplication backup appliance product called StoreOnce, which leverages Ibrix technology for scale-out deployment, but this report does not evaluate backup appliances.

IBM Scale Out Network Attached Storage (SONAS)

IBM introduced SONAS based on its General Parallel File System (GPFS) in early 2010, and thus far has had limited market success. GPFS was designed for HPC research environments, offering highly parallel storage throughput for large Linux compute clusters. SONAS is a turnkey software and hardware solution for the broader market beyond HPC. IBM has improved SONAS's manageability by using the highly visual graphical user interface (GUI) adopted from its XIV storage arrays. SONAS is highly rated in this report for its high-capacity scalability and high-throughput performance. It also offers some distinctive features, such as a cross-geography namespace for global collaboration and integrated tape tier support under the same namespace. However, its storage efficiency and resiliency depend on the different storage arrays used at the back end, and it lacks deduplication and compression.

NetApp Clustered Data Ontap

This report evaluates only NetApp Clustered Data Ontap v.8.x, which adds a global namespace, load-balancing capabilities and federated management to the feature set that has made its nonclustered file systems popular. Customer adoption of Clustered Data Ontap has been driven mostly by new deployments and end-of-life refreshes due to disruptive migration hurdles. NetApp Clustered Data Ontap can support up to 12 failover node pairs, which can scale up to 50PB. It enables user-transparent migration among different node pairs to perform load balancing, easing management complexities with high availability in a large environment. NetApp has been in a market-leading position in consolidating Windows file servers for home directories, and its Clustered Data Ontap brings its Common Internet File System (CIFS) support into a more scalable environment. Among the critical capabilities evaluated in this report, Clustered Data Ontap is highly rated for its storage efficiency, resiliency, performance and security, as well as interoperability. The newest Data Ontap, v.8.1.1, launched in June 2012, introduced a new feature called Infinite Volume, which provides a single mount of 20PB storage with up to 2 billion files. However, Infinite Volume is currently available for NFS v.3 only, and data deduplication occurs at the much smaller FlexVol level than the Infinite Volume level. Clustered Data Ontap does not currently support some noncluster standard features that are important for backup and archiving, such as SnapVault and WORM, and increases management complexity, compared with NetApp's nonclustered storage.

Nexenta NexentaStor

This research evaluates only the NexentaStor product with namespace cluster plug-in software. Nexenta was founded in 2004, and in 2008 launched its open-source product, NexentaStor, which is based on the open-source project illumos. Illumos is a follow-on to OpenSolaris, and contains ZFS, initially developed by Sun Microsystems. To compensate for the lack of a native parallel file system, NexentaStor has a namespace cluster plug-in software option that enables horizontal scaling and global namespace capabilities. NexentaStor has more than 4,000 commercial deployments, of which only a limited number of customers use the global namespace cluster plug-in. The product provides unified block and file storage, and supports unlimited snapshots, synchronous replication, thin provisioning and virtual machine integration (VMware, Citrix Systems and Microsoft), which make NexentaStor appealing to enterprise IT buyers. However, Nexenta lags behind in manageability and security. The company's support infrastructure needs to improve to meet customers' expanding global requirements.

Quantum StorNext

Quantum is an established maker of data protection and management products particularly known for its tape drives and libraries. In 2006, Quantum acquired Advanced Digital Information, through which it inherited StorNext, a shared SAN file system product. Quantum has enhanced StorNext over the years with the ability to handle bigger datasets and more IP network-centric workloads, and the ability to embed more flexible, automated storage tiering. The product has been built to address high-performance streaming of rich media, cross-OS file sharing and long-term archiving in industries such as life sciences, energy, media and entertainment, and government. In August 2012, Quantum released v.4.3 of the product with an enhanced MySQL database to ease the handling of billions of files and PBs of storage. StorNext is available as a software-only solution, and as an appliance with dedicated hardware for metadata controllers, NAS gateways and archival storage. The product has tight integration with tape and Quantum's object storage, and takes advantage of policy-based tiering to lower the total cost of ownership. However, StorNext lacks thin provisioning and snapshots. To become more general-purpose and to broaden the product's data center appeal, Quantum must expand its list of ISV partners beyond the niche vertical industries.

Red Hat Storage Server

Red Hat acquired Gluster, an open-source and privately held startup firm focused on scale-out storage, in 4Q11. Red Hat relaunched GlusterFS as Red Hat Storage Server, with preintegrated software consisting of Red Hat Enterprise Linux (RHEL), GlusterFS and the extensible file system (XFS). Red Hat Storage Server can be run on most industry-standard x86 servers with Ethernet and InfiniBand support on bare-metal hardware, or as an Amazon Machine Image in the public cloud. Red Hat Storage Server supports unified file and object storage at the software layer, but is still an evolving product. Enterprise IT buyers need to be aware that data protection features such as snapshots are not yet available, and that management, reporting and logging capabilities need to evolve further. However, Red Hat's acquisition allows the Gluster team to tap into a wider open-source community pool and engineering talent. Red Hat has unveiled an aggressive product road map for the next 12 to 18 months to address the limitations.