Start-ups and market leaders are focusing on management to ease problems with scaling beyond a single NAS device.
By Lisa Coleman
Long known for its low cost, heterogeneous file sharing, and simple installation—among other benefits—network-attached storage (NAS) is difficult to manage when scaling beyond the first few appliances.
The problems are multifaceted. Generally, NAS boxes are deployed as "islands" of storage, where each island is managed individually. While administrators can see one view of all the devices, they cannot manage the devices as one unit; for example, replication, permissions, and storage allocations are difficult because each device must be managed individually.
"A lot of the innovation in NAS centers on management, because it's a huge problem," says Jamie Gruener, an analyst with The Yankee Group consulting firm.
According to Gruener, NAS management should entail device-level integration, including the ability to move files between heterogeneous filers and eventually from filers to archive devices. Stronger device-level integration would be a plus since most management features today are vendor-specific, says Gruener. "You have to be able to look at your entire environment, which you can't do today."
In an attempt to mitigate NAS management issues, some vendors have introduced distributed file systems that make many NAS devices appear as a single image. File switches are another recent innovation in NAS management. The switches serve as a central point for management and access (see below).
"Historically, when you add NAS devices to your environment your management system goes up linearly," says Randy Kerns, a partner with the Evaluator Group.
Another basic NAS management problem that still must be solved is data migration, according to Brad O'Neill, a senior analyst with The Taneja Group. Users want migration technologies that are heterogeneous, affordable, and highly available. When an environment is scaled up, it is a time-consuming process to migrate data to a new filer, partition it, and bring it online. Another function that many users need is a heterogeneous global namespace for aggregating files, enabling administrators to access, move, and re-concatenate directory trees regardless of their filer residency, according to O'Neill.
"NAS management is just starting to go through the same evolution we saw with SAN management in the late nineties," says O'Neill. "We need to move from vendor-specific solutions to a total infrastructure perspective."
"Users do not want to buy different software from different vendors to manage different hardware," says Bill North, research director of storage software at International Data Corp. (IDC). "They want a common interface."
NetApp opens up
NAS market leader Network Appliance recently opened up its proprietary DataOnTap APIs and is delivering them via the ManageOnTap API suite and a software developers kit (SDK).
The company claims that more than 40 vendors (e.g., AppIQ, BMC, Computer Associates, CreekPath Systems, Fujitsu Softek, NuView, Tek-Tools, and Veritas) are developing management code based on its APIs. The products range from anti-virus software to storage resource management (SRM) and file virtualization.
"We're breaking tradition by providing complete access to the APIs. Any ISV, customer, partner, or competitor has access to the APIs," says Narayan Venkat, senior manager of product marketing at Network Appliance.
The ManageOnTap API suite includes APIs for managing systems, storage, files and blocks, virtualization, and quota and resource management. It also supports SNMP, NDMP, and Win32. CIM interfaces will be provided in future versions of the SDK.
EMC—the other market leader in NAS—has been tackling NAS management for years and has also (selectively) shared its APIs. EMC has an array of products such as StorageScope, OnCourse, Workload Analyzer, and ControlCenter to help manage NAS environments that include its Celerra file servers as well as other vendors' NAS servers.
File virtualization on tap
One vendor with early access to Network Appliance's APIs is NuView. The two companies signed an OEM agreement last fall, and Network Appliance is reselling NuView's StorageX management software under the name Virtual File Manager.
The software allows storage administrators to create and manage logical relationships independent of the physical location of files residing on Windows-based storage systems. By creating a global namespace, administrators can benefit from better file-level management, movement, and replication.
Unocal, an energy resource and project development company, used NuView's software to prevent management problems during a major server consolidation project. The company consolidated storage from 100 to 40 servers, including Windows-based NAS filers from Dell attached to a SAN.
"Imagine an environment where you have 60 servers disappearing, and all the storage associated with them allocated into a NAS environment. That makes for a very complex NAS environment," says John Mowery, a principal consultant at Unocal.
One of the big hurdles Unocal had to overcome was making it easy for users to find their data after the consolidation. NuView's software allowed Mowery to create a hierarchical structure out of 200 shares, thereby creating a storage namespace that would be more logical for users to navigate.
"It's easier for users to find their data. Instead of users trying to decipher cryptic share names, we created a structure based on how the organization operates. We created references to their data that are pertinent to their business unit," Mowery explains.
The consolidation and data migration were transparent to Unocal's users. NuView's software also helped Mowery cut down migration time from four to six hours per share to 15 minutes per share.
Another company delivering file virtualization and aggregation is 1Vision Software. The company's vNAS software aggregates multiple NAS devices at the file system level. The software, which is designed for Windows-based NAS, merges file systems from multiple NAS devices into a single file system. Once aggregated, the NAS appliances appear as one large storage device on the LAN, according to Dave Howard, founder of 1Vision. When the software is installed on each NAS device, vNAS makes all unutilized space available to users.
Introduced last month, Z-Force's ZX-1000 file switch can be placed in front of Windows, Linux, Free BSD, or proprietary NAS devices to aggregate them into a large virtualized NAS array, allowing for incremental scalability and easier NAS management, according to company claims. The switch can attach to any CIFS-compatible device.
For the past four months, Sanborn—a geographic information systems company—has been using the ZX-1000 to front-end eight Dell PowerVault 715N NAS devices with about 3TB of capacity. Sanborn also uses a BlueArc SiliconServer with 6TB of active data. Overall, the company has several hundred terabytes of data and about 20TB of online storage for detailed mapping applications. Sanborn migrated from a tape-based system to a large-scale, centralized NAS environment with the Dell-BlueArc combination.
Stephen Rochford, director of IT at Sanborn, wanted to take advantage of his cheaper storage and yet still have performance equal to what he was achieving with the BlueArc system, which is what Z-Force promised him with its file switch technology.
The switch eases management because it virtualizes the Dell NAS devices into one, he explains. In contrast, with the BlueArc system his users see five different volumes and have their projects spread across them; therefore, the users must know which volume contains their data set. Rochford may also front-end his BlueArc NAS system with Z-Force's switch.
"The Z-Force technology gives you one mount point for your users. And, assuming it continues to prove itself on a performance and reliability basis, and the management interface gets better, I can start putting more and more storage behind it and I'll have only one place where users can access all that storage," says Rochford.
The ZX-1000 file switch stripes and mirrors data across different NAS devices in a server set. Striping files across multiple NAS devices improves file response time. The various NAS devices serve data in parallel to the file switch, according to Dan Cloer, director of product marketing at Z-Force. The switch creates a common data namespace for the different NAS devices.
"Management becomes a lot easier, because it's load-balanced and scalable," says Cloer.
The 3U, 19-inch rack-mount switch has two Gigabit Ethernet ports on the front and two on the back. The initial release supports CIFS. Z-Force has done interoperability testing with Samba for Unix clients and is developing support for NFS v3.0. The switch has a list price of $47,000.
Data migration blues
One common problem with NAS is migrating data from one filer to another when a new filer is brought online. How do you avoid downtime? For clustered solutions, this can be a problem because they do not scale well. For example, if there are 10 different filers with 10 different mount points, an IT administrator must go to each client and add all 10 mount points. While it can be laborious for the administrator, it is also a management problem for users who have to remember which drive has the requested data. In addition, recognizing a new mount means filers must be shut down.
Rainfinity is one vendor that is focusing on making data movement easier. It uses switching technology to address NAS management and data migration. The company recently introduced its 2U RainStorage NAS management appliance and has partnered with Network Appliance.
One of Rainfinity's first customers is Qualcomm's CDMA Technologies Division (QCT), which is using two pairs of RainStorage appliances in front of 24 Network Appliance filers that hold about 19TB. When the RainStorage appliance is plugged into the network it appears as a Layer 2 switch, enabling data movement without disrupting end users. Installation of the appliance requires establishing a separate Virtual LAN (VLAN). When data is moved, the ports associated with the appliance move into the VLAN and the appliance is in-band. The appliance moves out-of-band when the move is complete.
"The reason we got the RainStorage appliance was that downtime was not acceptable for our customers," says Keith Jolley, a staff engineer manager at QCT. Prior to installing the appliance, QCT would run out of space on a particular filer and its clients would suffer downtime.
The RainStorage system allows data to be moved between filers so users can continually update and access files. First, the appliance comes in-band and moves source and target storage appliances into the VLAN. It then begins a CIFS or NFS copy. Concurrent client/application I/O operations are synchronized while replicating data in the background. Data consistency is ensured by duplexing I/O while avoiding collisions between client access and data movement.
The RainStorage appliance, priced at $80,000, supports NFS and CIFS, has mirrored RAID-1 SCSI disks, and includes two Gigabit Ethernet interfaces for clustering.
No more head games
At the Salk Institute's Computational Neurobiology Lab, a center for brain research, performance is critical in processing computational neuroscience experiments. But management also plays a key role.
The lab uses a 64-processor Linux cluster supported by Spinnaker Networks' SpinServer 3300 NAS system. The storage is shared among 120 researchers on the Linux cluster. Clients include Linux, FreeBSD, Windows, MacOS, and OS9 platforms.
"All these platforms are hitting the same files, and our goal is to be able to mix Unix and Windows on the same file system," says Lee Campbell, a scientific systems analyst and bioinformatics specialist at the Salk Institute, in San Diego.
Windows and Unix have different permission lists, which creates problems when mapping is performed, says Campbell. "If a Unix user tries to create a file, what permission does it leave for the Windows user? We had problems with our previous servers, but the Spinnaker server solved them," says Campbell. The SpinServer software handles both CIFS and NFS and includes functionality for controlling Unix permissions when Windows creates a file.
Campbell also is able to improve management of file space growth. Each department within the lab has its own virtual file system, which can be easily changed because it is a group quota.
"Instead of having to go in there physically and grow file systems, you can just change a number. That makes management easier and more flexible because the virtual file system greatly simplifies the work of allocating file space to different groups," says Campbell.
Spinnaker's SpinFS global distributed file system allows all Spinnaker storage to be managed from a single console with a single management view.
SpinServers can be clustered together to form a single storage pool with non-disruptive file redistribution and system configuration. A global namespace is implemented across multiple servers and storage to eliminate islands of storage.
Spinnaker's non-disruptive data movement functionality is built into its file system, explains Jeff Tabor, senior product manager at Spinnaker. In contrast, older NAS file systems would require major operating system rewrites to achieve a "true" distributed architecture, according to Tabor.
NAS start-ups are devising ways to use storage more efficiently and simplify managing islands of storage. Several other NAS start-ups are expected to come out of stealth mode in the near future, with more options for addressing NAS management.
Recent NAS announcements
Last month, Storage Computer introduced a free Linux-based NAS kernel in a bid for a piece of the Linux-based NAS market. Its CyberNAS Red runs on RedHat Linux, supports NFS and CIFS, and offers volume and partition management. The core volume manager provides NAS file sharing and mounting of drives. In addition, a NAS selector feature allows multiple systems to be mounted. Account and IP domain configurations are also supported. For downloadable software, go to www.storage.com/nasred.
NSI Software is offering Double-Take for Windows Workgroup NAS Edition for data replication on Windows-based NAS devices.
The software provides continuous real-time data replication and automatic fail-over capabilities to provide high availability, enhanced backup, and disaster recovery. The software is priced at $2,495.
InoStor recently began shipping ValuNAS, a Serial ATA NAS appliance with the company's RAIDn technology, which protects data against multiple, simultaneous disk failures.
The ValuNAS 6000 model features up to 960GB of capacity, while the ValuNAS 9000 has up to 2.25TB. The 2U rack-mount devices support RAIDn as well as traditional RAID levels 0, 1, 4, 5, and 0+1. Both models include hot-swappable Serial ATA drives, Ethernet connections, redundant power supplies, snapshots, and support for third-party backup agents. ValuNAS runs on Linux-based iceNAS software. A 720GB version is priced from $5,999.
Snap Appliance recently introduced the Snap Server 4200 NAS system for workgroups. The filer runs on Snap's Linux-based GuardianOS and includes hot-swap disk drives, dual Gigabit Ethernet ports, and Active Directory support. In addition, the 4200 includes data-protection features such as embedded anti-virus software and server-to-server data replication. A 1U, 320GB configuration is priced at $3,195.
Moving toward Linux-based NAS
Although Windows-based NAS continues to gain market share (about 41% of total unit shipments), Linux remains a strong contender.
IBM recently dropped its low-end, Windows-based NAS 100 and NAS 200 devices. IBM will pursue a different platform—possibly Linux—for NAS, but not at the low-end. It will most likely be a midrange product, according to an industry source.
The low-end market is very competitive and low margin, says IBM spokesman Clint Roswell. "We found that most of the people who want NAS want even more capacity and functionality, so we're thinking about doing something in the future with NAS but not with Windows," he says.
IBM is not dumping its entire line of Windows-based NAS, however. Its NAS 300G gateway offers NAS-SAN connectivity with a NAS head front-ending a SAN.
Hewlett-Packard is also developing a Linux-based NAS system. The company is working with Cluster File Systems and Intel to develop a Linux-based clustered file system, called Lustre. The technology will allow scaling to thousands of nodes and is currently being tested at the Department of Energy. HP plans to introduce Lustre in its NAS 8000 servers early next year, with scalability up to 16 nodes.