SAN software, phase 1: Storage sharing
The approaches vary, but the goal`s the same: to provide heterogeneous sharing across storage area networks.
If you want to build a functional SAN--replete with heterogeneous file/resource sharing and data protection--you have a number of software options. Among early SAN adopters, storage integrators, and industry analysts, the vendors who are mentioned most frequently in this context include
DataDirect Networks, Mercury Computer Sys-tems, MountainGate, and Transoft Networks, which was acquired earlier this year by Hewlett-Packard.
In Macintosh environments, CharisMac Engineering (Newcastle, CA) is also a player, although its SAN sharing software does not currently support any Unix implementations, just Mac and Windows NT platforms.
Most of the other software vendors support NT, Macintosh, and one or two variations of Unix. Because most of these SAN software packages are currently deployed in video/imaging/entertain-ment applications, Unix support is often limited to Sun`s Solaris and/or Silicon Graphics` Irix. SAN sharing software usually runs on Fibre Channel storage networks, although most packages can also run over SCSI or IBM`s Serial Storage Architecture (SSA).
Using different methods, at divergent price points, all of these vendors provide the ability for heterogeneous platforms to share data at either the file, volume, or resource (subsystem) level, while providing mapping mechanisms for data protection and security--a key challenge in mixed environments, particularly those that include NT.
Windows NT is a rogue operating system in mixed environments because it tends to grab whatever disk resources it can, even if those resources are already "owned" by, say, one of the Unix platforms on the SAN. As a result, LUN (logical unit number) mapping--also referred to as masking, zoning, blocking, or partitioning--is necessary or some other technique for resource allocation.
All of the vendors provide this capability to varying degrees as part of bundled software packages. Transoft, however, offers a stand-alone product--called SAN Manager--that provides storage sharing (as opposed to true data sharing) with LUN-level management.
Introduced in March, the software allows administrators to manage a storage pool that is presented as a single logical unit (e.g., a single system image) and to allocate resources securely to different hosts without reboots, according to Doug Anderson, product manager at HP/Transoft, in Santa Barbara, CA. SAN nodes can only "see" storage that has been assigned exclusively to them and cannot access other storage LUNs assigned to other SAN nodes.
Host software-based LUN mapping can work in concert with, but also competes with, three other methods of protecting data in a heterogeneous SAN (see sidebar, "Four methods of LUN management"). In addition to Transoft, Dell offers host software-based LUN management, but Dell`s software works only with Dell hardware and Windows NT.
Hewlett-Packard plans to bundle SAN Manager with its high-end RAID arrays, although the software has been qualified with a wide variety of other vendors` RAID subsystems. However, on the host side, SAN Manager works only with NT and Solaris, although versions for HP-UX and AIX are in development.
SAN Manager is priced at $7,495 per SAN (up to 25GB) for an unlimited NT node license. A Solaris/NT license is priced at $9,995 per SAN. For additional capacities, licenses are priced on a per-GB basis; e.g., $2,495 for 100GB, $9,995 for one terabyte.
Transoft/HP also offers volume-level SAN sharing software in bundled hardware/software configurations called FibreNet, which is designed primarily for video/audio/imaging collaborative workgroup environments with multiple readers and one or more writers, as opposed to more general-purpose SAN applications.
The base product provides volume-level locking, and is based on the NT File System (NTFS). FibreNet supports NT, Irix, and Macintosh platforms, and competes primarily with software from Mercury Computer and to a lesser degree DataDirect and MountainGate.
Mercury`s SANergy is billed as a "SAN operating system," and provides file sharing and LUN partitioning. (Mercury may unbundle the LUN partitioning module later this year.) Like SAN Manager, SANergy software runs on each host platform connected to the SAN.
SANergy works with the native NT File System (NTFS), extending it to support multiplatform environments, including NT, Mac, Solaris, Irix, and AIX. Compaq`s True64 (formerly Digital Unix) and HP-UX versions are due next.
At the heart of SANergy is the NT-based Metadata Controller (MDC), which provides a central coordination point to handle metadata, transaction rollbacks, namespaces, etc. Because it potentially represents a single point of failure, the MDC has been criticized as a drawback to the SANergy approach.
However, Mercury plans to correct this perceived flaw in October with a hot-failover option, which allows another machine to take over in the event of an MDC failure, according to Barry Burke, vice president and general manager in Mercury`s Shared Storage Business Unit, in Chelmsford, MA.
Also due in the October time frame is support for Windows 2000 (not a trivial task given the modifications Microsoft has made to the NT file system in the new release), as well as a Solaris-based MDC that will use a Unix file system.
Like most other SAN software packages, SANergy has been deployed primarily in video, entertainment, pre-press, and Web hosting applications, although Mercury and the other vendors are attempting to broaden appeal into more mainstream SAN applications.
Roll your own
Unlike Mercury and Transoft, which rely largely on NTFS, DataDirect (Chatsworth, CA) and MountainGate (Reno, NV) have taken a ground-up approach to heterogeneous SAN sharing by creating their own native distributed file systems.
DataDirect`s Concurrent Data Net-working Architecture (CDNA) is in part a result of its acquisition almost two years ago of ImpactData, which had spent more than four years developing a file system.
CDNA was launched this spring and until recently has been limited only to DataDirect`s RAID arrays. However, the company`s game plan is to license CDNA to broaden its disk array compatibility. Last month, Storage Technology Corp. signed on as the first major OEM.
According to Bob Woolery, senior director of marketing at DataDirect (formerly Megadrive), a fully distributed file system includes a number of advantages, including:
Scalability. With a distributed file system, you can scale beyond workgroup environments into enterprise environments. In addition, a storage-oriented file system can form the building block for functions such as access privileges, storage management, and SAN-wide monitoring.
Error recovery. CDNA provides built-in error recovery and fault tolerance, down to the cable level.
Using a native file system eliminates the need for existing NT and Unix file systems, although CDNA supports a variety of host platforms, including NT, Solaris, Irix, and Linux. An HP-UX version is due next year.
As an enterprise-level distributed file system, the CDNA approach can be costly. Recognizing this, DataDirect provides a three-tier approach to SAN-based data sharing. At the small workgroup level, the company`s SANshare (which is based on CharisMac`s FibreShare code) provides sharing between Macintosh and NT platforms. For mid-range requirements, DataDirect resells Mercury`s SANergy.
MountainGate is another vendor that chose to create its own file system from the ground up. Available for about a year in its current version, MountainGate`s CentraVision File System (CVFS) is independent of specific RAID subsystems and protocols.
Like Data-Direct, Mountain-Gate preaches the necessity of a native file system. "In order to have true native access in heterogeneous environments, you need a native file system," contends Joe Moore, MountainGate`s manager of business development. In addition to the advantages cited by Data-Direct`s Woolery, Moore says the following:
"Implicit data conversion," which means that files are converted "on the fly" to standard file formats. The benefits here are cost savings, because only one version of the file needs to be stored on disk, and a sharp reduction in overhead.
True concurrency (e.g., a native file system allows multi-read/write capabilities to the same file at the same time).
CVFS currently supports only SGI Irix and NT. Next up are Macintosh, Solaris, and Linux.
Vendors such as DataDirect and MountainGate make no bones about the high cost of their native file system approaches. For example, CVFS licenses range from $2,750 per Macintosh or NT client, to $12,000 per seat for multiprocessor Irix environments.
Although relatively small vendors such as DataDirect and MountainGate have gone to great pains to write native file systems, one big vendor--Veritas Software--is also readying a distributed file system.
Veritas` entry in this market is due some time next year, and is expected to provide stiff competition to existing file systems, especially in more mainstream environments.
As usual, there is debate about definitions. For example, Mercury`s Burke argues that the defining characteristics of a distributed file system are:
No single centralized control of the file layout, metadata or physical drive mapping (e.g., striping, etc.).
Each system can directly open/read/write/ extend/truncate/close files, which usually requires some form of coordination with the other systems through techniques such as a distributed lock manager.
Each system directly manages user/ application access control and data security issues; again, with some form of coordination with the other systems.
A truly distributed file system is extremely complex because it needs to manage the coordination of security, access control, file and byte-range locking, file extents, free block lists, etc. And the complexity increases considerably as you add new operating systems to the mix.
As a result, these approaches are pricey, and may be applicable only in environments that need the utmost in scalability, fault tolerance, and heterogeneous platform support.
Because file systems and storage sharing are so critical to heterogeneous SANs, it`s no surprise that the Storage Networking Industry Association (www.snia.com) is working on standards in this area. The specifications are being developed by SNIA`s File System Working Group. The next formal meeting is in October.
"The problem with all the approaches today is that you`re relying on one vendor to solve the problem for you," says Tim Williams, founder and CEO at CrosStor Software (formerly Programmed Logic Corp.) He adds that another problem with existing solutions is that they only support a limited number of operating systems.
"The goal is to have heterogeneous sharing be as interoperable as NFS is today," says Williams. "We need a common standard, which is the SNIA`s goal."
Williams says that, eventually, support for the forthcoming standard protocols will come with the operating systems, just as NFS comes with all major operating systems.
CrosStor (South Plainfield, NJ) has proposed extensions to existing network-attached storage (NAS) protocols that will enable the protocols to operate over SANs. First up is CIFS/SMB, to be followed by extensions to NFS. At the same time, CrosStor is in the process of "SAN-enabling" its NAS software.
The market for SAN-based storage sharing and distributed file systems is still too small for industry analysts to track, but all observers predict healthy growth. Dataquest, a market research firm in San Jose, lumps this type of software into its "core" segment of the overall storage management software market, which includes file systems, volume management products, and emerging technologies such as SAN-oriented software.
Revenues for this segment of the software management market totaled $209 million in 1998, led by Veritas` File System and Volume Manager products, according to Carolyn DiCenzo, a principal analyst with Dataquest. DiCenzo predicts a compound annual growth rate of 21.8% for this market segment, reaching $561 million in 2003.
She expects the growth to be fueled in part by emerging SAN-based file systems such as DataDirect`s CDNA and MountainGate`s CVFS and the forthcoming distributed file system from Veritas.
In the CDNA architecture, data passes directly between clients and SAN-attached devices without passing through a file server. The Common Data Interface (CDI) is a middleware agent that allows Unix, NT and Mac platforms to access network data as if it were locally attached. The Network Data Director is an intelligent controller that manages traffic flow over the SAN.
At the heart of the SANergy architecture is a Windows NT-based Metadata Controller (MDC), which acts as a "server" for shared volumes. The MDC manages user authentication, access control and file system locking operations, with all I/O being directed over the SAN.
The SAN Manager drag-and-drop interface displays a single system image organized as two primary panes: nodes and storage devices. The node pane displays all SAN nodes and their LUN assignments. The storage pane shows the LUNs, organized into user-defined storage groups. A third pane can be opened to display Share Groups for clustered servers.
Four methods of LUN management
Avoiding data or file system corruption is a necessity in heterogeneous storage area networks because some operating systems--particularly Windows NT--attempt to grab all available storage, potential resulting in data loss or corruption. There are four main methods of avoiding this trap, most of which are based on LUN mapping, also referred to as masking, zoning, blocking or partitioning. Each can be used alone, or in conjunction with one or more of the other methods. The techniques vary, depending on where the management functionality resides.
Host-based file sharing software. This is the method used by HP/Transoft`s SAN Manager. LUN mapping software runs on all SAN-attached servers and clients. Because it`s priced on a per-platform basis, this approach can be expensive in large SANs.
Disk array controller firmware. Products such as EMC`s Volume Logix and MTI Technology`s Data Shield are implemented as firmware in those companies` RAID controllers. This approach provides a fine degree of granularity, allowing RAID systems to control which hosts have access to which LUNs, but drawbacks include high cost and vendor lock-in. (For more information, see InfoStor, March Special Report, p. 16.)
Switch zoning. Most Fibre Channel switches provide fabric zoning capabilities, which allow you to control device access through switch ports. This approach maps at the device (switch and subsystem) level, as opposed to the LUN level, providing a relatively course level of granularity.
Host-bus-adapter mapping. HBA vendors such as Emulex and JNI offer LUN management with their adapters. This is a relatively new approach that is done via HBA configuration software, allowing the adapter driver to determine which SCSI IDs and associated LUNs are presented to the host operating systems. Like host software-based LUN mapping, this approach requires new drivers for different operating systems.