The Network is the Storage
To reap the benefits of SANs, you have to figure out how they fit in with your existing LAN, WAN, and enterprise infrastructures.
By Vance McCarthy
The era of storage-area networks (SANs) is starting to take shape. Work is now under way that will make SANs a viable option for just about any enterprise--whether centralized, LAN, or WAN.
The SAN concept, which is being fueled primarily by advances in Fibre Channel, accommodates high capacity and split-second access by marrying highly scalable hosts and storage servers with high-speed (and sometimes long-distance) networking. The ability to daisy chain a variety of host storage devices across the campus--or across the country--promises more scalable, more manageable, and more highly available data storage than ever before.
The basic benefits of a SAN approach to storage are so promising that more than 50 storage and networking companies late last year formed an industry association--the Storage Networking Industry Association (SNIA)--to tackle a variety of thorny issues surrounding SANs and network-attached storage (NAS).
The primary goal of SNIA is to nail down the technical details and develop specifications for a suite of de facto standards for SANs and NAS. In November, the SNIA mapped out a list of several "must have" technologies that are needed before SANs become widely adopted (see sidebar "SNIA Faces Challenges").
In addition to these longer-term goals, SNIA participants are focusing on how SAN-enabled technologies will be rolled out to the public. They want to make sure SANs will be suitable for the three major types of networks--LANs, enterprise networks, and WANs. (For more information about SNIA, go to www.sresearch.com/snia.)
SANs and LANs
For the most part, advocates of upgrading from SCSI to Fibre Channel have made the most progress in advancing their technologies to the gigabit-per-second SAN.
In a Fibre Channel SAN, a variety of hosts and storage subsystems are connected through a gigabit-speed hub (in an arbitrated loop configuration) or switch (in a fabric configuration). Via hub and switch networking technologies, optimally tuned for data storage and retrieval, large numbers of users can simultaneously access data stored on a network of storage devices. The goal is to break through SCSI`s inherent limitations in data delivery speeds and the number of users (initiators) that can simultaneously access any storage device.
This approach to SANs also allows users to continue to scale their networks with more storage capacity without reconfiguring their LANs or buying additional RAID servers (which offer storage bundled with sometimes unneeded CPUs).
Vendors such as Arcxel, Brocade Communications, Gadzoox Networks, Emulex, McData, and Vixel, along with other hub and switch vendors, have joined together under the aegis of the Fibre Channel Association (FCA) to promote future standards for this emerging high-speed storage technology.
One missing piece, according to many of these companies, is a way to manage all these connections. "Management is key to a high-speed SAN, especially when you`re providing fault tolerance and a high quality of service," says Bill Lozoff, director of marketing at Gadzoox. "Another key element is being able to readily grant access to a large number of users, and yet maintain data integrity through file locking and other version-control techniques."
Pending a formal standard for these software technologies, Gadzoox has licensed high-availability code from Veritas Software. Gadzoox` Ventana Management Architecture, now shipping with the company`s Gibraltar line of Fibre Channel hubs, provides simplified troubleshooting; status reports on hub configuration and port availability; and a graphical view of data throughput over the SAN.
SAN and the Enterprise
For some vendors of large storage devices, such as EMC, the advent of the SAN answers other questions. For EMC, the key advantage of SANs is to expand the capabilities of centralized data stores.
According to EMC, the distance limitations imposed by traditional SCSI methods of connecting servers and storage systems have posed significant constraints on large, fast-growing organizations that are trying to manage data centers and multi-site computer operations. Most current SCSI technologies limit attachment distances to 25 meters. Fibre Channel can support connections up to 500 meters, and work has begun on an extension that would support links up to 10,000 meters.
Such increases in distance make it easier for managers to continue to pursue strategies that centralize their data in mammoth data stores, but still have the flexibility to access data islands--or conduct data mirroring--away from the central store.
But as distances increase, so does the importance of throughput speeds. A recent EMC white paper puts it this way: "By extending channel performance by a factor of 5x (100MBps vs. 20MBps for standard SCSI), Fibre Channel will allow customers to further exploit the performance and functionality of enterprise storage systems for key applications such as OLTP, data warehousing, and Internet-enabled systems."
The key here is that with a high-capacity data store at the center, a SAN can extend the access of mission-critical corporate data well outside the walls of central headquarters, without having to replicate the data for each work group. In short, for companies that need to centralize their storage, SANs can extend the accessibility of data without increasing the time and expense required to manage multiple data stores.
For SAN management, EMC provides a proprietary management suite--including client agents--for its Fibre Channel-enabled Symmetrix Enterprise storage systems.
SANs and WANs
SANs also are attractive to vendors focused less on centralized storage and more on long-haul data delivery and mirroring. For example, Computer Network Technology (CNT) last fall unveiled a SAN strategy for its UltraNet family of storage systems. A new product family slated to be rolled out through the middle of this year (including UltraNet Storage Director, Storage Gateway, and Storage Multiplexer) will enable remote clustering and high-speed availability of mission-critical data across a WAN using any protocol or operating system.
In CNT`s vision, a SAN`s ability to provide high-speed access over multiple protocols and operating systems makes it an attractive way to better manage large data stores, such as those found at sites that rely heavily on System/390 mainframes.
CNT officials see a SAN as providing a high-speed network that connects directly between the storage devices and servers, forming a storage "subnet." One key benefit, they contend, is that a SAN can scale to multi-gigabit performance, a level unmatched by traditional server-attached storage.
"For example, we have five 30GB servers running our databases, and we have to back those things up," says Jim Morin, CNT`s director of corporate planning and market development. "The traditional way would be to hook up tape backup drives to each server. But that would require lots of tape handling, and probably even some swapping of tapes."
A SAN provides a better alternative, Morin says. "Users can centralize a tape library and put that on a SAN so all servers can get to it, and in the bargain have their own dedicated storage subnet, which reduces the expense and complexity of backup."
Wide-area technologies make SANs an even better business case, Morin adds. "With data compression and ATM, we can set up high-speed WAN connections for backup to remote servers across the country or even overseas. And, there`s the added benefit that the available bandwidth can be shared with other applications." Morin envisions companies using other high-speed WAN connections, such as DS-3 or SONET, to manage the backup and/or mirroring applications to be conducted at night or when business traffic is low.
CNT plans to complement its planned rollout of SAN-ready UltraNet products with its Channelink line of mainframe connectivity and software products. Channelink provides WAN support for channel-to-channel traffic, as well as load leveling, pipelining, alternate path routing, and data compression.
SANs Not Yet Sans Confusion
A variety of vendors are jockeying to develop and market (not necessarily in that order) SAN solutions that will bring higher speeds, higher capacities, and higher reliability to your network-attached storage.
The technologies may seem a bit intimidating: So many choices from so many vendors, none of which are totally standards-compliant. Within the next 12 months, the promise of SANs will begin to address the needs of many IT managers plagued with the problems of increasing amounts of data and dwindling resources to cope with the storage explosion. But in the near term, SANs will probably raise more questions than answers.
A storage-area network differs from server-based storage primarily in the externalization of storage onto the SAN as a shared resource attached to multiple servers.
What Is a SAN?
According to Strategic Research Corp., a market research and consulting firm in Santa Barbara, CA, a storage area network is "a high-speed network that establishes a direct connection between storage elements and servers." A SAN can be local or remote, shared or dedicated. Key elements of SANs include:
- Externalized storage
- Centralized storage
- Remote clustering
SANs enable storage to be externalized from the server, allowing large storage devices to be shared among multiple host servers without negatively affecting the performance of hosts or the primary network. As such, the architecture separates network traffic onto LANs and data access onto SANs.
Although the concept has received a lot of recent attention, the architecture is not new. SANs have evolved from environments such as MVS mainframe DASD and the VAX/VMS clustered architecture. (The storage interconnect in any clustered system is essentially a SAN.) Outside of these environments, EMC pioneered the SAN concept in an architecture that is sometimes referred to as SAN-attached storage (SAS), although EMC generally doesn`t use the terms SAS or SAN.
Another form of the SAN architecture is network-attached storage (NAS). NAS arrays typically connect directly to networks via Ethernet and contain single-board servers with a proprietary operating system and file system. The NAS concept was pioneered by vendors such as Auspex and Network Appliance Corp. and is being adopted by a wide range of storage vendors.--DS
In the external disk-array market, the clear trend is toward storage-area network (SAN) and network-attached storage (NAS) configurations, and away from server-based arrays.
SNIA Faces Challenges
At the Storage Networking Industry Association?s (SNIA) first meeting last November, some 50 networking and storage vendors highlighted several key technologies needed for the advancement of SANs. The goal of the SNIA is to develop standardized technologies where appropriate. Here are some of the challenges they face:
-EFile systems?A powerful yet flexible file system is required to distribute data stores in an intelligent and manageable way. Techniques for mirroring and load balancing may also come out of what directions are taken here.
-EMulti-initiator clusters?While Fibre Channel increases the number of initiators now possible with SCSI, questions remain as to whether multi-initiators can be handled by technologies other than Fibre Channel. This could also determine how OopenO these approaches will become.
-EStorage networking routers?How do vendors address and map data stored across multi-vendor environments or across a mix of mainframe, SCSI and Fibre Channel devices? This issue also looks at SANs across the wide area, in terms of supporting multiple multiplexers, extenders, hubs, switches, routers, and gateways.
-EStorage network management?Methods of applying traffic management, asset management, and central resource management are required to enable load balancing for baseline and peak traffic. This also requires mapping these techniques to existing proprietary APIs for resource management.
-EConcurrent/direct backup?Techniques are required to create backups of data on the fly, while an application is being updated. This approach will likely need to encompass various operating systems, file systems, and database requirements that have emerged from work on clustering, disaster recovery, and remote data access.
Vance McCarthy is a freelance writer, specializing in networking issues, in Foster City, CA.