The growth of the Internet and the increasing demand for digital information have caused storage needs to expand dramatically. Yet, the digital economy's demand for 24 x 7 access to information leaves little time even for routine maintenance—let alone disruptive and time-consuming processes such as adding storage.
The need for an efficient way to manage storage has spurred interest in both Network Attached Storage (NAS) and Storage Area Network (SAN) solutions. Collectively referred to as storage networking, both technologies promise to ease the management burden by consolidating server storage and letting administrators manage one homogeneous storage pool from a central location.
NAS and SAN
NAS systems are essentially specialized file servers whose OSs are customized for speedy file service. NAS systems connect to a LAN and provide data access at the file level. Many NAS systems support both the Common Internet File System (CIFS) and NFS protocols so that both Windows and UNIX clients can access the same files. NAS systems range from basic 40GB workgroup models, such as Quantum's Snap Server 1000 ($499 direct), to large, full-featured enterprise-level systems such as Network Appliance's NetApp F880c, which offers a capacity of as much as 12TB, two clustered dual-processor systems, Fibre Channel—based RAID arrays, and redundant hot-plug power supplies (F880c pricing starts at about $200,000). NAS systems are available from many vendors, including Compaq, EMC, Dell, Hewlett-Packard (HP), IBM, Procom Technology, Raidtec, RAIDZONE, and LSI Logic.
SAN storage is designed for applications such as database-management systems and online transaction processing (OLTP) programs, which demand high-volume, block-level disk access rather than file-level access. SANs consolidate data into shared storage arrays, removing the storage burden from your servers and linking the arrays, servers, and your backup tape libraries into a back-end network. SANs typically support a variety of OSs, so you can use UNIX servers along with Windows systems. However, each OS can access only those storage resources assigned to it. SANs typically are Fibre Channel based, and a Fibre Channel switching fabric or hubs tie the devices together. Consequently, you need to install Fibre Channel host bus adapters (HBAs) on each server on the SAN.
Like NAS systems, SANs are available from many vendors, including Nishan Systems, Compaq, Hitachi, Dell, IBM, EMC, XIOtech, HP, Brocade Communications Systems, Gadzoox Networks, McDATA, QLogic, Vixel, and SAN Valley Systems. SAN packages typically include Fibre Channel HBAs, Fibre Channel switches or hubs, SAN management software, and sometimes a bridge to interface SCSI devices to the Fibre Channel network. Some components, such as the HBAs, switches, or hubs, might be from third-party vendors, but the SAN vendor usually provides a single point of support.
Small SANs—those that support as many as 20 servers—might have a Fibre Channel—arbitrated loop hub and one storage array with a capacity of 1TB or 2TB. Midsized SANs might support as many as 40 servers, one or two Fibre Channel switches, and a storage array with a capacity of roughly 4TB to 7TB. Very large SANs could have several Fibre Channel directors (i.e., switches with high-availability features and many ports) and storage arrays and support several hundred servers. SANs are extremely expensive—a small one can cost a few hundred thousand dollars, and large ones can cost a million dollars or more.
Although both NAS and SAN solutions provide efficient storage management, NAS systems are easier to install and configure and generally are less expensive than SANs. Many enterprise-class NAS products provide better performance, scalability, and reliability than general-purpose file servers.
SANs provide outstanding scalability—you can easily add storage to arrays or attach new arrays to SAN hubs or switches. In addition, SANs move storage traffic off the LAN, improving application performance. This performance improvement is especially evident during a SAN's characteristic LAN-free backup, in which data moves directly from the storage arrays to the SAN-connected tape library without tying up LAN resources, as Figure 1 shows. SANs also offer outstanding reliability because you can construct the network with redundant paths, as Figure 2 shows, and can mirror data to several arrays to ensure data access should a disk or switch failure occur.
As I implied earlier, however, SANs are difficult to assemble and configure. A lack of interoperability between third-party Fibre Channel switches makes creating a switching fabric that uses products from multiple vendors difficult or impossible. Interoperability is also a problem between Fibre Channel HBAs and network-management frameworks that provide SAN management capabilities—a situation that limits your HBA choices to those that the software supports.
Interoperability concerns are probably the most significant hurdles facing the SAN market; however, that picture is slowly changing. Vendors are beginning to hammer out and support interoperability standards and certify their SAN management software on other vendors' hardware. The ANSI Fibre Channel Switch Fabric-2 standard defines a common Fibre Channel switch interoperability mode, which several vendors now support. In addition, the Storage Networking Industry Association (SNIA) recently developed a Fibre Channel HBA API, which lets network-management applications manage any compliant HBAs, and an Extended Copy command designed to let SAN storage-management applications initiate a LAN-free backup with other vendors' storage arrays. (For more information about SNIA initiatives, visit http://www.snia.org.)
Some vendors certify their SAN storage-management applications on competitors' storage arrays. Such certifications ensure a basic operational level and management capability but don't ensure support for all functions. At press time, EMC and Compaq had gone the extra mile and agreed to cross-license their SAN APIs to allow more manageability between the two vendors' storage products.
Under its SANmark open device interoperability certification program, the Fibre Channel Industry Association (FCIA) develops test suites for Fibre Channel HBAs, switches, and end nodes (e.g., storage arrays, tape libraries, disk subsystems, Fibre Channel—bridge devices). The tests ensure that the products within each category can communicate with one another according to the ANSI Fibre Channel protocol standards. Currently, these tests don't test the interoperability of storage-management applications with hardware; however, the FCIA plans to expand the program in 2002 to include application/hardware interoperability. For more information about the SANmark program, see http://www.fibrechannel.com/sanmark/index.master.html.
The Convergence of NAS and SAN
Although NAS and SAN differed clearly in the past, new NAS products from Compaq and EMC are beginning to blur the lines between these two complementary technologies. Compaq's recently released StorageWorks NAS Executor E7000, for example, can leverage the storage resources in Compaq's StorageWorks SANs. The Windows-based E7000 comes with both Fibre Channel and Ethernet network interfaces. When NAS products share the SAN's storage resources, administrators can use the SAN's virtualization software to assign a portion of the storage pool to the NAS device (I discuss storage virtualization later). Thus, StorageWorks users can easily manage NAS and SAN storage, and the E7000 can benefit from the SAN's speed and scalability. This ability to use the SAN's storage arrays also lowers the E7000's cost (pricing starts at $47,000 for a two-processor configuration with 2GB of RAM) for companies that elect to use both Compaq's NAS and SAN products.
EMC has taken a similar approach with its enterprise-level EMC Celerra File Server. Like Compaq's E7000, the EMC Celerra separates processing and storage functions, leveraging the storage in EMC's SAN-based Symmetrix family of storage arrays. One of EMC Celerra's strengths is its high availability—in its most basic form, the product comes as a two-node clustered solution with two rack-mounted file servers and a control station that manages the servers. A Symmetrix storage array can cluster as many as 14 file servers sharing as much as 28TB of storage. But you don't need to invest in a SAN to use the EMC Celerra—EMC also sells Symmetrix and EMC Celerra together in the same rack-mount cabinet.
EMC also produces the EMC CLARiiON IP4700, which the vendor describes as a midrange NAS solution. The IP4700, which starts at $75,000, is an integrated NAS device with room for approximately 7TB of internal storage. Should you decide to implement a SAN later, you can convert the IP4700 into an EMC CLARiiON FC4700 SAN storage array by installing another version of the IP4700's proprietary OS and replacing the Ethernet interface with a Fibre Channel HBA.
Storage virtualization represents the physical disk space from multiple arrays as one storage pool, thus letting you centrally manage data and share it over a network of heterogeneous servers. Virtualization lets you manage storage resources without requiring that you know the physical disk configuration. In addition to letting you assign portions of the storage resource to individual servers, virtualization lets you easily maintain backup schedules, reallocate storage among the servers, and quickly add new storage without bringing down the storage array and without concern about disk sizes, types, or configuration.
Both NAS and SAN vendors offer virtualization solutions. For example, Network Appliance claims that its NetApp DataFabric Manager can support hundreds of its NAS appliances. SAN vendors have offered host- and array-based virtualization products for some time. Host-based products provide a unified view of network storage but must be installed on each server. Array-based packages give each server a logical view of the storage assigned to it, but you usually must manage each storage array separately. One exception is XIOtech's REDI SAN Links array-based virtualization software, which presents a logical view of all SAN storage for the vendor's MAGNITUDE storage arrays.
The need for a unified solution that can scale up as you add new servers and storage arrays has focused attention on network-based virtualization solutions. HP recently introduced its HP Surestore SANLink appliance, which sits on the storage network between your servers and storage arrays and provides all servers a logical view of the entire network's storage resources. (SANLink sells for approximately $225,000.) HP has tested the product with storage arrays from Hitachi, Dell, LSI Logic, and EMC. Compaq also plans to offer a network-based virtualization product by midyear.
Different vendors define virtualization differently, so you'll need to ask questions and compare offerings before settling on a NAS or SAN provider. For instance, although virtualization products might let you assign storage to each server from one logical pool, can you move data from one array to another within the virtual pool without disrupting server access to the data? And does the virtualization solution truly mask the configuration complexities of the storage arrays?
The Future of Storage
Direct-attached storage is inexpensive in the short term but is difficult to manage effectively when you're adding servers and you have a small staff. A combination of NAS and SAN products makes sense for many organizations, especially if the NAS appliance can leverage the SAN's storage resources.
Throughout the past year, vendors have introduced 2Gbps Fibre Channel HBAs, switches, and storage arrays at prices only slightly higher than current 1Gbps SAN equipment. These high-speed Fibre Channel SANs are particularly well suited for use with performance-critical applications and applications that require assured delivery over distances of up to 10 kilometers. But creating an enterprisewide Fibre Channel SAN with thousands of servers might not yet be feasible. The cost of such a network would be extremely high, and creating the Fibre Channel switching fabric needed for such a large SAN would be technically difficult.
Even small Fibre Channel SANs require IT talent with specialized knowledge to configure and manage them and can be expensive to create. But with the cost of Gigabit Ethernet server NICs undercutting the cost of Fibre Channel HBAs and the precipitous drop in the price per port of Gigabit Ethernet switches, interest in IP-based storage networks is growing. The Internet Engineering Task Force's (IETF's) new iSCSI standard for carrying SCSI commands and storage data over IP networks is gaining broad industry support, setting the stage for the introduction of IP-based SAN storage solutions. Vendors such as IBM, Cisco Systems, and Nishan Systems have already introduced iSCSI-based products, and industry observers expect HP to jump in soon. Look for other vendors to follow.
In the short term, 2Gbps Fibre Channel technology will predominate, especially for applications that demand guaranteed packet delivery or the utmost in performance. Fibre Channel technology is very reliable, and most larger IT shops are likely to let iSCSI products shake out before adopting them. Early on, 1Gbps IP-based SANs will appeal to small organizations that have less-demanding applications and that need to rein in their storage-management workload but can't afford specialized IT talent, management tools, and Fibre Channel hardware. We'll also see mixed Fibre Channel—and IP-based storage-network environments.
Over time, however, IP-based storage networks will gain favor as they develop a track record. One reason for this expectation is that technology typically advances more quickly in the Ethernet world than it does in the Fibre Channel domain. For example, 10 Gigabit Ethernet products debuted this year, whereas Fibre Channel just recently made the leap from 1Gbps offerings to 2Gbps offerings.
One hurdle that NAS and IP-based SANs need to overcome is TCP/IP processing overhead. As IP storage networks are asked to carry increasing amounts of data, this overhead can significantly affect the performance of the servers or appliances that do the processing. Existing Gigabit Ethernet server NICs can't off-load this processing, but Alacritech has introduced the Alacritech 1000x1 Single-Port Server and Storage Accelerator, which off-loads TCP/IP processing from the host processor. Although Alacritech has set no retail price for the card because initial shipments of the product are going directly to system manufacturers, a company representative says that system manufacturers can acquire the card for less than the cost of a 1Gbps Fibre Channel HBA.
Making a Decision
When deciding on your future storage strategy, your applications' requirements and the goals you've defined for your storage initiative will determine the solution that best meets your needs. For example, if you'll be mining data from a relational database, a SAN's high-performance block-level access is an obvious choice, whereas a NAS solution is more appropriate for serving Web pages and for other file-service tasks. If high availability is a top priority, you'll need to select a product that's been tested with your clustering solution.
If you decide on a SAN, you'll want one that permits a network topology with redundant data paths. Although vendors are heeding customer demands to make SAN hardware and software more interoperable, assembling your own solution remains a difficult proposition. Even if you have the inhouse expertise, acquiring all SAN components from one provider is still the safest approach. But if you know you'll want to add third-party products to that SAN in the future, your best bet is a standards-compliant offering that's been certified with a wide range of products. And regardless of whether you select a NAS or a SAN solution, don't forget to investigate which antivirus and backup offerings vendors have certified, particularly if you're considering a product that uses a proprietary OS.
|CONTACT THE VENDORS|
ALACRITECH 1000X1 SINGLE-PORT SERVER
AND STORAGE ACCELERATOR
Alacritech * 408-287-9997 * http://www.alacritech.com
STORAGEWORKS NAS EXECUTOR E7000, STORAGEWORKS SAN
Compaq * 281-370-0670 * http://www.compaq.com
EMC CELERRA FILE SERVER, EMC CLARIION IP4700, EMC CLARIION FC4700, SYMMETRIX
EMC * 508-435-1000 * http://www.emc.com
HP SURESTORE SANLINK
Hewlett-Packard * 650-857-1501 * http://www.hp.com
NETAPP F880C, NETAPP DATAFABRIC MANAGER
Network Appliance * 408-822-6000 or 800-443-4537
SNAP SERVER 1000
Quantum * 408-944-4000 * http://www.snapappliances.com
REDI SAN LINKS, MAGNITUDE
XIOtech * 952-983-3000 * http://www.xiotech.com