By Dave Simpson
The drumroll for IP storage area networks (SANs) has been long and loud, but end-user adopters have been few and far between. But that may be changing. Last month, Carlson Companies rolled out what may be the first data- center-class IP SAN at the company's Minneapolis headquarters.
Carlson is a $6.8 billion travel, hospitality, and marketing company with consumer brands such as Radisson Hotels & Resorts, Seven Seas Cruises, T.G.I. Friday's, and Wagonlit Travel. The company has approximately 188,000 employees.
Previously, Carlson relied on a mix of direct-attached storage (EMC Symmetrix arrays attached to HP-UX servers) and network-attached storage (NAS). But rapid growth dictated a more scalable, and manageable, architecture.
Carlson's IT organization first concluded that it required a SAN. "We determined we needed to go to a storage network-whether it was Fibre Channel or IP-for the advantages of a shared storage environment and to provide future scalability," explains Mark Price, Carlson's director of IT.
The company's Shared Services group, which provides IT services for all of Carlson's companies, evaluated Fibre Channel and IP options for building the SAN. "One of the biggest issues we wanted to address from a corporate point of view was to provide storage services to our remote offices," says Price, "and one of their biggest problems was backup and recovery. We wanted to expand outside our data center, globally, for all our remote offices and data centers."
The main reason for not going with the more traditional approach-Fibre Channel-was simple, according to Gary Johnson, an architectural consultant with Carlson: "We hadn't invested anything in Fibre Channel networking capabilities-either in skill sets, expertise, or infrastructure-and we already had TCP/IP infrastructure and the requisite expertise." Johnson says that one of the key advantages of an IP SAN is the ability to use the expertise and technology they already had in place. In contrast, "Fibre Channel would have required specialized support and would have necessitated managing two separate infrastructures," Johnson explains.
Carlson's IT group determined that the initial costs of going with Fibre Channel would have been about the same as with IP, but that the IP approach would save money over the long run. "Another reason we went with IP was cost avoidance, in the areas of personnel, testing and monitoring equipment, management, etc.," says Price.
On the IP SAN, Carlson currently has HP-UX servers and Solaris servers connected to a Hewlett-Packard XP512 disk array (which replaced the direct-attached EMC Symmetrix arrays). In the next phase, the IP SAN will include 36 HP-UX servers and 12 Solaris servers (see figure). The HP-UX servers are equipped with Agilent's Fibre Channel host bus adapters (HBAs), and the Sun servers have Fibre Channel HBAs from JNI. Eventually, Carlson plans to add Windows NT servers to the IP SAN for applications such as Exchange and SQL Server.
Multi-protocol storage switches from Nishan Systems sit between the servers and redundant Cisco 6509 Gigabit Ethernet switches, and between the Cisco switches and the Fibre Channel-based HP XP512 array. Nishan's switches use the iFCP protocol for translation between Fibre Channel and IP. (In addition to Fibre Channel and iFCP, the Nishan switches can be configured for other IP storage protocols such as iSCSI.) Carlson configured the switches with 16 ports per device. List price for each Nishan 4300 switch is approximately $40,000. The Nishan storage switches function as edge switches for Fibre Channel storage and hosts and enable the integration of storage into the IP infrastructure.
Carlson's management framework is fairly straightforward: The company uses HP OpenView (which they were already using on their Ethernet data network) and tools from Concord to manage the network portion of the IP SAN. It also uses Hewlett-Packard's management tools for the disk array.
Johnson says that Carlson is getting wire-speed performance over the Gigabit Ethernet network. Compared to the previous direct-attached configuration, Carlson has realized a 56% decrease in processing time on its 10TB+ Oracle database applications (although it is unclear how much of that performance improvement is attributable to the IP SAN vs. direct-attached storage and how much is attributable to the change in disk arrays).
Wire-speed Gigabit Ethernet is more than sufficient for Carlson's transaction-intensive database applications, but Johnson says that performance isn't the real issue. "The biggest advantage with IP storage is not performance, but the ability to manage and allocate bandwidth."
Johnson adds that in Carlson's IP SAN, packet loss is handled in part by the Nishan switches and in part by the way the company designed the network. "We're not sharing the IP storage network with any other traffic," he says. "It's purely for storage."
Security is critical
Another advantage of the IP SAN approach is the ability to use familiar IP management tools. In Carlson's case, using existing IP security tools is especially important. "The ability to encrypt packet streams is a big issue for our company because we handle a large amount of transactions for other companies, including credit card companies, etc.," explains Steve Brown, senior vice president and CIO at Carlson, "so being able to encrypt those data streams to make sure that nobody can break in and grab information is extremely important." Carlson uses the same encryption tools in its IP SAN as it does on its primary Ethernet data network.
In the future, Carlson plans to centralize backup and recovery of remote sites at its central data center in Minneapolis. The company also plans to do hot backups between data centers.
So far, the company's IT executives report no problems running block-level storage traffic over IP. "But we see a lot of advantages, especially in the area of scalability," says Brown. "With IP, we're segregating out the servers and storage logically into separate networks, so that gives us unlimited scalability. The only limitation would be bandwidth."
Carlson uses SNIA Shared Storage Model
In planning its IP SAN and explaining it to upper management, Carlson Companies took advantage of the Storage Networking Industry Association's Shared Storage Model. "We used the SNIA model mostly as a tool to think through the different scenarios, and we mapped our architecture to it," says Gary Johnson, an architectural consultant with Carlson.
The Shared Storage Model can be beneficial as a design framework, particularly at companies that want to mix various storage architectures. At Carlson, for example, network-attached storage (NAS) is used with an IP storage area network (SAN), which replaced a direct-attached storage architecture. As shown in the diagram, Carlson's business applications include database applications, e-mail, Web, and other corporate production applications.
The SNIA Shared Storage Model presents a layered framework for demarcating applications, file systems and records, block aggregation onto disk or tape, and connectivity to storage targets. "The SNIA model was very useful for clarifying the position of our NAS and IP SAN-based storage," says Johnson, "and it provided a coherent framework for depicting our different options for remote backup solutions as well."
For more information on the SNIA Shared Storage Model, visit www.snia.org.
Carlson used the SNIA Shared Storage Model to map its enterprise storage architecture, which includes an IP SAN and network-attached storage.