NASA, we have cheap storage

'We're taking a little bit of a gamble,' admitted Jeff Seaton, the chief technology officer for NASA's Langley Research Center.The Hampton, Va., center is phasing in a storage area network built from off-the-shelf IP-based networking gear. This new type of storage network, called an IP SAN, can be less expensive than typical storage networks, but because the standard underpinning of IP SAN operations, known as Internet SCSI or iSCSI, has only recently been ratified, few organizations have truly road-tested the technology.But last summer, when Seaton started to think about storage for the center's new research repository, IP SAN seemed like the best fit. Top-speed performance was not crucial. Keeping costs low, on the other hand, was essential. So was easy scalability. These requirements played to the strengths of IP SANs.Because the technology uses off-the-shelf networking components, IP SANs are less expensive than other SANs and technical expertise is easy to find. Experts say if customers don't mind that their SAN doesn't perform as quickly as other products on the market, they can save as much as 80 percent in procurement costs. For NASA, it was a reasonable trade-off.Although IP SAN vendors have been in the market for a few years, analysts believe 2005 may be the year that IP SANs enjoy mass deployment. In February, EMC Corp. of Hopkinton, Mass., introduced iSCSI versions of its Clariion midrange SAN systems. In April, IBM Corp. of Ar-monk, N.Y., decided to resell iSCSI systems from Network Appliance Inc. of Sunnyvale, Calif. Dell Inc. now offers rebranded versions of EMC's arrays through its own sales channel. And MPC Computers LLC of Nampa, Idaho, has also jumped into the market.'Right now iSCSI is kind of a land grab. Everyone is getting into the market,' said Jay Masterson, a systems engineer at MPC.Beyond the household names, a host of up-and-coming storage companies are also creating IP SAN products. One such company is EqualLogic Inc. of Nashua, N.H., which supplied Langley with its iSCSI-based system.Langley needed a SAN to hold a repository of researcher documents.'Our goal is to build a knowledge base here at the center'a repository of research and project information that will continue to live on even after individual projects are completed,' Seaton said. Using the new storage system, Langley researchers could access material over the center's network.Initially, the team dedicated to the project 2T of storage from servers already running its content management program, Xerox DocuShare.NASA knew, however, that this form of direct-attached storage would not scale very well. Eventually, the repository would require more storage space, and, as a result, more hard drives to manage. Plus, Seaton wanted to back up the material, which would require tapes. Guarding against disk failure would also require setting up a RAID configuration, resulting in more administrative headaches.Storage area networks were originally designed to eliminate such problems, said John Joseph, vice president of marketing for EqualLogic. Instead of having one server hoard all the data, why not put it on an array of hard disks, which can be easily managed by storage software? Multiple servers can then access the array through a network switch.In the past few years, the Fibre Channel SANs have filled this role well. Seaton even looked at a few Fibre Channel SANs that fell into the upper range of what he was willing to spend.One drawback to going with a standard Fibre Channel SAN, however, was that NASA would need specialized administration skills to run one. Seaton did not have such expertise on tap. IP expertise, on the other hand, he had in spades.'Going with an IP-based storage approach was much more attractive than adding Fibre Channel to the mix,' Seaton said.Eventually Seaton's team purchased an EqualLogic PS200E iSCSI array, which offered 3.4T of raw storage.The research center had the storage network up and running by November 2004. With the help of EqualLogic, the basic installation took only about half a day, Seaton said. The team built a subnet, hooking up a Cisco Gigabit Ethernet switch to the content manager servers, which run Red Hat Advanced Server and Microsoft Windows 2003, both on Dell Power Edge servers.Seaton has been happy with performance thus far, though he admits that he hasn't really stressed the system yet. The center is slowly phasing in adoption, with about 15 percent of potential users currently tapping into the system. The team will slowly increase the number of users over the next two years, checking to ensure the IP SAN handles the load. Thus far, the new technology is holding its own.An IP SAN's appeal lies in how it uses many off-the-shelf networking components, lowering the cost of components and presenting a familiar environment to system administrators.The cornerstone of the IP SAN is the iSCSI standard, approved by the Internet Engineering Task Force in February 2003. Computers have long used SCSI as the interface for internal components, such as hard drives and floppy disks. The iSCSI standard encapsulates SCSI commands in TCP/IP packets so they can go out over an IP-based network, hence the name IP SAN.Later that year, Microsoft Corp. released iSCSI initiator software for Windows Server 2003, thereby allowing any application running on Windows 2003 to access SCSI disk drives across the network. Storage providers also began offering initiators for Unix networks.When it was first introduced, iSCSI was immediately cast as an alternative to Fibre Channel protocols. Fibre Channel requires specialized cabling, switches and adapter cards. In contrast, an IP SAN could run over standard adapter cards and Gigabit Ethernet cables and switches.In some cases, storage vendors further cut the costs of IP SANs by using commodity hard disks running over the new Serial Advanced Technology Attachment bus. Because commodity PC manufacturers have already begun gravitating toward SATA disks, SATA drive costs should fall far below the more specialized SCSI storage drives.Because IP SANs use commodity equipment, customers don't have to worry as much about vendor lock-in, Joseph said. Fibre Channel started out as an industry standard, but vendors tried to differentiate themselves through their products and built equipment that had trouble working with other brands, if it worked at all.By using industry standard hardware, IP SAN managers can better mix and match components. And iSCSI companies will compete on the strength of their arrays and supporting software. For instance, EqualLogic stresses how easy its management software is to use.IP SANs only offer about 80 percent of the performance of Fibre Channel SANs, according to David Dale, chair of the IP Storage Forum of the Storage Networking Industry Association. This is due to slower disk speeds, especially in cases where SATA is used, and the fact that Gigabit Ethernet offers about half the throughput of fiber-based Fibre Channel.Vendors, however, are finding that, in many cases, customers don't mind taking a performance hit if the technology cuts costs. Of course, how much a customer saves can vary widely depending on what components they purchase.Jay Krone, EMC's director of Clariion Platforms Marketing, said the company's all-SATA 100i platform is its cheapest Clariion offering at a little less than $6,000. The company also offers the 100i platform running Fibre Channel, but Krone said the IP version has been the bigger seller, mostly because of the lower cost of the accompanying switches, cabling and adapter cards.'The big win for people is that all the connectivity components in iSCSI are much cheaper, as much as 50 percent less than a moderately complex SAN,' Krone said.To decide if the performance lag of iSCSI would be acceptable in your IT environment, Dale said an organization should look at the type of traffic its data will generate. For a basic document retrieval system, such as NASA's, an IP SAN would be the best bet because the loss of a few microseconds per retrieval would barely be noticed. On the other hand, a database that courts multiple transactions every second might be better suited to a Fibre Channel, which can offer faster disks and higher throughput.'ISCSI fits very well in an environment where you have a bunch of servers that may have been running applications that weren't the most important in the organization,' Dale said. 'The cost of the Fibre Channel SAN was in- compatible with those kinds of systems.' For the NASA deployment in Langley, Seaton knew users would only be retrieving documents, a task that an IP SAN should handle with no perceptible lag.'Right now we don't have any indicator that performance will be an issue,' he said.