Storage networks allow disparate storage resources to be centralized and organized within the data center. The available storage can then be centrally managed and allocated to users or applications. Fibre Channel (FC) has been the dominant storage area network (SAN) architecture for years, but its high cost and management complexity have deterred many SMBs from adopting this SAN technology.
Internet SCSI, or ISCSI, addresses these SAN problems by carrying SCSI commands and storage traffic over existing IP networks. Transporting "SCSI over IP" leverages ubiquitous Ethernet networks, and that translates to lower costs, easier maintenance and simpler management, which in turn makes SAN deployment more attractive to SMBs. This article examines the basics of iSCSI, enumerates deployment roadblocks and anticipates future trends.
Understanding iSCSI SANs
The development of iSCSI is a good example of problem-solving by using existing technologies in new ways. Traditional FC storage networks use the SCSI command set, carrying SCSI storage commands over physical FC infrastructures. As developers sought ways to overcome the cost, configuration and management complexities associated with FC networks, it made sense to place SCSI commands across other network architectures. Thus, iSCSI is SCSI block-storage access that is mapped to run across an IP network -- almost universally an Ethernet network. Although the iSCSI standard was ratified in 2003, iSCSI technology has only recently started receiving broad attention.
iSCSI SANs offers several benefits to an organization; cost, labor/management, and reach. The ubiquitous nature of Ethernet means that IP networks can be deployed quickly and easily in organizations of all sizes. Ethernet is also readily understood, so IT personnel can deploy and maintain an IP environment without specialized FC SAN training. It's not necessary to hire more IT personnel to implement and manage the SAN. Where FC SANs are typically small islands of technology located in a data center, the use of IP networks also gives iSCSI a global reach leading from a LAN to a WAN and onto the Internet -- allowing storage to be located almost anywhere.
Although the terms "iSCSI" and "storage over IP" are often used interchangeably, this is technically incorrect. While iSCSI may be the dominant subset of IP storage, there are other IP storage technologies to consider, such asFCIP (Fibre Channel over IP) for exchanging data between FC networks across the Internet and iFCP (Internet Fibre Channel protocol) for extending FC networks across the Internet.
Debunking common IP network myths
The low cost of iSCSI is usually its strongest selling point, but the issue of iSCSI cost is a bit more complicated than people think. People buy iSCSI because they want a cheaper alternative to FC. "Users have $2,000 Windows servers that they'd love to put on a SAN, but they can't afford to spend $2,000 more to connect them to a Fibre Channel SAN," says Stephen Foskett, director of strategy services at GlassHouse Technologies Inc. Leveraging an existing Ethernet network can indeed be far less expensive than deploying and interfacing to a new FC SAN, but businesses will ultimately spend capital on iSCSI storage arrays or specialized iSCSI adapters to enhance a server's connectivity.
The IP network itself is often a point of confusion. While iSCSI will certainly work over an existing Ethernet network using hardware, storage and software already on hand, it's vital to keep iSCSI traffic off the main production network. Not only can iSCSI traffic congest an everyday network, it's necessary to keep sensitive corporate data segmented from the general user population. "It is a SAN, and you don't want data essentially 'leaking out' of the data center," Foskett says.
Right now, the "sweet spot" for iSCSI is the midmarket. Where iSCSI is most readily embraced is in a Windows storage environment by organizations that have not previously implemented a SAN -- typically SMBs. However, iSCSI is also making inroads in the enterprise within the department or large workgroup where it can serve as primary storage to support Exchange or Oracle databases, and other block applications. Larger enterprises are even adopting iSCSI as secondary or even tertiary storage. "It's being used as an alternative to direct attached, Fibre Channel SAN attached, or even NAS [network attached storage]," says Greg Schulz, founder and senior analyst with the StorageIO Group.
Implementing iSCSI SANs
It takes three components to establish an iSCSI SAN: a network, a target and an initiator. The network component is relatively easy; virtually any IP network, such as Ethernet, will work. ISCSI is ideally suited to Gigabit Ethernet (GigE) and faster deployments, but it's crucial to keep the iSCSI SAN separate from the production LAN through a VLAN. Either that, or implement a different LAN segment and switch just for the iSCSI SAN. This separates storage and user traffic, and reins storage within the SAN.
The "target" is basically a storage destination. In the early days of iSCSI, routers were popular targets, allowing iSCSI hosts to access FC storage. Although this approach is still a viable transition technology, it has largely been abandoned, due to the availability of dedicated hardware and software products. Hardware iSCSI targets include dedicated storage arrays from vendors such as EMC Corp., EqualLogic Inc. and Hitachi Data Systems.
Today it is increasingly common to use a software-based iSCSI target running on a PC -- turning the server and its storage into an iSCSI target. Many open source products can be applied to Linux and BSD environments. Commercial software can also be employed, including iSCSI Storage Server for Windows from FalconStor Software Inc., SANmelody from DataCore Software Corp. and Microsoft's iSCSI target software, released as part of Windows Storage Server.
We should emphasize that there is no such thing as an iSCSI hard drive; any disk, such as SATA, in a server or array fitted with target software can be treated as "iSCSI storage." According to Schulz, "You can roll your own iSCSI storage, or you can redeploy a server to be an iSCSI array."
Each system that will access iSCSI storage will also need a hardware or software "initiator." Today, free iSCSI initiator software is already available for download with Windows 2000, Windows 2003 and Windows XP Pro. Software is the most popular initiator choice. "The majority, maybe 90% to 95% of all iSCSI implementations, involve a software initiator," Schulz says. Hardware initiators are also available as controller cards from many FC and TCP/IP offload engine (TOE) controller vendors, such as QLogic Corp. and Broadcom Corp.
Areas of concern
Most analysts and users agree that iSCSI is easier to deploy and manage than FC -- especially in the IP network and initiator. Some training and configuration may be needed to optimize iSCSI targets, but this depends on the target. A dedicated iSCSI array may demand more training and management than an iSCSI server deployed in-house. Ultimately, an iSCSI storage administrator will still need to be concerned with LUNs, volumes, provisioning and other common SAN issues. Pilot deployments can often help organizations identify potential problem areas and determine the tangible value of iSCSI.
Network architecture and performance is also critical to iSCSI. The technology is not tolerant of network interruptions, and an iSCSI array can easily crash as a result of interruptions or bottlenecks. Consequently, an iSCSI implementation should include an aluation of network performance and reliability. Old or outdated hardware components should be overhauled to improve network performance or ease bottlenecks, and failover measures should be deployed wherever possible to ensure network reliability. Typically, iSCSI is not recommended for use in large transactional environments where performance is critical.
ISCSI storage performance is also related to the number of spindles (drives) available in the network. you can achieve excellent iSCSI storage performance by load balancing more storage nodes (clustering) rather than placing more drives into fewer large iSCSI boxes. "Most high-end iSCSI arrays balance I/O across many more spindles than a Fibre Channel array," Foskett says. "If you look at the performance tests, you'll see that iSCSI outperforms Fibre Channel [due to the design of the storage array]."
The future of iSCSI SANs
ISCSI will have to find a way to flourish in the shadow of NAS, which is currently easier to deploy and manage -- particularly when coupled with virtualization techniques. But analysts say that iSCSI has a bright future in the SMB and enterprise, mainly because there is so much potential for growth. According to Foskett, about 80% of all servers (primarily low-end Windows servers) are not connected to shared storage, so iSCSI can expand shared storage dramatically by interconnecting that untapped wealth of storage.
The eventual move to 10 GigE and faster wireless networking standards will also impact iSCSI further in the future, enabling additional bandwidth for iSCSI and ultimately challenging the dominance of FC. "We're just scratching the tip of the iceberg with iSCSI," Schulz says. "[The year] 2008 is when I think iSCSI will really start to come into its prime." ***