iSCSI: Storage networking for the Internet generation

iSCSI is a natural successor to, and can be complementary with, Fibre Channel SANs and IP-based NAS.


After years of significant advancements in system processor and disk drive technologies, businesses are still faced with the dilemma of how to provide their end users with real-time data access. The explosive growth of the Internet has also challenged companies to make their data available 24x7 over global IP networks.

This article discusses how the prevalent storage interconnect technology (SCSI) and other storage networking technologies (e.g., storage area networks [SANs] and network-attached storage [NAS]) are evolving and merging with high-speed data networking to address this storage revolution for "anytime, anywhere" data access. Specifically, this article focuses on the emerging Internet SCSI (iSCSI) standard for accessing block-level data over IP networks and its movement into SAN and NAS markets.

SCSI evolution

Since the early 1980s, SCSI has been the dominant protocol for moving block-level data between servers and storage devices. The SCSI interface has evolved to keep pace with the demands of more-sophisticated and faster systems. The standard recognizes most storage peripheral types (e.g., disk and tape) and takes advantage of newer system hardware and more-intelligent controllers.

The data path has been widened and data-transfer speeds have been increased to keep pace with system requirements. This year, a new SCSI technology road map was adopted that will provide up to 640MBps performance as well as increased manageability and more features for high-end desktop connectivity.

Despite these advances in SCSI, its inherent limitations in distance, number of devices supported, and flexibility cannot keep up with the demands of distributed networked system environments. A key issue has been the sharing of storage devices between multiple systems.

SANs and Fibre Channel

To address the problem of data sharing, the market adopted the concept of a SAN. Fibre Channel is currently the de-facto standard for SANs due to its high performance, connectivity, and ability to support block-oriented storage protocols.

The Fibre Channel standard achieves its high performance by assigning much of the protocol processing to hardware. To overcome the scalability issues inherent in SCSI, Fibre Channel supports access to as many as 16 million SCSI devices in an extended SAN. Fibre Channel is widely deployed today in SANs running at a 1Gbps data rate, and 2Gbps products began shipping this year. With full-duplex transfer capability, 2Gbps Fibre Channel can theoretically support up to 200MBps throughput on a single link. Actual throughput has been measured at 195MBps, using less than 10% of the server's CPU.

NAS: Networked file access

Not all data is accessed as blocks. Many applications are designed to access data as files. To accommodate the need for shared file-level data access, the market adopted NAS. NAS devices allow users to attach scalable storage containing files directly to existing LAN (i.e., IP/Ethernet) network infrastructures, providing easy installation and maintenance. NAS devices support local file systems (e.g., NTFS and UFS), which can be remotely accessed by multiple NAS devices using file interchange protocols such as NFS (Unix) and CIFS (Windows). The file system accesses data that resides on disks either internal or external to the NAS device, attached via SCSI or Fibre Channel interconnects. NAS continues to grow in popularity as more businesses look to consolidate storage access over existing IP networks.

iSCSI: Storage access over IP

Although Fibre Channel SANs fill in many of the feature and functionality gaps left by SCSI, they leave others open. Fibre Channel does not inherently provide a way to access block-level data over an IP network. Also, SCSI-based storage devices in remote sites are left marooned, with no way to be accessed by systems or SANs that reside in a data center. Other emerging SAN IP standards such as iFCP and FCIP only address Fibre Channel SAN-to-Fibre Channel SAN interconnection over an IP network. The greater population of block-level data that resides on SCSI devices is not addressed in those implementations. Storage vendors are looking to iSCSI to remedy these issues.

Despite its growing popularity in IP-based infrastructures, NAS does not support direct access to the block-level data that resides on SCSI-attached devices or Fibre Channel SANs without first going through the file system. A number of NAS suppliers are looking to iSCSI to remedy this issue as well.

The requirement for a standard method to access the abundance of block-level storage data over IP-based networks brought about the development of the iSCSI protocol. iSCSI is an emerging standard that defines the encapsulation of SCSI packets in TCP for reliable transport and for routing using IP. The iSCSI protocol enables existing IP networks to move block-level data over LANs, WANs, or across the Internet, without changing the network infrastructure, the host software or operating system, or target storage devices (e.g., disk drives and tapes). The iSCSI standard is being developed by the Internet Engineering Task Force's (IETF-www.ietf.org) IP Storage Working Group (www.ietf.org/html.charter/ips-charter.html). Ratification of the standard is expected by year-end.

iSCSI enables standard SCSI commands to be passed between host systems (initiators) and storage devices (targets) over a TCP/IP network. To understand how iSCSI works, see the sidebar on p. 38. It is important to keep in mind that in iSCSI, the commands that are passed between systems are standard SCSI commands. Connectivity between systems is over a standard Ethernet/IP network infrastructure rather than SCSI cabling or a Fibre Channel network.

Software vs. hardware iSCSI

SCSI command encapsulation and data preparation can be accomplished through software on the host using standard TCP/IP and off-the-shelf Gigabit Ethernet network interface cards (NICs), but not without a performance penalty due to the high CPU utilization incurred by intensive TCP/IP protocol processing. iSCSI host bus adapters (HBAs) take the protocol processing off the host system and put it into firmware on a Gigabit Ethernet HBA. This implementation, sometimes referred to as a TCP/IP off-load engine (TOE), can substantially lower CPU utilization and free up system CPU cycles for application processing.

The figure illustrates where the iSCSI, TCP, and IP stacks reside on a Gigabit Ethernet HBA. An HBA implementation with a TCP/IP off-load engine can potentially achieve wire-rate speeds with less than 10% CPU utilization.

Host bus adapter implementations of iSCSI will include TCP/IP off-load engines and iSCSI agents.
Click here to enlarge image

Today, many iSCSI implementations are software-based, running over a Gigabit Ethernet NIC. However, some HBA suppliers are making available both initiator and target implementations running the iSCSI, TCP, and IP protocols in firmware on the HBA. The firmware is software upgradable, so there is no need to swap out the HBA as newer features are incorporated into the iSCSI specification.

Many vendors that supply target systems (filers, SCSI drive devices, storage routers, and switches) are incorporating iSCSI ports into their products. Today, the iSCSI target and initiator implementations support 1Gbps Ethernet, but next year, 10Gbps Gigabit Ethernet initiator and target implementations will come onto the market.


The emergence of iSCSI products in the current Internet-based market is extremely timely and, some would argue, not soon enough. With the prevalence of IP/ Ethernet infrastructures in practically every business's data center as well as in the majority of remote branch sites, iSCSI is a natural fit. The demand to access block-level data-whether it's on a direct-attached system, Fibre Channel SAN, or a NAS device-over the existing base of IP networks and the Internet is exploding. Applications such as remote backup and storage virtualization are taking advantage of ubiquitous IP networks to gain access to block-level storage devices. There is a need to provide low-cost SANs for isolated SCSI devices.

Remote branch sites with Ethernet LANs but no Fibre Channel SAN are ideal candidates for these environments. Another use of iSCSI is in remote branch sites that require access over an IP network to block-level data on a Fibre Channel SAN in a data center.

NAS vendors are jumping on the iSCSI bandwagon by incorporating iSCSI ports into their filers in order to exchange block-level data with devices on either an iSCSI or Fibre Channel SAN. End-to-end IP storage networking solutions are emerging from storage router and switch vendors that will use the iSCSI protocol to enable remote data access from hosts and storage systems no matter where they reside-on Fibre Channel SANs or Ethernet networks. Again, the collaboration of these vendors in forums such as the IETF and the SNIA IPS Forum will ensure these iSCSI LAN, SAN, and NAS solutions will interoperate.

As with any new technology, initial iSCSI market projections are conservative. Most industry analysts agree that iSCSI is a natural complement to SAN and NAS environments. Some even estimate that iSCSI will promote the growth of SAN and NAS environments by opening up easier access to storage and providing better support for mission-critical, storage-based applications such as backup and storage pooling. As the ratification of the iSCSI standard comes to closure, it is expected that from 2002 to 2003, the introduction and adoption of iSCSI solutions will accelerate and gain momentum throughout the storage networking industry.

Jane Shurtleff is director of marketing for Emulex's IP Storage Networking Group (www.emulex.com) in Costa Mesa, CA.

How iSCSI works

Click here to enlarge image

  1. An application or user issues a request for data, a file, or an application. The operating system generates the SCSI commands and data request. The command and request are encapsulated and a packet header is added. The packets are transmitted over an Ethernet connection.
  2. At the receiving end, the packet is disassembled or, if it was originally encrypted, the packet is decrypted, separating the SCSI commands and data.
  3. The SCSI commands and data are sent to the SCSI storage device. Similarly, data is returned in response to the request using the iSCSI protocol.

The iSCSI standard

The iSCSI protocol specification identifies the reliable transmission of the iSCSI protocol over IP networks, address and naming conventions, session management, error handling, and security. The Internet Engineering Task Force's IP Storage Working Group began drafting the iSCSI specification in late 1999. The specification is due to be completed by year-end, with products that fully support the Version 1 iSCSI specification available in the first half of 2002.

The early influx of iSCSI products does require interoperability among products from different vendors. The Storage Networking Industry Association (SNIA-www.snia.org) has formed the IP Storage Forum (www.snia.org/english/products_fs.html) to promote the iSCSI standard and interoperability testing.

In July, SNIA and the University of New Hampshire co-sponsored iSCSI conformance and interoperability testing among more than 25 vendors in the first iSCSI "plugfest." The intent was to do a check-point against the draft of the iSCSI specification to ensure it was enabling vendors to develop interoperable products. Current iSCSI hardware and software implementations are being developed to be upgradable to comply with the iSCSI standard once it is completed.


This article was originally published on November 01, 2001