Simplifying storage: combining the iSCSI standard with SAN functionality

Computer Technology Review, May, 2004 by Tom Major

The rapid proliferation of data, along with the impact of data growth on IT managers, continues to garner significant mindshare. Today, information is at its genesis digital. Rich media, e-mail, CRM systems, and increasing regulatory requirements are contributing to data growth of 50% to 100% at many enterprises. The 24x7 nature of business today and the need for 100% data availability also fuel rapid data growth. Unfortunately, IT budgets are not growing in concert with data. In other words, IT professionals are being asked to do more with less.

Shared, networked storage is frequently touted as a way to address cost, growth, management and availability issues associated with storage by facilitating:

* Server and storage consolidation

* Increased utilization of available storage

* Increased performance

* More efficient backup and data protection processes

* High availability of data

Users have had Fibre Channel storage area networks (SANs) and network-attached storage (NAS) available to them for years--in theory, offering most or all of the benefits outlined above. So why is direct-attached storage (DAS) still the dominant storage architecture? With connection costs high (i.e., $1000 HBAs and $1000 per port switches) Fibre Channel SANs have proven too expensive for the masses. Additionally, Fibre Channel's complexity discourages many from implementing the technology. NAS has proven to be difficult to scale and addresses only file storage requirements. As we move into 2004, iSCSI is a new standard protocol being touted as the answer to the call for affordable, scalable, easy-to-manage storage. Is that realistic? Can a protocol do all this?

The History of a Standard

First, let's look back at the development of iSCSI. iSCSI was derived from SCSI, a protocol used for accessing and writing block-level data on locally attached disk drives. Access to block-level data is a fundamental requirement of file systems, databases, and many applications. In addition to being a protocol, SCSI is also a physical bus with distance limitations. As systems were architected that moved the physical storage outside of the computer, other protocols were developed that incorporated the SCSI protocol capabilities, but extended the physical connection distance. Fibre Channel is just such a protocol, as well as a physical network. Fibre Channel found a home in large IT shops, where the budget and expertise to install and manage this complex network was available.

It would seem natural that storage would be networked using similar network technologies as those used in the corporate computing network. The commodity nature of Ethernet networks made them affordable, and expertise around these technologies was almost ubiquitous. Unfortunately, speed (10 Mb/s) and lack of an appropriate protocol initially inhibited the use of IP/Ethernet networks for storage.

Over the ensuing years, Ethernet networks gained the speed (100 Mb/s and 1Gb/s) necessary to support storage traffic, leading vendors to continue the quest to network storage over Ethernet. NAS implementations made use of TCP/IP-based Ethernet networks by moving the file system to the storage device and using file-based protocols (CIFS and NFS). NAS delivered on the promise of easy to use storage at an affordable price. However, unlike Fibre Channel SANs, this architecture didn't support delivery of block data to the databases and applications (i.e., Microsoft SQL Server or Exchange) that required block-level access.

To address this inability to support many database and application networked storage requirements, LeftHand Networks developed a blockbased protocol for TCP/IP networks in 2001. This precursor to iSCSI, called Advanced Ethernet Block Storage (AEBS), sparked the delivery of the first, native IP-based SAN.

Simultaneously, vendors worked for several years through the Internet Engineering Task Force (IETF) to define a standard for a block-based storage protocol specifically designed for native TCP/IP networks. The standard, iSCSI, was ratified in June 2003.

Some offerings make use of iSCSI to connect IP-based servers into Fibre Channel SANs and call this an IP SAN. This article considers the use of iSCSI in constructing complete, or native, IP-based SANs.

The Benefits of a Standard

With iSCSI and TCP/IP-based storage networks, users are assured of the benefits that come with standards:

* Compatibility with a wide range of components: Ethernet has been a standard for many years, and components are truly plug-and-play. Users own and are familiar with a variety of brands of Ethernet switches and other IP-based networking equipment. IP-based storage works with standard network interface cards (NICs), eliminating the need for expensive host-based adapters (HBAs) required with Fibre Channel.

* A large base of management software and tools: Native IP-based storage allows IT managers to handle storage traffic in much the same manner in which they handle other network traffic. Existing software applications and tools for bandwidth provisioning, traffic management, security, and overall network management can be used with the storage. Adapting storage traffic to IP offers a unified view of all traffic and eliminates the need for separate management systems (i.e., storage SAN and corporate LAN).