Update On IP-based Storage.

Pick up any computer trade publication and you likely see articles discussing Storage Area Network (SAN) products and strategies. Briefly, SANs allow efficient and flexible connection of servers and storage by de-coupling nonvolatile storage (hard disks and disk arrays) from the servers used by LAN client machines for data storage services. To allow servers to subsequently re-access the data, a new network--the SAN--interconnects the storage devices to the servers.

It is generally agreed that SANs are good and one of the primary benefits of SANs is reduced operational costs through flexible storage and maintenance procedures. The question is how does one implement the reconnection of the storage to the servers following the de-coupling?

* What hardware architecture should be used?

* What communication protocol should be used?

When first available, Fibre Channel technology comprised a high-bandwidth (100MB/sec) solution well suited to SAN applications. So it's natural that most SANs today have been implemented using Fibre Channel to achieve interconnection between servers and storage repository units. Moreover, since disk storage arrays almost always use SCSI command protocol architecture, Fibre Channel standard groups necessarily developed a Fibre Channel SCSI encapsulation protocol to convey SCSI commands and responses between servers and storage repository devices over the Fibre Channel network.

IP-Based SAN Inevitability

As enterprises began deploying Fibre Channel-based SANs, it soon became clear that an individual SAN usually comprised an autonomous "SAN island" in which storage (and data) could be shared among the servers in the SAN, but not outside the SAN. These autonomous islands would collectively defeat the ultimate goal of accessing any data from any authorized server as well as diminish many data administration efficiencies. To unify the islands into single enterprise-wide data repositories, enterprises are looking to their existing Ethernet LANs and TCP/IP MAN/WAN infrastructures. As this occurs, the previously distinct lines between LANs and SANs will blur as IP networking begins straddling both LANs and SANs as a common denominator.

However, since the onset of interest in merged storage and networking, IP-based SANs have become inevitable. IP-based SANs allow direct communication between servers and IP-based storage devices using existing LANs, or even the Internet as the transport mechanism for, say, remote backup and archival storage. And since the leading technology to transport IP communication is Ethernet, Ethernet-based IP SANs are inevitable.

Presently, it's a fact that Fibre Channel will likely deliver superior performance today over Ethernet in most SAN applications. However, the Ethernet roadmap is well defined through 100-gigabits and hundreds of millions of dollars are being invested annually in order to bring interoper-able ten-Gigabit Ethernet technology and products to market. This will translate into more ten-gigabit Ethernet products from more suppliers sooner than will be possible with ten-gigabit Fibre Channel. The cost and price leverage resulting from higher volumes can be seen in the price differences for gigabit Ethernet and one gigabit Fibre Channel switches--Fibre Channel switches today cost at least three times as much per port compared to gigabit Ethernet switches.

However, room exists for both Fibre Channel and Ethernet IP-based SANs since different environments have different performance and cost objectives. These two approaches provide a flexibility that allows enterprises to match their performance and price requirements to a best solution.

The Internet Engineering Task Force (IEIF) And Standards

The IETF is a large, open, international community comprising network designers, operators, vendors, and researchers. This powerful Internet oversight group monitors Internet architecture evolution as well as its operations in order to maintain Internet order and operational continuity. IETF activities span numerous Working Groups formed by interested parties. One such group is the IP Storage (IPS) Working Group.

The IPS Working Group exists to develop IP encapsulation standards for existing protocols such as Fibre Channel and SCSI. The IPS Working Group operates in a non-disruptive manner since other standards organizations, such as the ANSI T11 and T10 respectively, control these protocol standards. The IPS Working Group's clear incentive to transport SCSI over IP is that it allows Ethernet the opportunity to become an alternative to Fibre Channel for implementing SANs.

In addition to the IETF, another industry group, the Storage Networking Industry Association (SNIA), is developing additional standards to facilitate the management and successful evolution of storage networks. One such working group in SNIA is developing a Fibre Channel Host Bus Adapter Application Programming Interface (HBA API), which will facilitate the management of Fibre Channel HBAs by standardizing properties, discovered end nodes, LUN mappings, etc. Other recent initiatives within SNIA have formed to develop additional standards for IP storage including an API and MIB. Much collaboration is expected between the Fibre Channel and IP storage groups within IETF and SNIA since many of the companies involved in Fibre Channel today (either as a system provider, user, or supplier) also have interests in IP storage technology.

The IETF Process

The IPS Working Group performs most interim work using email communication and physically meets three times a year. Individual members first submit material for consideration as individual draft submissions. The December meeting, held in San Diego, CA, considered several potential IP storage drafts. After achieving a rough consensus on any particular standard, the group then produces an IETF Working Group Internet Draft. Currently an Internet Draft exists for iSCSI, the most likely IP storage standard to emerge soon.

The company names you might find associated with various draft work meeting include Adaptec, Agilent, Cisco, EMC, Hewlett-Packard, IBM, LightSand Communications, Nishan, Pirus Networks, Quantum, SanGate, SANRAD, Stonefly Networks, Sun Microsystems, etc. Many other enterprises participate using the IPS reflector. Clearly there is a broad level of participation in the IPS Working Group.

When a draft is considered acceptable by consensus, the Internet Drafts can next become Request for Comments (RFC) proposals which, when modified to incorporate those comments, can become standards. Presently, various Internet Drafts are before the IETF PS Working Group. These include, but are not limited to:

* iSCSI--Encapsulating SCSI over TCP/IP networks.

* FCIP--Tunneling Fibre Channel traffic over TCP/IP networks.

iSCSI will likely follow the timeframe goals the IPS Working Group adopted in its charter--a February 2001 requirements draft and a May RFC proposal. Other draft standards in progress include FCIP and iFCP. In essence, iFCP is a proposal to support Fibre Channel protocol using Ethernet IP as a substitute network. There is also a proposal to merge the FCIP and iFCP standards proposals but no decision on that has been made.

Despite proposals in some areas, many serious unresolved issues exist for IPS, including IP segment loss recovery, device naming conventions, discovery procedures, data packet numbering, message framing, security, and management information blocks (MIBs). Simply stated, there is still a great deal of work to do.

Products will be introduced and deployed before the completion of all this standardization work. Much product development work is now being done to the iSCSI Internet draft, and as the RFC version emerges, products will emerge conforming to this version of the standard. Since much of the IPS layer implementations will be in software and firmware, these initial products are expected to be upgradable to conform to later revisions of the IPS standards.

Compared to many other technologies, SANs are a relatively recent development. However, because of the critical role they promise to play in mission critical enterprise computing applications, SAN implementation approaches are receiving intense scrutiny. Consequently, they are evolving rapidly as they achieve interoperability with Internet and existing enterprise infrastructures. An essential accelerant in this evolution is the participation of the IETF in developing standards for Ethernet P-based SANs. An important one is known as iSCSI.

The IETF's IPS Working Group is presently reviewing several Working Drafts that will eventually standardize how Ethernet IP networks natively transport encapsulated SCSI. This will be a fully described, standardized IP transport methodology. Presently, much of the work is in its infancy and much work remains to be done. But before all the work is complete, implementations will appear that will gracefully transition to final standards because implementations are largely software based.

When the work is complete, the adopted Ethernet IP-based approach will immediately assume the level of a global standard for IP SAN applications. Alternatives to or deviations from this standardized approach will likely experience the usual fate-market confusion and poor acceptance.

Understandably, this development of using standardized IP network technology for storage networking will likely play a major role in reshaping the competitive SAN market.

Dave Schwaderer is the technical marketing manager at Adaptec (Milpitas, CA).