Users can take advantage of IP and Ethernet, but new processors may be required to take full advantage of IP SANs. By Ron Kroesen
End users and storage integrators find themselves in a quandary these days. They know that networked storage has become a need-to-have rather than a nice-to-have capability. After all, the amount of digitized data continues to mushroom, and Internet-intensive applications like online transaction processing and e-commerce have created a volume of content that has exceeded the capabilities of direct-attached storage.
Although it's clear that networked storage has been elevated to the status of a must-have technology, choosing the right approach can confront users and integrators with a complex set of choices.
For example, while network-attached storage (NAS) provides a partial answer as a great remote file I/O solution, it does not address the performance and low-latency requirements for disk access through block-level I/O. This weakness opened the door for storage area network (SAN) technology, which can provide performance, capacity, and flexibility advantages.
iSCSI uses TCP/IP for reliable data transmission over potentially unreliable networks.
Fibre Channel has become the dominant interconnect protocol for SANs. But it still has a few potential drawbacks such as distance limitations that can cause problems in a widely dispersed SAN.
Perhaps more importantly, Fibre Channel SANs present a significant expense at a time when end users are examining every IT expenditure with a magnifying glass. Fibre Channel SANs require their own special network transport, creating two separate network technologies (Fibre Channel and Ethernet) within an enterprise. This translates into further expense, as IT departments need additional administrators to deal with multiple network technologies.
Despite its shortcomings, however, Fibre Channel continues to maintain its grip, particularly at large enterprises. Being the first SAN technology, Fibre Channel has carved out a solid installed base. User sites that are committed to Fibre Channel aren't likely to walk away from that investment.
Still, other networked storage alternatives—under the umbrella term of IP storage—have emerged that promise to expand the size and scope of the market for SANs. IP storage, which uses IP over Ethernet networks, delivers a networked storage alternative that takes advantage of inexpensive and mature technologies (Ethernet and IP) and offers compatibility and interoperability with existing IT infrastructures.
One of the key IP storage alternatives is iSCSI, which, according to the Internet Engineering Task Force (IETF), provides block-oriented storage transport over IP networks. Blocks of data in SCSI format are encapsulated in TCP packets that run on IP networks (see diagram).
iSCSI is one Upper Layer Protocol (ULP) used in IP storage. Another approach—FCIP—allows encapsulation of Fibre Channel data in IP format mainly for connecting Fibre Channel SANs over IP networks.
In a recent report, Bear Stearns argued the proposition that iSCSI could expand the total market opportunity available for networked storage by offering an easier and more-affordable infrastructure for server-storage connectivity. By providing the opportunity to cost-effectively extend the benefits of consolidated storage to remote sites and departmental-level environments, iSCSI can increase the scale and scope of the core data-center SAN to include a wide range of previously isolated servers and clients. Furthermore, iSCSI's use at the edges of a data-center SAN and as an economical means to bridge distant SANs over metropolitan- and wide-area networks can potentially help grow the storage market by expanding SAN access to any user on the network.
Other arguments in favor of IP storage have focused on the premise that one networking technology is easier to deal with than two, if they deliver comparable performance and feature sets.
Cost of implementation is also an important factor in IP storage. Ethernet is an inexpensive technology with near ubiquitous deployment. IT professionals have developed a comfort level with Ethernet, having worked with it for almost two decades.
Finally, LAN/WAN connectivity can be easily supported with iSCSI, and scalability to 10Gbps Ethernet is ensured in the near future.
IP storage gains momentum
The iSCSI specification is expected to be ratified by the IETF in early 2003. The completion of the standard, combined with other developments, should spur end-user adoption of IP storage.
Among those other developments is the announcement from Cisco Systems that it will acquire SAN switch developer Andiamo Systems (see "Cisco squares off against Brocade, McData," InfoStor, October 2002, p. 1). Cisco recently released a line of Fibre Channel switches, developed by Andiamo, that it will market under the MDS 9000 product label. Multi-protocol switches that support Fibre Channel, iSCSI, and FCIP are expected in the first half of 2003.
Cisco also sells the SN 5428 storage router. The router makes it possible for IT managers to connect Fibre Channel devices and have them appear as iSCSI devices on a TCP/IP network, or just to use it as a low-end multi-protocol switch.
IP storage has also received a boost from vendors such as Nishan Systems. The company sells multi-protocol switches that support Fibre Channel, iSCSI, and iFCP, and lists Wells Fargo, JP Morgan, and Carlson Companies among its early customers.
The Carlson installation is the first data-center IP SAN deployment (see "Carlson's data center relies on IP SAN," InfoStor, August 2002, p. 1). The Carlson network architecture is based on Nishan's multi-protocol switches, using the iFCP protocol, connected to Fibre Channel devices.
Combining Fibre Channel and IP
The "peaceful co-existence" strategy employed by Carlson recognizes the complementary nature of IP storage and Fibre Channel and allows users to protect their existing investment in Fibre Channel technology while beginning to implement IP storage within the corporate enterprise.
Given the complementary capabilities of Fibre Channel and IP storage and the possibilities that iSCSI SANs have for growing the networked storage market, it's worthwhile to consider some of the recent technology developments that can help iSCSI maximize its potential.
Specifically, a new type of processor may be required for IP storage devices. Optimized for IP storage, these processors differ from traditional architectures, including general-purpose processors and packet-based network processors. In some applications, traditional processors don't meet all the requirements of IP storage, in part, because they are tied to specific methods of memory access, operating systems, and context switching.
Storage network processors can be used in any equipment that connects to the IP storage network, including servers, host bus adapters (HBAs), disk arrays, NAS appliances, inter-protocol bridges, and virtualization boxes.
These processors can deliver line-speed performance, even in heavy OLTP traffic applications.
The processors are optimized for all the tasks of handling and terminating storage traffic, including the processing of both packets and ULP Protocol Data Units (PDUs), maintenance and management of connections and context, connection setup and teardown, management of packet losses, and the arrival of out-of-order packets.
SANs have enormous potential to solve storage problems, but to fulfill that promise, new processors may be required that are tailored for the performance, flexibility, scalability, and data integrity needs of IP storage networks.
Ron Kroesen is vice president of marketing at Silverback Systems (www.silverbacksystems.com) in Campbell, CA.