1.1. Space and Time
Enterprise networks are facing an explosion of data demands from their users. Applications have become more sophisticated, content more rich and diverse, consumers of data more numerous, and expectations for data access driven higher. The catalyst for this explosive growth in information and information access has been forged by a fundamental change in how data is used and by a profound change in the underlying infrastructure that delivers it.
For most of a generation, large computer systems centralized the data services in the enterprise and tightly controlled information content, format, and access. User requests for changes in data formatting (for example, adding a new field in a customer information display) might be queued for months as rows of monastic COBOL programmers patiently scribed code. Remote access to data relied on expensive, point-to-point leased lines and proprietary communications equipment that inherently limited access. In the end, these monolithic computer systems and networks failed to accommodate user demands for more flexible access to more diverse information and have been superseded by more powerful, more affordable, and more dispersed computer systems and networks. UNIX and Windows platforms have displaced mainframes, and networks built on open standards have displaced proprietary ones.
Today, the typical enterprise data center is a maze of 19" racks housing discrete UNIX and Windows servers and their attached storage. A rack or group of racks may be dedicated to a specific department of the enterprise, such as finance, engineering, or human resources. Whereas previously these departments would have shared computing time on a large mainframe, they now control their own computing resources. And although previously they would have had to stand in line with program change requests in hand, they can now install their own specialized applications and generate their own data content. Data generation is further facilitated by new enabling technologies based on visual programming utilities and Hypertext Markup Language that place powerful content creation tools into the hands of any literate person. Multiply this scenario by the thousands of enterprise networks that have undergone this transformation, and it is easy to appreciate the surge in the volume of new data anticipated during the next few years.
Concurrent with this metamorphosis of computing environments, the nature and use of data itself has changed. Information previously presented as dry text has become rich in content, graphical, animated, and dynamic. Nested links to more granular information allow the user to navigate to the desired level of detail or to click through to associated topics. Static columns of numbers in a spreadsheet may be accompanied by colored charts and graphs; a text document may embed a spreadsheet, graphics, or logo; a human resources record may include an employee photograph as well as scanned documents; and a user may casually append a multimegabyte PowerPoint presentation to an e-mail message.
With the ability of more people to generate more data that has more content, enterprise networks are challenged with providing the space to store data and reducing the time to deliver it. For enterprise networks, space and time are indeed a continuum, because ever-increasing storage requirements stress the ability of networks to provide prompt delivery, and more ubiquitous and efficient networks stress the ability of storage systems to absorb newly generated data. Previously viewed as two separate entities within the enterprise, storage and networking are melding into a single concern.
Today's enterprise networks have, after a long struggle between contending technologies, almost universally converged on the Transmission Control Protocol/Internet Protocol (TCP/IP) as the transport protocol of choice, and Ethernet as the underlying infrastructure. Analogous to the Betamax versus VHS wars in the video-recording industry, TCP/IP and Ethernet were not necessarily the best technologies to triumph over their contenders. They do have, however, the proven advantages of open systems compliance, resiliency to disruption, flexibility in deployment, ubiquity, and affordability as networks scale. Over time, the TCP/IP protocol has been enriched with auxil iary management, security, and encryption capabilities, whereas Ethernet has matured from a shared, low-speed topology to a high-performance, multi gigabit switched transport.
Until recently, networks terminated at the user, at the far end of the system, and the server or computing resource terminated at the point of data access. Network managers have been responsible for allocating bandwidth and monitoring the transport of data from source to destination, but have not been responsible for what happens to the data once it arrives. Likewise, systems and storage administrators have been focused on the placement and security of data under their charge, but have been largely oblivious to the external network infrastructure. As networking and storage technologies merge, these previously separate areas of responsibility are also merging.
During the past 20 years, the more dynamic development of enterprise networking has been in stark contrast to the more conservative pace of storage systems. For decades, the dominant paradigm in storage has been defined by channel architectures at the high end and the Small Computer Systems Interface (SCSI) at the medium to low end of enterprise systems. SCSI is both a protocol and a cabling standard. The SCSI protocol is optimized for the movement of large blocks of data between storage systems and hosts, typically servers. Parallel SCSI cabling provides a parallel data path for the efficient transport of bytes of data over limited distances. Although parallel SCSI performance has gradually increased from 20 megabytes per second (MBps) to higher speeds, it has maintained a rigid master/slave relationship between servers and attached storage resources. Because the server in this case is the owner of and sole gateway to its attached storage, access to the data residing on that storage is vulnerable to server failure.
Perhaps the most significant advance in storage technology was initiated by the creation of serial SCSI, or SCSI-3. Although previous developments provided greater efficiency in data movement, lower profile disk drives, and greater disk capacity, none of these accomplishments altered the server/storage paradigm. SCSI-3 enabled the SCSI protocol to be transported on more flexible networking infrastructures, and thus cut the thick umbilical cord of parallel cabling that previously bound storage to individual servers. With storage now a peer on a common network with multiple servers, its data is accessible even if individual servers fail.
Networking for storage resources on its own, however, would not be viable if it lacked the performance of the technology it hoped to replace. For serialized SCSI to replace parallel SCSI, a quantum leap in network perfor mance was needed. This breakthrough was achieved by gigabit networking—first in the form of Fibre Channel, followed by Gigabit Ethernet. At gigabit speeds, it is possible to conquer time, delivering data as quickly as server resources can digest it. And with large storage pools now networked to multiple servers, space can be mastered as well.
Storage networking has undergone a fairly rapid adoption cycle from early market products to acceptance as a mainstream enterprise solution. Because of the central role that data plays for institutions and commercial enterprises, issues of data placement, security, and transport are universal. Today, all major suppliers of host systems and storage resources provide solutions based on storage networking as their flagship offerings. Although these solutions are being adopted initially by the top-tier enterprise customers, they are steadily penetrating the mainstream market. This rate of adoption will be accelerated by the emergence of new storage network solutions such as storage data over IP.
The first-generation storage networks for block data transport were enabled by Fibre Channel technology. Fibre Channel is a gigabit transport that provides high-performance, long-distance fiber-optic cabling (as much as 10 km), flexible peer networking, and upper layer protocol support for SCSI-3. Although Fibre Channel supports a variety of upper layer protocols, including IP, engineering development has focused on storage applications. A variety of products, including Fibre Channel host bus adapters (HBAs) for servers, Fibre Channel–attached storage arrays, Fibre Channel–attached tape libraries, and Fibre Channel switches and hubs, have enabled the first enterprise-class storage network solutions.
The challenge Fibre Channel has faced is common to any new technology that must engineer a total solution from the ground up. Creating standards for physical and logical layer interfaces, developing and testing entirely new architectures, undergoing long periods of interoperability testing and validating standards compliance, establishing agreement among the developers of hardware and software products for practical implementations, and demonstrating to the market that the solutions offered are viable, scalable, and enduring have been difficult tasks. Add to this considerable set of challenges the fact that storage networking itself introduces entirely new concepts and issues unrelated to the underlying plumbing. Management of storage networks, for example, is complex simply because networking and storage management tools must be combined. The management complexity often associated with current Fibre Channel storage area networks (SANs) will remain, even if Fibre Channel disappears.
While Fibre Channel developers have been struggling with these issues, new IP-based storage network initiatives have emerged. IP and Ethernet could not offer a viable alternative to Fibre Channel as long as Ethernet's performance was limited to Fast, 100-megabit per second (Mbps) Ethernet. Storage traffic for enterprise applications requires at least ten times that bandwidth. Ethernet vendors, however, have benefited directly from the development of Fibre Channel technology. The physical layer and data encoding mechanisms of Fibre Channel were borrowed (or stolen, depending on whom you ask) to accelerate the development of Gigabit Ethernet. With the appropriate bandwidth now in hand, Gigabit Ethernet is now positioned to compete with Fibre Channel for the storage networking market. The development of 10 Gigabit Ethernet will provide something of an advantage until Fibre Channel reaches parity.
What is wrong with Fibre Channel? Probably nothing that additional years of engineering could not fix. The interoperability and maturity of Fibre Channel end systems, however, has not been matched by Fibre Channel fabrics, largely because of competitive interests among the fabric vendors themselves. Without interoperability between director-class and departmental switches, it is not possible to build an enterprise-class storage network. Inaddition, security mechanisms such as encryption and authentication, transport management platforms, and traffic prioritization methods, which are already available for IP and Gigabit Ethernet networks, are still under construction in Fibre Channel protocol standards. With the tens of thousands of engineers at work on IP solutions arrayed against less than a thousand Fibre Channel engineers, it will be difficult for Fibre Channel vendors to deliver full and robust solutions quickly to an increasingly competitive market.
Given the availability and stability of Fibre Channel end systems, it is likely that Fibre Channel–based storage and IP-based storage networking products will be successful for some time to come. No one technology displaces all others, at least not overnight. IP-based storage solutions will become more pervasive over time, though, if for no other reason than the historical momentum that IP and Ethernet have gained throughout mainstream data communications. Customers will always gravitate to solutions that are more familiar, more affordable, and more compatible with their existing infrastructures.