The Evolution Of Network Storage.
This article is the first in an ongoing series related to the newly formed Open Storage Networking (OSN) initiative. Please see the June issue of CTR ("OSNI Puts FC's Feet To The Fire," page 1) for a complete discussion of OSNI.
The convergence of storage and networking is upon us. Network Appliance and Quantum/ATL recently announced the formation of Open Storage Networking (OSN) initiative with participation by Amdahl, Cisco Systems, Foundry Networks, Legato Systems, and VERITAS Software. This was the latest in a series of storage initiatives extending across traditional industry and technology boundaries with the objective of providing truly open storage networking for continuous access and availability of data within network environments. The OSN initiative is focused on the utilization of Gigabit Ethernet for high performance storage networking and invariably invites comparisons with alternative implementation technologies such as Fibre Channel SANs. Although there has been a great deal of discussion in recent months on the relative advantages and disadvantages of specific storage network architectures such as SAN and NAS, I would like to look beyond the current issues, which are typically addressed in the context of local networ ks to anticipate those issues which will be key considerations for distributed networks.
A Historical Perspective
We are at the beginning of a third era in the evolution of storage solutions. This era--the era of network storage--will have substantial short-term implications for end users, as well as long-term implications for the continuing development of the Information Technology (IT) industry.
The first era was that of internal storage. In these architectures, storage was highly integrated with processor technology to realize the efficiencies necessary to provide the performance improvements that were demanded as the utilization of IT accelerated. The cost of storage technology was sufficiently high to encourage the development of storage-efficient architectures and applications and to constrain the scale of storage growth to a level that could be implemented internally.
The second era was that of external storage. As the cost of storage technology declined and as the standardization of external storage channels stabilized with protocols such as ESCON and, later, SCSI, system architectures evolved to the point that many applications required augmenting internal storage with storage that was external to the processor. The growth of external storage was also fueled by the rise of heterogeneous computing environments, which could share external storage much more effectively than internal storage. EMC, today's premier storage systems company, built its market on the growth of external storage.
The third era is that of network storage. The continued decline of storage technology costs, the emergence of high-bandwidth networking, and the development of access-dominated applications have led to the emergence of a series of new storage architectures, which--while they differ in design concept and technological approach--share the common characteristic of being integrated into the system architecture by means of general purpose networks rather than dedicated storage channels. While all major storage suppliers today identify network storage as an important and durable element of system architectures, Network Appliance, today's premier network storage provider, began addressing this opportunity over eight years ago.
Today, network storage is implemented in one of two major forms: either Storage Area Network (SAN) or Network Attached Storage (NAS). SANs are basically an extension of the dedicated storage channel architecture, utilizing Fibre Channel technology to provide enhanced physical distribution of storage resources, enhanced sharing and access to storage resources, and enhanced manageability. NASs, on the other hand, are an extension of client networking to provide enhanced independence between the storage resources and the server operating systems and file systems, enhanced sharing and access to data resources in heterogeneous environments, and enhanced scalability.
Network storage has grown extremely rapidly in recent years (approximately 90% from 1998 to 1999, according to IDC) and is projected to grow at a 66% CAGR over the next three years. While network storage represented only 15% of the external storage market in 1999, it is projected by Dataquest to grow to over 45% of the external storage market by 2002.
SAN And NAS: Similarities And Differences
As implemented today, SAN and NAS architectures can be distinguished in three basic areas: network transport, network protocol, and data structure.
* Network Transport
SANs and Fibre Channel networks are often used interchangeably. While Fibre Channel is not the only possible transport for SANs, it does remain an excellent choice because of its bandwidth (1Gbps with 2Gbps within a year) and its distance capabilities (up to 10Km). NAS architectures, on the other hand have typically been implemented using the Ethernet transport--historically fast Ethernet (100Mbps), but today, more typically, Gigabit Ethernet (1Gbps) with the expectation for 10 Gigabit Ethernet within a year.
While both Fibre Channel and Ethernet technologies are developing rapidly, industry analyst IDC has projected that Ethernet will maintain a significant cost advantage (at equivalent bandwidth) compared to Fibre Channel over the next five years. The balance between Fibre Channel and Ethernet transports (as well as others) will continue to evolve over the next several years. It is very likely, for example, that, at bandwidths above 1Gbps, all networks will use optical rather than copper physical transport. The network transport, then, while a significant element of the system implementation, does not drive the architecture of storage networks.
* Network Protocols
SAN architectures use a straightforward extension of the SCSI protocol to transport data efficiently over local networks. Latency is a critical issue, particularly for enterprise storage applications in which acceptable limits for network latencies are similar to device latencies in hard drives--approximately five milliseconds. In this environment, the efficiency of the SAN protocol can be very important. The sparseness of the protocol, however, makes it difficult to extend a SAN beyond a local networking environment.
NAS architectures, on the other hand, use the TCP/IP (which, for simplicity, we will call IP) protocol. IP is the protocol used in most Ethernet environments for Unix and Windows file transfers. It is also the protocol used in transport data structures through the Internet. It has spawned an industry of hubs, routers, switches, and other devices to manage it in all kinds of physical network environments. It is the closest thing to a ubiquitous standard in the IT industry today. The sophistication of the IP protocol can introduce additional latency in storage networks, but it can also significantly improve the integration of local storage networks into distributed enterprises such as a Metropolitan Area Network (MAN). For distributed networks, network latency is often limited by communication (i.e., "speed of light") latencies, which can be greater than storage device latencies. TCP/IP has been used in wide area networking (X.21, X.25) for ten or more years.
* Data Structures
Consistent with their heritage as extensions of storage channels, SAN architectures transport data blocks related to the physical storage devices in the network. Once again, this provides for very efficient data transport in highly structured network environments. NAS architectures, on the other hand, transport data files, using one of several well-established network extensions of traditional file systems (NFS, in the case of Unix, and CIFS, in the case of Windows), which permits more loosely structured network architectures. Either approach may be advantageous in specific local network environments, but the ability to transport complex data structures will be a critical issue for distributed networks.
The Impact Of The Internet
While the growth of the Internet has brought a new and different class of storage-intensive architectures to the market, it has also had a very synergistic impact upon the evolution of storage architectures within "classical" local IT systems. The access- and communication-intensive nature of the Internet has driven its evolu-tion on a path, which is significantly different from that of more classic IT systems. Specifically, the rapid adoption of HTML (and now, XML) data structures, the non-deterministic nature of the Internet's search-engine-based "file system," and the extensive use of distributed replication in the form of caching all reflect the unique characteristics of Internet functionality and economics. At the same time, the Internet has enormously increased the rate of generation (actually, local replication) of data in almost all IT environments and has extended the access of organizational databases of organizations engaged in e-commerce from hundreds or thousands of users to literally millions o f users. Both the Internet and local IT architectures will continue to develop and, since the interaction between them will continue to increase over time, it is reasonable to expect that the evolutionary paths of these developments will begin to reflect their significant interaction.
Implications For The Future
So what does this mean for storage? I believe that there are several critical trends to consider in projecting the continued evolution of storage
* The seemingly insatiable corporate appetite for storage, driven by applications, including email, e-commerce, and data replication, will demand the scalability and manageability advantages of network storage.
* The rapid development of technology, which makes network storage more effective.
* The need to protect storage investments--which have grown to 75% of overall IT investments--will require adherence to open standards.
* The economics of distributed networks are improving rapidly, making approaches that encompass local and distributed networking increasingly important.
* TCP/IP is winning the protocol battle. Its limitations are primarily related to processing intensity, while its advantage is its ubiquity in local and distributed networks.
Network storage will, therefore, continue to evolve from its current state. As high bandwidth communication becomes more affordable, network storage architectures will extend beyond local environments to incorporate truly dis-tributed enterprises. Open initiatives such as the OSN will, I believe, play an important role in guiding that evolution toward effective solutions for the demanding storage needs of tomorrow's enterprise.
Kevin Daly is the president and CEO at Quantum/ATL (Irvine, CA).
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Technology Information|
|Publication:||Computer Technology Review|
|Date:||Jul 1, 2000|
|Previous Article:||Switch Interoperability Brings Fibre Channel Forward.|
|Next Article:||Identity Crisis Jolts Workstation World.|