Storage Networking--Promises, challenges And Coming Convergence.

Storage technologies and architectures are in a period of accelerated evolution, driven by the spiraling growth of data. Historically, adding storage to a server meant shutting down all applications on the sewer and temporarily disrupting end user access to all information held by that system, but today, as information systems are no longer closed entities but are open to the knowledge workers throughout the enterprise and even customers and suppliers, it is critical that information be accessible on a continuous basis. The storage network has emerged as a primary component of the IT infrastructure that has, as its primary goal, the delivery and protection of information.

The trend toward the networking of storage must encompass a mix of computing platforms, communication protocols, storage devices, and network topologies. Various standards, communications types, file system protocols, and interface buses exist to connect hosts to storage devices and form a storage network. The Storage Networking Industry Association (SNIA), an organization made up of over 100 global IT companies, and the Fibre Channel Industry Association (FCIA) are working to promote the use of Fibre Channel in a multiple vendor environment and to create standards for the high-level protocols that allow SAN devices to inter-communicate.

The SNIA has proposed the following definition for storage networking: "Storage networking is the practice of creating, installing, administering, or using networks whose primary purpose is the transfer of data between computer systems and storage elements." There is no specific reference to a particular interconnect technology.

Today, the terms SAN and NAS are used when we speak of storage networks. In reality, these terms are specific references to a certain upper level protocol on top of a certain kind of "wire." SAN refers to a storage network using the SCSI command set on Fibre Channel media. NAS refers to devices that attach to an IP network via Ethernet media.

Both variations of a storage network enable a switched topology, offering extended distance, scalability, and addressability. Both storage network designs can also be based upon a 1Gbps foundation and both interconnect technologies enable the implementation of functions that are critical to support the delivery and protection of information for the enterprise.

The emergence of Fibre Channel and Gigabit Ethernet are foundations for high-speed connectivity that can be considered for the "plumbing" of a storage network. Both technologies provide an equivalent link bandwidth of 1Gbps. Fibre Channel is the basis for the Storage Area Network (SAN), but Gigabit Ethernet can also be used to form a dedicated network for data traffic between computers and storage devices. Storage devices that attach to Ethernet technology are referred to as Network Attached Storage (NAS) devices. NAS subsystems utilize network file sharing protocols to transport data on an Ethernet storage network. This contrasts to the Fibre Channel SAN that requires the use of the SCSI channel command set for data transportation. Network file sharing protocols are based on mature, stable technology with many benefits and, in many cases, they are not yet available with Fibre Channel SANs.

Emergence Of Fibre Channel

The Fibre Channel switching, hub, and routing technology provides a connectivity basis that promises to allow an IT manager to share, manage, and protect data assets more strategically and efficiently. Although offering very high bandwidth and much promise for the future, the immaturity of the Fibre Channel SAN is currently beset with many interoperability issues that must be addressed before a truly "open" SAN, which is able to connect heterogeneous devices, is a reality.

Meanwhile, IP-based storage networks, like Gigabit Ethernet, with network-attached storage (NAS) devices, offer much of the vision of the SAN today. NAS appliances are highly optimized, single purpose servers dedicated to data storage. NAS appliances process requests for data using stable and mature networking protocols such as Network File Sharing (NFS) and Common Internet File System (CIFS).

Fibre Channel is a communications protocol that can provide high-performance data transfers for channel and Local Area Network (LAN) traffic among workstations, mainframes, servers, data storage systems, and other peripherals using Small Computer Systems Interface (SCSI) and Internet Protocol (IP). Although the Fibre Channel has been defined with the capability to support IP, the Fibre Channel hardware used in a SAN supports only the SCSI protocol. Within the next 12 months, Fibre Channel components will be available that will allow the integration of the IP-based NAS into the Fibre Channel SAN. A Fibre Channel storage network of the future will be able to operate with both SCSI and networking (IP) protocols.

Fibre Channel has emerged to help circumvent the speed, distance, and connectivity limitations of the traditional parallel SCSI environment. Traditional SCSI can connect a maximum of 16 nodes, while SCSI Fibre Channel can connect up to 126 nodes in a single loop and more than 16 million nodes in a switched topology.

Fibre Channel is designed as a communications protocol that merges characteristics of the channel and the LAN. A channel is a hardware intensive design that supports very high effective data rates, but at the cost of limited distance and addressability. On the other hand, a LAN needs to be more flexible and, therefore, has a more software intensive design with the benefit of extended distance and connectivity and at the cost of increased latency. In turn, Fibre Channel is hardware intensive to promote high speed and it is designed to scale device connectivity at kilometer distances (See Fig).

The Fibre Channel is a layered architecture. The first three layers are designated as FC-PH or the physical layers.

FC-0 defines the link speed. The fastest link speed available today is 1Gbps (or 100MB/sec), but 2Gbps and 4Gbps have also been adopted. Expect to see 4Gbps Fibre Channel links shipping within 1218 months.

FC-0 also defines the options available for physical media. The physical media can be electrical or optical fiber. The choice of physical media will primarily dictate maximum distance (i.e., the maximum distance for electrical fiber is 25 meters, whereas the maximum distance for multimode optical fiber is 500 meters, and for single mode optical fiber 10,000 meters).

FC-4 provides for Upper Level Protocol (ULP) mapping capability. Channel, network, and multimedia protocols can be mapped onto the Fibre Channel physical layers. From a storage networking perspective, the most important ULP is SCSI and IP (Internet Protocol).

IBM's FICON is the mapping specification for the ESCON command set. With this announcement, IBM has signaled that the Fibre Channel storage network may extend throughout the enterprise to include the mainframe, plus Unix and Windows servers. As of this writing, the FICON ULP has not been adopted by ANSI as part of the Fibre Channel standard.

Network-Based File Sharing -- Mature And Stable

In contrast to the use of Fibre Channel-based SCSI as the data transport mechanism, the sharing of files using standard TCP-IP networks has been done for more than 15 years. In 1985, Sun Microsystems released the first commercial version of NFS (Network File Sharing) protocols. Sun made the NFS source available to the industry at low cost and, as a result, NFS is supported in virtually every operating system in the world. The version currently shipping is 3.0 and it represents a stable, standardized mechanism for sharing files in a cross-platform environment. NFS is predominantly used in the Unix environment, although NFS stacks are available for Microsoft Windows and OS/390 platforms.

In a NFS environment, the file server can designate certain file systems as exportable. Client servers or workstations can request these files and applications use these files as if they were locally attached.

Also in the early 1980s, Intel and Microsoft introduced the Server Message Block (SMB) protocol. SMB provides NFS functionality for the Windows environment. SMB has since been renamed CIFS (Common Internet File System) and was ratified as a X/Open standard in 1992. Today, it is the file sharing mechanism for Windows/NT and Windows2000 and has also been implemented on some Unix platforms.

Storage networking performance--the speed of the network and server overhead--has been a major inhibitor to the widespread use of network file sharing. A component of performance is link speed. With the introduction of Gigabit Ethernet and Fibre Channel, network link speeds are now equivalent to channel link speeds. This enhances the viability of using the network for storage access.

The Network "Appliance"

The emergence of the "appliance" addresses the second half of the performance issue--server overhead. An appliance is a special purpose, highly optimized computing platform that is designed to perform only a limited number of functions. This is in contrast to a general purpose computer and an operating system that must be able to provide services and facilities for a very diverse set of applications. By necessity, the design of a general purpose operating system involves many trade-offs.

There are many examples of appliances today, including network routers, switches, PDAs, and file servers. With the emergence of filer appliances, network-based storage is a more viable alternative.

As stated earlier, Fibre Channel storage network is very efficient and, thus, can sustain a very high effective bandwidth. On the other hand, today's Fibre Channel storage network has many of the attributes of direct connect. SAN storage devices are mostly dedicated to a specific server, even though the Fibre Channel topology supports multiple server "visibility." Data sharing is still primarily accomplished via the network. With the use of the SCSI channel command set, Fibre Channel storage subsystems have a volume level orientation.

The lack of a common cross platform file structure is inhibiting true data sharing in the Fibre Channel SAN, as in the traditional parallel SCSI environment. There are ongoing projects from several major companies to create file systems that can support true data sharing, but industry-wide standards for Fibre Channel file sharing are not expected for another couple of years.

NAS subsystems, by contrast, attach via an IP network and data is accessed using network-based client/sewer protocols such as NFS and CIFS. A NAS subsystem can present a common view of data to all requestors. The integration of file system logic and the file level orientation of the NAS subsystem is the primary distinguishing attribute of this device as compared to the SAN. The embedded file system logic also offers opportunities for the NAS designer to transparendy extend file system capability to enhance application performance and availability. True data sharing can be supported because the NAS is accessed by using standardized networking protocols.

The data sharing and extended functionality of the NAS can be relatively easy to implement. A NAS subsystem can plug into the networks that enterprises already have and understand. Existing networks may not have to be changed at all. The primary consideration for this implementation is the interference that NFS/CIFS packets may have on normal network traffic. This implementation may overload and negatively impact overall network performance. On the other hand, it is also possible to create an independent IP network-based on Gigabit Ethemer technology whose function is to allow the attachment of NAS devices to sewers. This implementation promotes secure access to data and also isolates I/O traffic from normal network traffic. Conceptually, this framework favorably compares to a Fibre Channel SAN.

Link Rate Convergence--Ethernet Vs. Fibre Channel

Both the Fibre Channel SAN and the Ethernet NAS network leverage high-speed media for fast access to data. Though different, the SAN and NAS topologies are ultimately complementary. Today's NAS device may become a node on tomorrow's SAN. Today, both approaches to networked storage are enabling companies to go beyond the limitations of server-attached storage.

Although Ethernet and Fibre Channel support equivalent peak link rates, the Fibre Channel can achieve higher effective use of the total bandwidth. Within the next 12-18 months, Fibre Channel will be available at the 4Gbps link rate. At the same time, Ethernet is expected to be available at 10Gbps. So where networks were much slower than channels in the past, they are now converging in terms of effective link rate.

Fibre Channel Exploitation

In 1998 and 1999, the underlying components of the Fibre Channel SAN infrastructure were just beginning to be installed. As with any new technology, there is a gradual exploitation and realization of the benefits of the technology. The Gartner Group research firm predicts a gradual exploitation of this infrastructure over the next several years. Over time, Fibre Channel storage networks will embed high availability, data replication, and fast recovery services to support the information needs of the enterprise. A NAS storage network offers much extended functionality today and can support the ability to share a single instance of data across homogeneous and heterogeneous platforms. NAS-extended file system functionality offers the ability to perform "virtual" data replication. Virtual replication, or snapshots, allows the logical duplication of files without a 100 percent physical replication. NAS Snapshots can be used to reduce the backup window and be directly accessed by end users for fast recovery of data .

Frederick M. Shields is the storage manager of marketing at Amdahl Corporation (Sunnyvale, CA).

How SAN & NAS Compare

COMMON CHARACTERISTICS

Any-to-any switched connectivity

Increased scalability and connectivity

Enhanced consolidation capability

Extended distance

1Gbps bandwidth

Enabling structure for

Centralized backup

Disaster recovery

HA clustering

Data sharing

SAN                                 NAS
Description: A Fibre Channel        Description: Platform independent
based network of storage            disk storage device, which
devices. Fibre Channel connectivity contains a special-purpose operating
provides direct connection          system that processes
from any server to any storage      IP-based file requests.
device on the SAN. Devices on a     The device must connect to a
SAN communicate using the           network that supports TCP-IP.
SCSI command set.
Performance                         Performance
1 Gbps bandwidth                    1 Gbps bandwidth
Reduce LAN/WAN contention by        Highly optimized File-serving
putting Backup traffic on the SAN.  operating system.
How is data accessed?               How is data accessed?
In blocks, SCSI command set.        In files. IP file-sharing protocols
                                    (NFS and CIFS)
Connectivity                        Connectivity
Typically constructed around        Token Ring, Ethernet, and
1 Gbps Fibre Channel.               Gigabit Ethernet. ATM/OC3
Cross-platform data sharing         Cross-platform data sharing
Not possible today. Current SANs    Today Windows NT, UNIX, and
are homogeneous environments.       Netware servers can share
Multi-platform data sharing is      devices and data
approximately two years away.
Cost                                Cost
Fibre Channel technology is high    More mature, more widely
Cost today.                         manufactured technology is
                                    substantially less on a component
                                    to component basis.