Standards, interoperability, and reliability propel Fibre Channel to the forefront

Standards, interoperability, and reliability propel Fibre Channel to the forefront

In high-speed, backbone local area networks, Fibre Channel technology outclasses Fast Ethernet, switched fddi, and ATM in overall performance, scalability, and life-cycle costs

tim donaldson

ancor communications inc.

To meet the technical computing application demands for increasing bandwidth in the enterprise, local-area network (LAN), fiber-optic backbone environment, Fibre Channel switching technology offers superior capabilities compared to other approaches, such as Asynchronous Transfer Mode (ATM), Fast Ethernet, Gigabit Ethernet, and switched Fiber Distributed Data Interface (fddi). It can handle different kinds of traffic, operate at gigabit speeds, perform reliably under real-world conditions, integrate readily into current infrastructures, compete affordably in cost, and accommodate the digital voice, video, and data communication services expected in the next few years.

Although ATM, Fast Ethernet, Gigabit Ethernet, switched fddi, and Fibre Channel each offers specific strengths, network planners need to carefully consider potential compromises as well. For examples, fddi can be expensive for general use in the enterprise network; ATM offers the promise of voice and video over a common network plus performance over a wide area network (WAN), but may be hampered by cost, complexity, and quality-of-service issues; and Gigabit Ethernet standards (as well as products based on these standards) are not anticipated to be available until 1998.

In choosing the proper high-performance network technology, network planners and users are looking for ways to extend the utility and performance of their existing technology and equipment. Many cannot wait for proposed technologies, such as Gigabit Ethernet, to turn into reality; but today`s selected technology solution must also work in tomorrow`s high-performance network environment.

In addition, the ability to support WAN, LAN and channel protocols makes Fibre Channel technology well-suited to network backbone and channel applications. This interoperability uniquely positions Fibre Channel as a common gigabit-speed connectivity technology across the enterprise. Furthermore, Fibre Channel can handle the growing traffic in client/server networks, as well as connect directly to large storage devices and other peripherals.

Fibre Channel has emerged as an available and key network solution for demanding, high-bandwidth communications applications because it is an ansi-standard, switched protocol that allows concurrent communications among servers, workstations, mainframes, data storage devices, and other peripheral devices (see figure). The first network technology to deliver true gigabit performance, Fibre Channel is 10 times faster than other current network technologies, such as Fast Ethernet and fddi. Moreover, it was created from the beginning with emphasis on user needs, performance, scalability, and interoperability.

Protocol support

Because today`s networks usually run a variety of LAN, WAN and channel protocols, the backbone switch should run as many upper-layer protocols as possible. Fast Ethernet and switched fddi can handle network protocols like IP (Internet protocol) and IPX (Internet protocol extended), but cannot run channel protocols such as scsi (small computer system interface). In operation, ATM manages IP traffic through two competing methods: Classical IP and LAN Emulation.

Unlike the other technologies, Fibre Channel can support both channel and network protocols. It uses unique physical, encoding and framing methods that map both channel and network traffic into Fibre Channel frames. This approach enables Fibre Channel to handle a scsi operation between a computer and a raid (redundant array of independent disks) storage device, or a file transfer from one tcp/ip (transmission control protocol/Internet protocol) device to another simultaneously over the same switching fabric. In addition, Fibre Channel supports such channel protocols as IPI (intelligent peripheral interface), scsi, hippi (High-performance Parallel Interface), and the Bus and Tag protocol for IBM mainframes. Network protocols supported by Fibre Channel include ieee 802.2 Ethernet, IP, and ATM.

In many enterprise networks, handling bandwidth-hogging, delay-sensitive data, voice, and video traffic already is a key requirement. Both ATM and Fibre Channel can handle data, voice, and video traffic. Currently, however, the ATM Forum has specified a voice standard only for constant-bit-rate traffic. For video, ATM uses the mpeg (Motion Picture Experts Group) format.

Unlike the connectionless architecture of ATM, Fibre Channel offers both connection-oriented and connectionless transmissions, as well as broadcast. Furthermore, Fibre Channel uses three distinct classes of service to accommodate different types of transmission needs, including multimedia across the network. Class 1 service is a circuit-switched connection with an end-to-end physical path set up before transmission of data begins, and is well-suited for voice and video applications. Class 2 is a connectionless frame-switched service that does not have a dedicated connection but does guarantee delivery of data. Class 3 is a connectionless datagram service that does not confirm delivery.

Speeds, feeds and needs

In backbone LANs, ATM is frequently associated with Synchronous Optical Network (Sonet) OC-3, 155-Mbit/sec and OC-12 622-Mbit/sec transmissions. Switched Fast Ethernet and fddi top out at 100 Mbits/sec. The present speed leader, Fibre Channel, runs at 266- and 1062-Mbits/sec per port, full duplex (see Table 1), with ansi standards in place for 2- and 4-Gbit/sec performance.

Raw speed, though, is only part of the transmission equation. In the real world, network planners need to know how the network handles traffic, guarantees delivery, and holds up under congestion. Departmental network switching has provided traffic relief for shared-media 10- and 100-Mbit/sec Ethernet networks buckling under the bandwidth demands of today`s computer, imaging, and other data-intensive applications.

In the backbone network, however, the requirements are more complex. Backbone switches need to handle different protocols and different types of traffic and architectures; they must also deliver true performance, compatibility, and quality-of-service to ensure overall network integrity. And, they need to scale to serve growing user needs.

How, and how efficiently, the network handles congestion directly impacts network performance. Fibre Channel bundles data in variable-size frames (containing as many as 2112 data bytes and 36 overhead bytes), using "credit-based" flow control to manage communication between sending and receiving nodes. In this "end-to-end" flow control scheme, the receiver issues a "credit" that determines how much traffic the source can transmit. Frames then remain at the source until their destination is ready. This approach ensures data integrity, even in a bursty LAN environment.

Fast Ethernet switches use either backpressure to slow traffic at the source and avoid congestion, or large buffers that interleave inbound and outbound traffic. The 100Base-T Fast Ethernet devices use the csma/cd (carrier sense multiple access with collision detection) protocol to govern collisions and retransmissions. And fddi adjusts the timed target rotation time (ttrt) to handle greater loads, although industry field tests indicate that fddi switches can drop to just a 22% frame delivery rate under 150% network overload conditions.

Using a rate-based system of flow control, ATM switches monitor the network using resource-management cells in an end-to-end loop that indicates how much bandwidth can be used for available-bit-rate traffic. When the switches detect congestion, they modify the rate specified in the resource-management cell, telling the source to slow down and back up traffic into "leaky bucket" storage buffers. When these buffers overflow, however, head-of-line blockages can occur that affect the entire buffer, resulting in the loss of cells.

Obviously, mission- or timing-critical voice and video applications (such as videoconferencing) cannot tolerate this spillage. The loss of a single cell in an ATM transmission can lead to higher-level packet retransmissions and exponential increases in traffic through the switch. A similar end-to-end flow control scheme has been offered by the ieee 802.3¥Task Group as a core proposal for the basis of the first draft of the Gigabit Ethernet standard.

Cell or frame size and payload efficiency also form important parts of the transmission equation. Fibre Channel uses variable-size frames of up to 2148 bytes, with 2112 bytes of effective payload for a 98% efficiency. The efficiency leader, switched fddi, offers a larger pay load of up to 4500 bytes, requiring only 22 bytes of overhead, for a 99% efficiency. Fast Ethernet also stacks up well at 98%, using the standard Ethernet payload of up to 1500 bytes with only 26 bytes of overhead. Far behind is 155-Mbit/sec ATM; it uses a 5-byte overhead for every 48-byte payload for an efficiency of just 83%.

Cost factors

In today`s high-performance switched networks, a useful comparison metric is to examine cost per megabit-per-second of throughput. This takes into account cost-per-port and network speed, thereby giving a valuable measure of how effectively--and profitably --network resources and users can be deployed in the real world (see Table 2).

Cost studies, based on published price information, demonstrate that fddi technology is expensive ($109.11/Mbit/sec) due to high switch port costs. On the other end of the scale, Fast Ethernet leverages available throughput with a low per-port cost to deliver a price per Mbit/sec of $12.83, although Fast Ethernet performance is limited by 12.18 Mbits/sec of available throughput. The ATM price per Mbit/sec is approximately $23.85. The incremental speed afforded by gigabit Fibre Channel switched networks delivers the lowest price per Mbit/sec of any high-performance network: $9.50 per Mbit/sec. This cost is 35% lower than Fast Ethernet, and less than half the cost of ATM.

Real-world applications

Today`s high-performance data applications--such as film and video production where servers and workstations run special, high-speed, and data-intensive applications for editing and special effects--are handled effectively by Fibre Channel technology. For example, the movie "Independence Day" includes more than 350 digital special effects and 12.8 Tbytes of stored data. For digital post-production work, film is digitized frame-by-frame, and each frame represents 40 Mbytes of data. Multiplying this number by a projection speed of 24 frames-per-second for a 70-mm feature film results in large quantities of digital data.

A post-production movie company, Pacific Ocean Post, turned to Fibre Channel technology to speed up film`s processing work. It used an Ancor Communications` FCS 266 Fibre Channel switch to connect the server and workstations to a scanner and film rendering device. The switch`s adapter cards and drivers deliver Fibre Channel bandwidth direct to the desktop. The company`s systems engineers estimate that the Fibre Channel network yielded a 200% improvement in throughput, and doubled (from 1500 to 3000) the number of frames that could be digitized over a 24-hour period.

Complex visualization, computer-aided design and manufacturing, and video/voice applications also are well-suited to Fibre Channel networks. Gigabit-speed Fibre Channel networks can run real-time uncompressed video, voice, and data applications; deliver broadcast-quality sound and pictures across the network; and eliminate the annoying jerkiness and poor audio usually associated with desktop videoconferencing. For scientific visualization and medical imaging, where quality and data integrity are mandatory, Fibre Channel backbone ensures a robust, high-speed link.

In addition to backbones, Fibre Channel technology also can be used in workstation clustering and network-attached storage applications. These applications require high-bandwidth and high quality-of-service connectivity solutions for archiving, data backup, and data analysis. In addition, Fibre Channel`s interoperability and high performance permit workstation clusters that can deliver mainframe, and even supercomputer, computing power at low cost. Using Fibre Channel to connect storage devices results in direct, high-speed access to information without the need for an intermediate server.

The move to more bandwidth-intensive mission-critical applications on the network bodes well for standards-based technologies such as Fibre Channel that offer true gigabit performance plus scalability to support a growing network. And with ansi standards already in place that support 2- and 4-Gbit/sec Fibre Channel performance, this technology should keep pace with user needs now and into the future. u

Tim Donaldson is vice president of sales and marketing at Ancor Communications Inc. in Minnetonka, MN.

More in High-Speed Networks