Gigabit Ethernet (GigE) is becoming more and more popular. The accelerating growth of LAN traffic is pushing network administrators to look to higher-speed network technologies to solve the bandwidth crunch. These administrators—who typically have either Ethernet or FDDI backbones today—have several alternatives to choose from. Although each network faces different issues, Gigabit Ethernet meets several key criteria for choosing a high-speed network:
• Easy, straightforward migration to higher performance levels without disruption
• Low cost of ownership—including both purchase cost and support cost
• Capability to support new applications and data types
• Network design flexibility
For gigabit Ethernet bandwidth pricing from multiple carriers, please complete the orange form.
One of the most important questions network administrators face is how to get higher bandwidth without disrupting the existing network. Gigabit Ethernet follows the same form, fit and function as its 10 Mbps and 100 Mbps Ethernet precursors, allowing a straightforward, incremental migration to higher-speed networking. All three Ethernet speeds use the same IEEE 802.3 frame format, full-duplex operation and flow control methods. In half-duplex mode, Gigabit Ethernet employs the same fundamental CSMA/CD access method to resolve contention for the shared media. And, Gigabit Ethernet uses the same management objects defined by the IEEE 802.3 group. Gigabit Ethernet is Ethernet, only faster.
It is simple to connect existing lower-speed Ethernet devices to Gigabit Ethernet devices using LAN switches or routers to adapt one physical line speed to the other. Gigabit Ethernet uses the same variable-length (64- to 1514-byte packets) IEEE 802.3 frame format found in Ethernet and Fast Ethernet (Figure 1). Because the frame format and size are the same for all Ethernet technologies, no other network changes are necessary. This evolutionary upgrade path allows Gigabit Ethernet to be seamlessly integrated into existing Ethernet and Fast Ethernet networks.
In contrast, other high speed technologies use fundamentally different frame formats. High-speed ATM, for example, implements a fixed-length data cell. When connecting Ethernet and Fast Ethernet to ATM, the switch or router must translate each ATM cell to an Ethernet frame, and vice versa.
As defined by the IEEE 802.3x specification, two nodes connected via a full-duplex, switched path can simultaneously send and receive packets. Gigabit Ethernet follows this standard to communicate in full-duplex mode
Gigabit Ethernet also employs standard Ethernet flow control methods to avoid congestion and overloading. When operating in half-duplex mode, Gigabit Ethernet adopts the same fundamental CSMA/CD access method to resolve contention for the shared media. The CSMA/CD method is illustrated in Figure 2.
The Gigabit Ethernet CSMA/CD method was enhanced in order to maintain a 200-meter collision diameter at gigabit speeds. Without this enhancement, minimum-sized Ethernet packets could complete transmission before the transmitting station senses a collision, thereby violating the CSMA/CD method. To resolve this issue, both the minimum CSMA/CD carrier time and the Ethernet slot time have been extended from their present value of 64 bytes to a new value of 512 bytes. (Note that the minimum packet length of 64 bytes has not been affected.) Packets smaller than 512 bytes have been augmented with a new carrier extension field following the CRC field. Packets longer than 512 bytes have not been extended. These changes, which can impact small-packet performance, have been offset by incorporating a new feature, called packet bursting, into the CSMA/CD algorithm. Packet bursting will allow servers, switches and other devices to send bursts of small packets in order to fully utilize available bandwidth.
Devices that operate in full-duplex mode (switches and buffered distributors) are not subject to the carrier extension, slot time extension or packet bursting changes. Full-duplex devices will continue to use the regular Ethernet 96-bit interframe gap (IFG) and 64-byte minimum packet size.
As in the transition from Ethernet to Fast Ethernet, the fundamental management objects familiar to most network managers are carried forward with Gigabit Ethernet. For example, SNMP defines a standard method to collect device-level Ethernet information. SNMP uses management information base (MIB) structures to record key statistics such as collision count, packets transmitted or received, error rates and other device-level information. Additional information is collected by remote monitoring (RMON) agents to aggregate the statistics for presentation via a network management application. Because Gigabit Ethernet uses standard Ethernet frames, the same MIBs and RMON agents can be utilized to provide network management at gigabit speeds.
Cost of ownership is an important factor in evaluating any new networking technology. The overall cost of ownership includes not only the purchase price of equipment, but also the cost of training, maintenance and troubleshooting.
Competition and economies of scale have driven the purchase price of Ethernet connections down significantly. Though Fast Ethernet products have been shipping only since 1994, even these products have experienced significant price declines over the past two years. Gigabit Ethernet will follow the same price trends as Fast Ethernet. Products on the market today provide cost-effective connections for gigabit transmission rates. The IEEE’s goal was to provide Gigabit Ethernet connections at two to three times the cost of a 100BASE-FX interface. As volume builds, reduced line width IC processes are implemented and lowcost opto-electronic devices are developed, the cost of Gigabit Ethernet interfaces will decline.
Switched Gigabit Ethernet connections are lower in cost than 622 Mbps ATM interfaces (assuming identical physical media interfaces), because of the relative simplicity of Ethernet and higher shipment volumes. Gigabit Ethernet repeater interfaces will be significantly lower in cost than 622 Mbps ATM connections, providing users with cost-effective alternatives for data center network backbone and server connections.
Over time, advances in silicon, including 0.35-micron CMOS ASIC technology, will provide even greater performance gains and cost reduction opportunities that will result in a new, even more cost-effective generation of Ethernet technology. Analysis indicates that 0.35-micron processes will achieve 1250 Mbps operation and economically fit one million gates on a single die. This is more than enough to fit a complete Ethernet switch, including management, a significant amount of buffer memory, and an embedded 32-bit controller, on a single die—with obvious cost advantages.
Finally, because the installed base of users is already familiar with Ethernet technology, maintenance and troubleshooting tools, the support costs associated with Gigabit Ethernet will be far lower than other technologies. Gigabit Ethernet requires only incremental training of personnel and incremental purchase of maintenance and troubleshooting tools. In addition, deployment of Gigabit Ethernet is faster than alternative technologies. Once upgraded with training and tools, network support staff are able to confidently install, troubleshoot and support Gigabit Ethernet installations.
The emergence of intranet applications portends a migration to new data types, including video and voice. In the past it was thought that video might require a different networking technology designed specifically for multimedia. But today it is possible to mix data and video over Ethernet through a combination of the following:
• Increased bandwidth provided by Fast Ethernet and Gigabit Ethernet, enhanced by LAN switching
• The emergence of new protocols, such as Resource Reservation Protocol (RSVP), that provide bandwidth reservation
• The emergence of new standards such as 802.1Q and 802.1p which will provide virtual LAN (VLAN) and explicit priority information for packets in the network
• The widespread use of advanced video compression
These technologies and protocols combine to make Gigabit Ethernet an extremely attractive solution for the delivery of video and multimedia traffic
Network administrators today face a myriad of internetworking choices and network design options. They are combining routed and switched networks, and building intranets of increasing scale. Ethernet networks are shared (using repeaters) and switched based on bandwidth and cost requirements. The choice of a high-speed network, however, should not restrict the choice of internetworking or network topology.
Gigabit Ethernet can be switched, routed and shared. All of today’s internetworking technologies, as well as such technologies such as IP-specific switching and Layer 3 switching, are fully compatible with Gigabit Ethernet, just as they are with Ethernet and Fast Ethernet. Gigabit Ethernet is available in a full duplex repeater (with the accompanying low cost per port) as well as on LAN switches and routers.
The simple migration and support offered by Ethernet, combined with the scalability and flexibility to handle new applications and data types, makes Gigabit Ethernet the strategic choice for high-speed, high-bandwidth networking.
Gigabit Ethernet is an extension to the highly successful 10 Mbps and 100 Mbps IEEE 802.3 Ethernet standards. Offering a raw data bandwidth of 1000 Mbps, Gigabit Ethernet maintains full compatibility with the huge installed base of Ethernet nodes.
To recap the recent history of the Gigabit Ethernet standards process, in July, 1996, after months of initial feasibility studies, the IEEE 802.3 working group created the 802.3z Gigabit Ethernet task force. The key objectives of the 802.3z Gigabit Ethernet task force were to develop a Gigabit Ethernet standard that does the following:
• Allows half- and full-duplex operation at speeds of 1000 Mbps
• Uses the 802.3 Ethernet frame format
• Uses the CSMA/CD access method with support for one repeater per collision domain
• Addresses backward compatibility with 10BASE-T and 100BASE-T technologies
The task force identified three specific objectives for link distances: a multimode fiber-optic link with a maximum length of 550 meters; a single-mode fiber-optic link with a maximum length of 3 kilometers (later extended to 5 kilometers); and a copper based link with a maximum length of at least 25 meters. The IEEE is also actively investigating technology that would support link distances of at least 100 meters over Category 5 unshielded twisted pair (UTP) wiring. This standards work will be completed this year. In addition, the task force decided to include a specification for an optional Gigabit Media Independent Interface (GMII) in the scope of its work.
One of the primary goals of the Gigabit Ethernet Alliance was to accelerate the standards activity for Gigabit Ethernet. Fast Ethernet took approximately 13 months to go from first draft to final approval, and Gigabit Ethernet required was approximately the same amount of time (Figure 3). The goal of the IEEE 802.3z Gigabit Ethernet task force accomplished its goal of completing the Gigabit Ethernet standard by 1998, although pre-standard products appeared in 1997.
Table 5 shows the Gigabit Ethernet distance specifications for fiber optic media. For the 62.5 micron diameter 160 MHz*km multimode (MM) fiber often called FDDI-grade fiber, the distance is specified at 220 meters. As is evident from the table, as the bandwidth of the fiber increases, the minimum range for MM fiber increases up to 550 meters. The longwave length transceiver (1000BASE-LX) reaches 550 meters for all media types. For single mode fiber with 1000BASE-LX, the distance is specified at 5 km.
The challenges associated with using lasers on multimode fiber have become more apparent as the operating speed has been increased. In addition, the IEEE 802.3z task force was particularly attentive to the characteristics of the installed base of network cabling, be it copper or fiber optic cabling. The group conducted numerous experimental and field tests to insure that the underlying signaling technology would work on the vast majority of installed cabling. Thus, the IEEE applied far greater scrutiny to the operating characteristcs of lasers on multimode fiber than has ever been applied before. This is the first time that laser launch has been extensively tested over long multimode fiber. These rigorous tests discovered a jitter component caused by a phenomenom known as differential mode delay (DMD).
This jitter effect regarding multimode fiber transmissions was resolved. For 1000BASE-SX, the solution was achieved by qualifying the launch of the laser transmitter, introducing conformance tests for stressed receiver sensitivity and stressed receiver jitter, and reallocating the jitter budget. In addition to these refined transceiver conformance tests, the link distance for the lowest bandwidth multimode fiber was specified at 220 meters. Other fiber types can go further, see Table 5. With 1000BASE-LX transceivers over multimode fiber, external patch cords are used to mitigate DMD. Existing technologies that use the combination of short wavelength lasers and multimode fiber, such as Fibre Channel (FC), have not seen DMD effects because of the short distances used in FC applications.
Since Gigabit Ethernet is Ethernet, the types of Gigabit Ethernet products are quite straightforward: Layer 2 switches, Layer 3 switches (or routing switches), uplink/downlink modules, NICs, Gigabit Ethernet router interfaces, and the buffered distributors. There are pure multiport Gigabit Ethernet switches with high performance backplanes, as well as devices that have both Gigabit Ethernet and Fast Ethernet ports in the same box. Gigabit Ethernet uplinks have appeared as modular upgrades for fixed-configuration Fast Ethernet devices or modular, chassis-based hubs to provide a high-speed connection to the network. Vendors of high performance routers can be expected to deliver Gigabit Ethernet interfaces as well.
Some Gigabit Ethernet vendors have developed a new device called a full duplex repeater or buffered distributor. The full duplex repeater is a full-duplex, multiport, hub-like device that interconnects two or more 802.3 links operating at 1 Gbps or faster. Like an 802.3 repeater, it is a non-address-filtering device. The buffered distributor forwards all incoming packets to all connected links except the originating link, providing a shared bandwidth domain comparable to a 802.3 collision domain. Unlike an 802.3 repeater, the buffered distributor is permitted to buffer one or more incoming frames on each link before forwarding them.
As a shared bandwidth device, the buffered distributor should be distinguished from both routers and switches. While routers with Gigabit Ethernet interfaces may have backplanes that support bandwidths greater or less than gigabit rates, the ports attached to a Gigabit Ethernet buffered distributor’s backplane share one gigabit of bandwidth. In contrast, the backplanes of high-performance, multiport Gigabit Ethernet switches will support multigigabit bandwidths.
Gigabit Ethernet provides high-speed connectivity, but does not by itself provide a full set of services such as Quality of Service (QoS), automatic redundant fail-over, or higher-level routing services; these are added via other open standards. Gigabit Ethernet, like all Ethernet specifications, specifies the data link (layer 2) of the OSI protocol model, while TCP and IP in turn specify the transport (layer 4) and network (layer 3) portions and allow reliable communication services between applications. Issues such as QoS are not addressed in the original Gigabit Ethernet specifications, but must be addressed across several of these standards. RSVP, for instance, is defined at the network layer to work alongside IP. Layer 3 (routing) services also operate at the network layer (Table 6).
Various implementations of Gigabit Ethernet may include one or more of these standards in order to provide a more robust or functional networking connection, but the overall success of Gigabit Ethernet is not tied to any one of them. The advantage of modular standards is that any one piece may evolve and be adopted at a pace determined by market need and product quality. Note that all of the standards are just as readily paired with Fast Ethernet and 10 Mbps Ethernet, so that all levels of Ethernet performance can benefit from all the standards work.
Applications emerging in the mid to late 90's demand consistent bandwidth, latency, and jitter from network connections. Such applications include voice and video over LANs and WANs, multicast software distribution, and the like. Standards bodies have responded with new open definitions such as RSVP and the current work in the IEEE 802.1p and IEEE 802.1Q standards groups. RSVP is gaining industry acceptance as a preferred way to request and provide quality of network connections. In order to have RSVP function and deliver defined and consistent quality to an application, each network component in the chain between client and server must support RSVP and communicate appropriately. Because of the need to have so many components supported by RSVP before meaningful results can be achieved, some vendors are advancing proprietary schemes to deliver some degree of QoS. Some of these may deliver QoS benefits to users, but will require certain portions of the network to be vendor-specific implementations.
802.1p and 802.1Q facilitate quality of service today over Ethernet by providing a means for “tagging” packets with an indication of the priority or class of service desired for the packet. These tags allow applications to communicate the priority of packets to internetworking devices. RSVP support can be achieved by mapping RSVP sessions into 802.1p service classes.
Layer 3 involves determination of the eventual destination of a packet–beyond its MAC destination address on the packet header. By examining the IP address (buried deeper in the packet), the IP subnet can be determined, allowing broadcasts to be contained to the appropriate subnets and packets to be forwarded accurately to intermediate nodes for most efficient transit through the network.
The classic Layer 3 device is the router, which makes Layer 3 decisions by implementing complex algorithms and data structures in software. While such complicated routing tasks formerly required complex and software intensive multi-protocol router products, vendors over the last year have shipped Layer 3 switches or routing switches that accomplish many of these tasks, while delivering arguably better price/performance than traditional routers. Narrowing the protocol supported to IP has allowed these devices to optimize tasks and accomplish more work with dedicated hardware.