Gigabit Ethernet

From Hill2dot0
Jump to: navigation, search

Continued advancements in LAN applications, the workstations that run them, and the nature and amount of communication between servers and workstations demanded greater speeds in the networks that serve them. By the late 1990s, Fast Ethernet networking had become a common desktop technology, and even faster backbone service was required.

Enter the Gigabit Ethernet (GigE) standard. Essentially an extension of 10BASE-T and 100BASE-T technologies and defined originally in IEEE 802.3z and IEEE 802.3ab, Gigabit Ethernet is capable of both half duplex and full duplex transmission modes; it is almost exclusively found, however, as a full duplex implementation. The family of GigE Physical Layer standards, collectively referred to as the 1000BASE family, support the use of both unshielded twisted pair (UTP) and optical fiber, and add a new media: twinaxial cable.

Gigabit Ethernet retains the traditional IEEE 802.3 frame type, which enables its simple integration into existing 10BASE-T and 100BASE-T networks. Initially, the physical plant used the same physical signaling as Fibre Channel, although later extensions added new signaling strategies to enable the use of UTP and twinax.

GigE Topologies

GigE Topologies

The 1 Gbps version of the Ethernet standard defines two topological modes—shared media and dedicated media. The distinctions between the two are detailed below.

Shared media

  • One collision domain per hub
  • Shorter media because of the collision domain
  • Half duplex operation only
  • If star-wired, uses a repeater

Dedicated media

  • One collision domain per segment port
  • Longer media options
  • Half duplex or full duplex options
  • If star-wired, uses a switch

Although both options are supported in the standard, modern GigE products are almost universally based on the dedicated media model and implemented as full-duplex.

Shared Bandwidth GigE

As with 100BASE-T, a great degree of the popularity of the Gigabit Ethernet idea is in its very name. Even though the earliest products might not bear much of a physical resemblance to the Ethernet that many data communications professionals know and love, it still uses the Ethernet frame structure and so is perceived as Ethernet at heart. Having the name “Ethernet” associated with a technology is almost a guarantee of its success, and GigE has indeed succeeded.

For some, however, it is not the frame structure that defines Ethernet; it is the CSMA/CD MAC scheme. To maintain this link, a lot of work was done to support the shared bandwidth model of Ethernet in the newer high-speed variant. Unfortunately, the problem of collisions and their impact on total network size returns. With Fast Ethernet, the designers had to choose between increasing the minimum frame size and decreasing the size of the network. To preserve backwards compatibility with the original Ethernet, the network shrank by an order of magnitude, to maximum network sizes of 200-800 meters.

Repeating this choice at 1 Gbps rates is not an option. It would translate to maximum network sizes of 20-80 meters, and note those are station-to-station, which would mean 10-40 meters cable limits for shared-media environments. Extending the frame size was also not an option because it would sacrifice backwards compatibility.

To avoid both these pitfalls, GigE introduced the technique of carrier extension. This means that if a small frame has been transmitted, the transmitter continues to transmit a signal so that the maximum collision window is covered. This “carrier extension” is effectively dead time, or additional overhead and reduces network efficiency.

To overcome some of the effects of underutilizing the bandwidth, due to limitation of Ethernet's maximum frame size, as well as the overhead of carrier extension, a mechanism called frame bursting was introduced. Once the first frame has been successfully sent (this could be a small frame with carrier extension), subsequent frames can be sent immediately—up to a limit of 65,536 bits (8 KB). Each series of frames sent in this manner is separated by a small gap consisting of the carrier extension pattern; this ensures that other stations on the network will know that a series of frames is being sent.

Most implementations of Gigabit Ethernet use switches for hubs and are operating in full-duplex mode, the same as full-duplex 10 and 100 Mbps versions. The minimum and maximum frame still remains the same. There has been talk over the years of standardizing “jumbo frames” for use with Gigabit Ethernet. This would better utilize the bandwidth and place lower loads on servers and switches. The IEEE has formed the Frame Expansion Study Group, which may take this into consideration; however, there is a great deal to discuss yet, and no standard has emerged.

GigE Media Options

GigE Media Options

IEEE 802.3z specified three physical options.

  • 1000BASE-LX uses a long-wave laser over either multimode fiber (MMF) or single-mode fiber (SMF) to achieve distances up to 5 kilometers. It would most commonly be found in a campus backbone or a metropolitan setting.
  • 1000BASE-SX uses a short-wave laser over MMF to achieve distances of 275 meters or 550 meters, depending on the type of MMF used. It might be used for fiber-to-the-desktop or up a vertical riser in a building.
  • 1000BASE-CX uses a specialty shielded balanced copper twinaxial jumper cable that was primarily intended for use within a data center or a wiring closet.

IEEE 802.3ab standardized 1000BASE-T, which uses an advanced multilevel signaling technique over four pair of Cat 5 UTP (or better) to a distance of 100 meters, as specified in ISO/IEC 11801:2000. It has penetrated the industry widely, and there are now many 10/100/1000 autonegotiating LAN adapters on the market.

As an aside, the use of the X in the higher speed IEEE 802.3 Physical Layer specifications indicates that part of the specification was derived from another technology. In the case of Gigabit Ethernet, the technology for the 1000BASE-TX standard was appropriated from Fibre Channel. In 100BASE-TX, it was borrowed from the specification for Fiber Distributed Data Interface (FDDI) over twisted pair.

Although these are the standardized media and distances, several vendors now offer versions of Gigabit Ethernet that operate over significantly longer distances using SMF. For example, Cisco’s 1000BASE-LX and 1000BASE-LH can extend to as much as 10 kilometers of standard SMF, while 1000BASE-ZX can extend to as much as 100 km over dispersion-shifted fiber (DSF). This means the equipment at either end of the link must either be from that vendor, or that vendor must provide an adapter or repeater than can convert to one of the industry standards to interconnect with a different vendor’s equipment.

GigE and Jumbo Frames

As Ethernet speeds have increased, some of the initial design decisions have come into question. The problem of minimum-size frames and the relative size of the network is virtually eliminated by the pervasive move to switched networks. However, now attention begins to focus on the other extreme: the maximum frame size.

Consider a device, like a server, that is receiving a steady stream of maximum size files because it is constantly engaged in file transfer activities. Most file transfers result in multiple maximum-size frames followed by one smaller one. On a 10 Mbps Ethernet, ignoring interframe spacing and assuming no collisions, the server is receiving at least 820 frames per second (820 frames x 1518 bytes per frame x 8 bits per byte is approximately 10 million bits). If any of the frames are smaller, the frame count will increase (assuming constant traffic).

When we go to 100 Mbps and 1 Gbps, this frame rate increases to 8200 and 82,000 respectively. The problem is that the server is taking an interrupt for every frame to process it. Imagine someone ringing your doorbell 82,000 times per second and ask yourself how much work you'd get done!

Increasing the maximum frame size would reduce the total number of received frames, reduce overhead, and render the attached systems and the network more efficient. These are the benefits being sought, but as of yet there is no standard for such frames. There is, however, a lot of discussion around such a standard, and there are products that support jumbo frames (as these are known) up to 9000 bytes. It is generally agreed that this is the upper bound without changing the CRC. With a 32-bit CRC, frames are protected up to about 12,000 bytes, after which the CRC loses effectiveness.

Such an increase, even when standardized, will wreak havoc with older LAN environments. It will only be useful in environments in which the entire new Ethernet environment supports the jumbo frames (e.g., a newly installed SAN), or in which the jumbo frames can be isolated from the regular frames using something like virtual LAN (VLAN) technology.

A final word about language: this is not the first time the “larger frame” issue has arisen. As the technology has matured and additional function has been added to LANs, some of these additional functions (e.g., VLANs) require the elements in the network to be able to add information to an Ethernet frame as it moves through a switched network. Although the information might be stripped before it reaches the destination, the switching nodes inside the network might be faced with abnormal-size frames if the initial frame was already at its maximum size. Such frames are not considered jumbo frames; they are only slightly above the traditional maximum Ethernet frame size. The term “baby giant” has been coined to refer to these.

See Also


<mp3></mp3> | Gigabit Ethernet (GigE)