Dr. Dobb's Journal May 2000
In late 1995, the IEEE 802.3 committee formed the High-Speed Study Group to investigate means for conveying packets in Ethernet format at speeds in the gigabit-per-second range. A set of 1000-Mbits/sec. standards have now been issued.
The strategy for Gigabit Ethernet is the same as that for 100-Mbits/sec. Ethernet. While defining a new medium and transmission specification, Gigabit Ethernet retains the CSMA/CD protocol and frame format of its 10-Mbits/sec. and 100-Mbits/sec. predecessors. It is compatible with the slower Ethernets, providing a smooth migration path. As more organizations move to 100-Mbit/sec. Ethernet, putting huge traffic loads on backbone networks, demand for Gigabit Ethernet has intensified.
Figure 1 shows a typical application of Gigabit Ethernet. A 1-Gbit/sec. LAN switch provides backbone connectivity for central servers and high-speed workgroup switches. Each workgroup LAN switch supports both 1-Gbit/sec. links, to connect to the backbone LAN switch and to support high-performance workgroup servers, and 100-Mbits/sec. links, to support high-performance workstations, servers, and 100-Mbits/sec. LAN switches.
Figure 2 shows the overall protocol architecture for Gigabit Ethernet. The MAC layer is an enhanced version of the basic 802.3 MAC algorithm. A separate gigabit medium-independent interface (GMII) has been defined and is optional for all of the medium options except unshielded twisted pair (UTP). The GMII defines independent 8-bit-parallel transmit and receive synchronous data interfaces. It is intended as a chip-to-chip interface that lets system vendors mix MAC and PHY components from different manufacturers.
Two signal encoding schemes are defined at the physical layer. The 8B/10B scheme is used for optical fiber and shielded copper media, and the PAM-5 is used for UTP.
The 1000-Mbits/sec. specification calls for the same CSMA/CD frame format and MAC protocol as used in the 10-Mbits/sec. and 100-Mbits/sec. versions of IEEE 802.3. For traditional Ethernet hub operation, in which only one station can transmit at a time (half duplex), there are two enhancements to the basic CSMA/CD scheme:
With a LAN switch (full-duplex operation), which provides dedicated rather than shared access to the medium, the carrier extension and frame-bursting features are not needed. This is because data transmission and reception at a station can occur simultaneously without interference and with no contention for a shared medium. All of the Gigabit products on the market use a switching technique, and so do not implement the carrier extension and frame bursting.
With a switching technique, full-duplex operation is employed, and the CSMA/CD protocol is not needed. The Gigabit specification expands on the pause protocol that is defined for 100-Mbits/sec. Ethernet by allowing asymmetric flow control. Using the autonegotiation protocol, a device may indicate that it may send pause frames to its link partner but will not respond to pause frames from its partner.
The current 1-Gbit/sec. specification for IEEE 802.3 includes the following physical layer alternatives (Figure 3):
The encoding scheme used for all of the Gigabit Ethernet options except twisted pair is 8B/10B. This scheme is also used in Fiber Channel. With 8B/10B, each 8 bits of data is converted into 10 bits for transmission. The 8B/10B scheme was developed and patented by IBM for use in its 200-megabaud ESCON interconnect system.
The developers of this code list the following advantages:
The 8B/10B code is an example of the more general mBnB code, in which m binary source bits are mapped into n binary bits for transmission. Redundancy is built into the code to provide the desired transmission features by making n>m. Figure 4 illustrates the operation of this code. The 8B/10B code actually combines two other codes, a 5B/6B code and a 3B/4B code. The use of these two codes is simply an artifact that simplifies the definition of the mapping and the implementation; the mapping could have been defined directly as an 8B/10B code. In any case, a mapping is defined that maps each of the possible 8-bit source blocks into a 10-bit code block. There is also a function called "disparity control." In essence, this function keeps track of the excess of zeros over ones or ones over zeros. An excess in either direction is referred to as a "disparity." If there is a disparity, and if the current code block would add to that disparity, then the disparity control block complements the 10-bit code block. This has the effect of either eliminating the disparity or at least moving it in the opposite direction of the current disparity.
The encoding mechanism also includes a control line input, K, which indicates whether the lines A through H are data or control bits. In the latter case, a special nondata 10-bit block is generated. A total of 12 of these nondata blocks are defined as valid in the standard. These are used for synchronization and other control purposes.
For 1000BASE-T, the encoding scheme used is PAM-5, over four twisted pair links. Therefore, each link must provide a data rate of 250 Mbits/sec. PAM-5 provides better bandwidth utilization than simple binary signaling by using five different signaling levels. Each signal element can represent two bits of information (using four signaling levels), plus there is a fifth signal level that is used in a forward error correction scheme.
The Gigabit Ethernet Alliance: http:// www.gigabit-ethernet.org/.
Frazier, H., and Johnson, H. "Gigabit Ethernet: From 100 to 1,000 Mbps." IEEE Internet Computing, January/February 1999.
DDJ