Successors of Ethernet

On a regular Ethernet segment, all stations share the available bandwidth of 10 Mbps. With the increase in traffic, the number of packet collisions goes up, lowering the overall throughput. There are two basic approaches to increase the bandwidth.

One is to replace the Ethernet with a higher speed version of Ethernet. Fast Ethernet and 100 VG-AnyLAN, two competing standards both operating at 100 Mbps belong to this category. Gigabit Ethernet has also emerged recently, extending the data rate to 1000 Mbps.

The other approach is to use Ether-switches that use a high-speed internal bus to switch packets between multiple (8 to 32) cable segments and offer dedicated 10 Mbps bandwidth on each segment/ port.

Switched Ethernet, fast Ethernet and 100 VG-AnyLAN are discussed here.

Switched Ethernet

Switched Ethernet gives dedicated 10 Mbps bandwidth on each of its ports. On each of the port one can connect either a thin segment, thick segment or a computer.

There are two techniques used in the implementation of Ethernet switches: store-and-forward and cut-through.

In store-and-forward, the entire frame is captured at the incoming port, stored in the switch's memory, and after an address lookup to determine the LAN destination port, forwarded to the appropriate port. The lookup table is automatically built up. On the other hand, a cut-through switch begins to transmit the frame to the destination port as soon as it decodes the destination address from the frame header.

Store-and-forward approach provides a greater level of error detection because damaged frames are not forwarded to the destination port. But, it introduces longer delay of about 1.2 msec for forwarding a frame and suffers from the chance of loosing data due to reliance on buffer memory. The cut-through switches, on the other hand, have reduced latency but have higher switch cost.

The throughput can be further increased on switched Ethernet by using full-duplex technique, which uses separate wire pairs for transmitting and receiving. Thus a station can transmit and receive simultaneously, effectively doubling the throughput to 20 Mbps on each port.

Fast Ethernet

The 802.u or the fast Ethernet, as it is commonly known, was approved by the IEEE 802 committee in 1995. It may not be considered as a new standard but an addendum to the existing 802.3 standard. The fast Ethernet uses the same frame format, same CSMA/CD protocol and same interface as the 802.3, but uses a data transfer rate of 100 Mbps instead of 10 Mbps. Fast Ethernet is based entirely on 10-Base-T, because of its advantages (Although technically 10-BASE-5 or 10-BASE-2 could have been used with shorter segment length).

Three signaling schemes have been proposed for fast Ethernet.

• 100 BASE-T4: This option is used for half-duplex communication using four wire-pairs of the existing category 3 UTP cable. It is also possible to use CAT 5 UTP or multi-mode fiber. Three pairs are used to transfer data in each direction and the fourth pair is used for collision detection. The hubs used are 100 BASE-T. The maximum distance between a station and hub is 100m.
• 100 BASE TX: This option uses two pairs of category 5 UTP, STP or multimode fiber for full-duplex data transfer. The maximum distance between a station and hub is 100m.
• 100 BASE FX: This uses two strands of multimode fiber, one for each direction for full duplex 100 Mbps data transfer in each direction. Here distance between a station and hub can be up to 2 km.

Gigabit Ethernet

As its name implies, gigabit Ethernet, officially known as 802.3z, is the 1 Gbps extension of the 802.3 standard already defined for 10 and 100 Mbps service. But, gigabit Ethernet is not simply a straight Ethernet running at 1 Gbps. In fact, the ways it differs from its predecessors may be more important than its similarities. Some of the important differences are mentioned below.

1. The cabling requirement of gigabit Ethernet is very different. The technology is based on fiber optic cable. Multi-mode fiber is able to transmit at gigabit rate to at least 580 meters and with single-mode runs exceeding 3 km. Fiber optic cabling is costly. In order to reduce the cost of cabling, the 802.3z working group also proposes using twinax (twin coax cable) for short-haul distances up to 30 meters.

2. Gigabit Ethernet also relies on a modified MAC layers. At gigabit speed, two stations 200 meters apart will not detect a collision, when both simultaneously send 64-byte frames. This inability to detect collision leads to network instability. A mechanism known as carrier extension has been proposed for frames shorter than 512 bytes. The number of repeater hops is also restricted to only one in place of two for 100 Base-T.

3. Flow Control is a major concern in gigabit Ethernet because of buffer overflow and junked frames in heavily loaded condition. The solution proposed by IEEE sub-committee is the 802.3x. The X-on / X-off protocol works over any full duplex Ethernet, fast Ethernet or gigabit Ethernet link. When a switch buffer is close to capacity, the receiving device signals the sending station and tells it to stop transmitting until the buffer becomes empty.

4. Finally one important feature, which Ethernet technology lacks is the Quality of Service (QOS). The gigabit Ethernet is a connectionless technology that transmits variable length frames. As such, it simply cannot guarantee that the real-time packets get the preferential treatment they require. The IEEE sub-committee is exploring specifications that will help Ethernet provide the required QOS.

100 VG-Any LAN

Developed by Hewlett-Packard and presently supported by over 30 companies, 100 VG-AnyLAN was adopted in early 1995 by the IEEE 802.12 sub-committee. Rather than using CSMA/CD medium access control, it uses demand priority protocol. It provides a 100 Mbps data rate with guaranteed bandwidth and bounded access delay for time critical applications. Basic features of 100 VG-AnyLAN are considered below.

Media: 100 VG-AnyLAN uses point-to-point link between a hub and a computer or another hub. Different links can be either 4-pair CAT 3 UTP, 2-pair STP not exceeding 500 m or 2-pair optical fiber not exceeding 2 km with a suitable mix.

Topology: The 100 VG-AnyLAN networks are centered around the concept of intelligent hubs, each of which is at the centre of a star topology. A hub has several local ports for interconnecting to individual workstations, servers, or lower level hubs. Optionally, it can also have one cascade port for connection to a higher level hub as shown in the figure above.

Medium Access Control: 100VG-AnyLAN uses Demand Priority Protocol, which combines the best characteristics of Ethernet (simple, fast access) and token ring (strong control, collision avoidance and deterministic delay).

Control of a demand priority protocol is centered in the hubs and is based on a request/ grant handshake between the hubs and their connecting nodes. Each node needing to send a frame indicates its need to transmit by sending a request signal to the hub and waiting for the hub to grant its permission to transmit its frame.

Hubs use a round-robin selection procedure to ensure that a fair opportunity for nodes to transmit data is available. Two priority levels - high priority and normal priority, are provided so that time critical traffic such as interactive voice, video, audio and multimedia can be given priority service with guaranteed low delay. This 802.12 standard can handle both Ethernet and Token ring frame formats, making it convenient for internetworking with them.