Table of Contents
| Table of Contents | ||
|---|---|---|
|
Physical Local Area Networks
Characteristics
Physical LANs are quite common in computer networks and have the following characteristics:
- Cover a limited geographic area
- Support high data rates
- Constructed with a reliable physical medium
- Are inexpensive to deploy
- Consist of a shared physical medium
- Support broadcast communications
- Store and forward and routing are unnecessary
There are many network protocols used in physical LANs, such as:
- CSMA/CD (Carrier Sense Multiple Access with Collision Detection): Ethernet, ALOHA
- Token ring: IBM, FDDI
Common Topologies
The data unit on a LAN is referred to as a frame (not a packet). When a frame is transmitted, it propagates over the entire physical medium. How a frame is forwarded depends on the LAN topology used. There are two popular LAN topologies:
- Bus
- Ring
Bus Topology
Each end host on a bus has a transceiver that transmits and passively monitors the bus for passing transmissions. Synchronization is performed on frame headers, which contain control information. In a bus topology, every end host connected to the bus receives the header of all transmissions on the bus. If an end host receives a frame header addressed to a different end host, the rest of the frame will be ignored.
Ring Topology
In a ring topology, each end host has a transceiver that breaks the circuit of the ring. Unlike a bus topology where each end host passively listens to the bus, each end host actively participates in the ring regardless of a frame’s destination. If an end host receives a frame with header addressed to itself, the end host will copy the frame from the ring; otherwise, the end host will forward the frame towards the next end host on the ring. In a ring topology, frames flow in a single predetermined direction.
Similar to cut-through switching, cut-through forwarding can allow an end host to begin transmission of a frame on its outbound transceiver before it receives the entire frame. If the end host determines after reading the frame header that the frame is destined for another end host on the ring, it can begin transmission.
The source end host of a frame will remove the frame from the ring and not forward it once it arrives back at the source host.
End Hosts
In a physical LAN, an end host contains a network interface card (NIC) for transmitting and receiving frames on the LAN. The NIC provides controlled access to the physical medium of the LAN. Thus, a NIC implements the physical and data link layers of the OSI model network stack or the physical, media access control (MAC), and logical link control (LLC) layers of the IEEE standards based network stack.
Media Access Control Techniques
There are a variety of methods for controlling access to the physical medium of a LAN when there are multiple end hosts sharing the LAN. These techniques are collectively known as media access control (MAC) and are highly dependent on the topology of the LAN. Oftentimes, the MAC technique can have a significant impact on the overall performance of the LAN.
Fixed Assignment MAC
A LAN with fixed assignment MAC pre-allocates fractions of the bandwidth capacity to end hosts. Two popular fixed assignment MACs are known as time division multiple access (TDMA) and frequency division multiple access (FDMA).
Time Division Multiple Access
In TDMA, each end host is pre-allocated a fixed duration time slot in which it (and only it) has the ability to transmit on the shared physical medium of the LAN. An example of TDMA is a shared T1 link where multiple telephone calls are multiplexed onto the single T1 link in predefined time slots.
Frequency Division Multiple Access
In FDMA, each end host is pre-allocated a unique frequency at which it can transmit data continuously. An example of FDMA is an analogue cellular telephone tower, which assigns connected cellular devices unique frequencies for conducting phone calls.
Note that in TDMA, if the traffic pattern of end hosts is bursty, many time slots may be wasted if the end hosts do not have any data to transmit in their predefined slots. Likewise, in FDMA, if the traffic pattern is bursty, end hosts may not utilize their allocated frequency all the time, resulting in wasted bandwidth. This is a problem inherent to fixed assignment MAC techniques.
Random Access MAC
The goal of random access MAC is to allocate the physical medium to end hosts based on demand. As such, the link capacity is fully allocated to all end hosts at any given time; however, note that only one end host can transmit on the physical medium at a time. Thus, in random access MAC, a mechanism is required to detect whether or not an end host is currently transmitting, as well as to detect and mitigate transmission collisions.
A collision is defined as the event when two or more end hosts transmit on a random access MAC physical LAN simultaneously, resulting transmission overlap perceived at any end host on the LAN. If a collision occurs in a random access MAC, all involved transmissions fail and must be retransmitted at a later time.
The advantage of random access MAC is the ability for an end host to utilize 100% of the link capacity as needed. However, the disadvantage is that if the demand for LAN access is high, many collisions and retransmissions can occur, resulting in a lower successful data transfer rate. Throughput is defined as the amount of data transmitted successfully (i.e. without collisions). If collisions are high, the throughput of a random access LAN can be a fraction of the total link capacity.
Ethernet is a widely used example of random access MAC. And, although not widely used in modern times, the ALOHA protocol is a classic example of random access MAC.
Ethernet
Ethernet is a CSMA/CD or Carrier Sense Multiple Access with Collision Detection protocol. As a random access MAC, it allows all end hosts to attempt transmission on the physical LAN at a given time. Probabilistically speaking, it is possible for more than one end host to begin a frame transmission at a given time. And, given the physical properties of the LAN, it is possible that multiple end hosts could begin transmission around the same time despite each detecting an idle LAN. To detect and mitigate such scenarios, Ethernet allows NIC transceivers to listen to the LAN while they are transmitting.
ALOHA
Unlike Ethernet, ALOHA does not provide the ability for an end host to listen to the LAN while transmitting. As such, when a collision occurs, it is not detected by any end host until after all colliding transmissions are complete.
Demand Assignment MAC
As the name implies, demand assignment MAC attempts to allocate the entire link capacity based on demand. It improves upon random access MAC by allowing a single end host to access the physical medium at any given time based on the amount of data the end host would like to transfer. There are two types of demand assignment MAC:
- Distributed control demand assignment MAC
- Centralized control demand assignment MAC
Distributed Control Demand Assignment MAC
In distributed control demand assignment MAC, a single token circulates between hosts on the LAN. Only the end host in possession of the token is allowed to transmit on the LAN. When an end host is finished with a transmission, it passes the token to the next host.
Centralized Control Demand Assignment MAC
In centralized control demand assignment MAC, a single token is allocated to end hosts by a centralized controller. Only the end host in possession of the token is permitted to transfer on the LAN.
The advantage of both distributed and demand assignment MAC is that the overall performance of the LAN is not sensitive to demand. However, the disadvantage is that end hosts must wait on the token in order to transmit. The amount of time spent waiting is undefined and could be long depending on the amount of data to be transmitted by the host in possession of the token. Thus, starvation is a possibility and unfortunate side effect of demand assignment MAC.
Characterization of Network Traffic
Traffic load and its characteristics can be a major contributor to the overall performance of a LAN. In an ideal network, all traffic would be predictable such that link allocation could be allotted perfectly with zero collisions, zero wait for link access, and 100% link capacity achieved. Unfortunately, in the real world, network traffic tends to be bursty. This is due to the behavior of applications and users that utilize the underlying network. For example, the release of a new television show by an online streaming provider could result in a temporarily large increase in traffic and network demand by watchers.
This bursty nature network traffic motivates the use of probabilistic models to drive the design and implementation of networks. Such models should be simple enough to evaluate for correctness, accurate enough to emulate real network traffic, and robust enough for use in a wide variety of real world scenarios.
Examples of Network Traffic
Digitized Audio/Video
In remote audio and video conferencing, audio and/or video is sampled and relayed over the network to participating parties. Pulse code modulation takes an analog signal and samples it at a particular rate and resolution to encode it digitally. This fixed-size digital interpretation of the audio/video is then relayed over the network at predictable intervals in order to facilitate real time communications.
Interactive Data Transactions
In ecommerce, users interact with remote web servers hosting content. For example, a user might browser different web pages searching for products, selecting more detailed views, which the web server then relays to the user on demand. This type of traffic pattern in irregular in time and size.