Lecture 13: Multiaccess Networks

Network Technology

Revision:

The Internet is composed of many interconnected networks and/or subnets.
Networks and subnets are interconnected by routers.

In today's lecture we investigate the technology of networks -- that is, the technology used to physically connect a group of computers. Most networks (sometimes called LANs) have the following characteristics:

Usually employ some form of shared medium -- the networked systems all "see" one another's transmissions. This is much less true than it once was, but the core technology is designed around the assumption of a shared medium.
Data rates are high, usually at least 10Mbps (ie, 10 million bits per second -- data is always transmitted "serially", or one bit at a time), commonly 100Mbps, occasionally 1Gbps.
Privately owned; no involvement of public telcommunications providers, usually limited geographical area such as campus, business premises, etc.

Ethernet/802.3

Ethernet (also known as IEEE 802.3 and ISO 88023) is the dominant LAN technology at present. Ethernet defines a Medium Access Control (MAC) technology for operation over various types of cabling in a bus architecture.

Until recently^[1], the predominant form of Ethernet used so-called thin wire coaxial^[2] cabling. A typical installation might have looked like:

Whilst this is no longer the most common hardware used for Ethernet/802.3, it exemplifies the shared medium idea that it is based on. Note the use of "T-connectors" on each computer, and the "Terminators" at each end of the network.

^[1] OK it's not so recent. Basically this cabling technology was no longer being installed by about the mid-1990s.
^[2] Historical note: this was also called "cheapernet" or "10Base2". Exercise: cheaper than what?

CSMA/CD Medium Access Control (MAC)

This is an algorithm (implemented in the network-card hardware) which defines how computers using Ethernet/802.3 can efficiently share a common medium: Carrier Sense, Multiple Access, with Collision Detection.

Multiple Access: All computers have equal access: there are no masters or slaves, etc. If the shared channel is clear (ie, not in use), a computer may begin to transmit immediately.
Carrier Sense: If the channel is busy, continue to "listen", and attempt to transmit as soon as it becomes available.
Collision Detection: If a collision is detected (ie, another computer started to transmit at the same time, so the signal is garbled), immediately cease transmission. Wait a random period of time, then start all over.

Ethernet/802.3 Repeaters

A group of computers connected to a thin-wire (coax) Ethernet cable was called a segment, thus:

A thin wire (coaxial cable) segment had a maximum length of 185 metres, plus a few other rules as to numbers of machines, and the distance between them. The 185m segment length can be extended using a repeater^[3], which behaves somewhat like a bi-directional amplifier. The resulting extended LAN acts like a single larger segment, and is referred to as a "collision domain", because every computer in each of the component segments still "sees" everyone else's tranmissions.

^[3] a maximum of 4 repeaters was permitted between any two stations on the network.

Twisted Pair (10BaseT) Ethernet

For various reasons^[4], this is now the preferred Ethernet technology. In this system, stations are "star-wired" to a central hub, using a high-performance 4-pair twisted pair cable, thus:

The hub acts as a repeater, so whilst this superficially looks like a group of point-to-point links, all stations still "see" each other's transmissions, just the same as the coax-cable bus topology. In other words, it's still a "shared medium" network, and the attached computers still occupy a single collision domain.

The cable is usually "Category-5" (always abbreviated to "Cat5") unshielded twisted pair (UTP) or better, although the older, lower performance Cat-3 cable is also sometimes encountered. In practice, there are various hardware technologies involved in a typical installation: in-wall fixed wiring, patch cables, patch panels, etc. There are also strong limits on "cable-run" lengths.

^[4] We will discuss these reasons in the tute.

Ethernet/802.3 Frames

Data on an Ethernet is transmitted in frames:

Preamble: 7 bytes of 0101010101... This is used to synchronise the receiver.
Start Of Frame: 1 byte, thus: 01010111.
Source and Destination Address: each 6 bytes (48 bits!), and are uniquely assigned by IEEE. This is called a station's MAC address (or MAC-level address). All stations on a segment examine the destination address of all frames to see if it was addressed to them.
Type field: indicates which higher-level protocol created this frame, eg 0x0800 for IP. In 802.3 (rarely used to carry IP packets) this field gives the length (in bytes) of the data field.
Data field: between 46 and 1500 bytes of data. NB: minimum frame size is thus 64 bytes. The data field usually contains an IP datagram.

Ethernet/802.3 Switches

An Ethernet switch has a similar function to a hub (see earlier) -- switches are sometimes called "switching hubs".

The difference is that a switch examines the MAC-level destination address of every frame it receives, and transfers it directly to the appropriate port, without other ports being involved in, or aware of, the communication. Many such transfers can occur simultaneously, which has the effect of increasing overall "system" throughput. Typical switches can also do full-duplex transfers, simultaneously transmitting and receiving frames.

Note:: A switch builds a table mapping source addresses to ports which it subsequently uses to make switching decisions. It's obvious that switches are significantly more complex than simple hubs, and this is reflected in their price. However, as with all technologies, the difference is rapidly diminishing.

Philsophical note: it's interesting that switches essentially remove the possibility of collisions. We should therefore imagine that network hardware is no longer executing the CSMA/CD protocol, because a host can send a frame at any time. This is ony partly true -- collisions can, in fact, still occur, and the important point is that an Ethernet switch can still interoperate with older, non-switching hardware. The retention of the same frame format is an important aspect of this interoperability.

Address Resolution Protocol (ARP)

Recall that IP datagrams are forwarded over networks from host to router, router to router and router to host, and that every host and router has (at least one) unique IP address. From today's lecture, we also see that they also have a separate, unique MAC address -- used to address frames in the network.

The Address Resolution Protocol provides a mapping between IP addresses and MAC addresses. For example, in the case of IP local delivery (see earlier) the router or host knows the IP address of the destination machine, and also knows (from examination of the network/subnet address) that it is connected to the same network. In order to deliver the datagram, it:

First broadcasts an ARP-request. In effect, the broadcast asks (eg) "Which one of you guys is 149.144.21.60?".
The host whose IP address is 149.144.21.60 replies with its MAC address.
The datagram is then encapsulated into a frame with the correct destination MAC address and placed "on the wire". The destination system notes its own MAC address as the frame's destination and picks up the frame, delivering the datagram to the IP software.

Systems which use IP keep an ARP cache of recent IP-to-MAC mappings to avoid the need for repeated ARP-requests. This "ARP Table" can usually be examined by the system manager. ARP is considered to be the "last hop" routing protocol for IP packets.

Newer Technologies

100 Mbps Ethernet (100baseT)

so-called "Fast Ethernet".

Modern switches have at least one 100BaseT port, and 100BaseT hubs are also invariably switches.
Modern "Ethernet cards" are usually auto-detecting "10/100Mbps" and will run at 100Mbps if possible.
"Full-duplex" systems can simultaneously send and receive at 100Mbps.
Requires Cat5 cable as a minimum, and is commonly used over dual fibres instead of UTP, giving a distance advantage.

Fibre Distributed Data Interface

FDDI operates at 100 Mbps. It has been the Big New Thing for more than a decade, but has never been widely adopted due to its complexity and fiendishly high cost. Dead technology now.

Asynchronous Transfer Mode (ATM)

This is a system which allows integrated voice/video/data networks, currently at bit rates between 25Mbps and 625Mbps, with the most common version running at 155Mbps. Complex and expensive, but becoming very popular for "campus-wide" networks -- eg, La Trobe's microwave network is actually an ATM LAN.

Gigabit Ethernet

A variation which is compatible with 10 and 100 Mbps Ethernet, but runs at 1000Mbps. Still very expensive (original version only ran over fibre, for example) but will probably become the dominant LAN technology this decade.

Note: we have not covered token ring LANs in this lecture. Maybe see the assignment topics if interested.