Linking the Components
Chapter 4 does a good job with the basics. Please get any questions asked and answered in class...
- The bus is the main 'linker of components' in a computer.
Depending on the platform at hand, the bus might be a many-conductor
ribbon or braid of wires, or it might be parallel 'traces' on one of
the layers of a printed circuit board.
Adapter cards plugged into a PC's PCI slots, for example, are connected to the same bus as the Intel or AMD CPU likely to be on the board. We're used to today's ubiquitous PCI, ISA, & EISA busses in PCs.
Other busses are adapted for a purpose. The small 104-conductor PC104 bus makes it easy to make wearable or embedded processors that run Linux or Windows and put them in a butt back. All standard keyboard, vga, and mouse adaptors, plus ide for hard disk/cd/dvd, Ethernet, and a cluster of serial ports can be included on one of these cards. If more room is needed, another card, perhaps manufactured to suit some device in which the unit is embedded, can be attached to the 104-pin bus at the edge. These can be programmed for practically any application. OS & application software can be burned into a large ROM, and RAM provided for user's data. Or, a large, non-volatile, 'silicon hard drive' can be substituted for a real, spinning, hard drive at the IDE interface.
A minicomputer or mainframe may have an extended, 'multi-bus' architecture. The bus may be called a backplane, and it will will accept several circuit boards. Each may be larger than a PC's mainboard, with one popular 'form factor' about the size of a pizza box. Here the 'CPU board' (or boards) may have several microprocessors on it. Other boards han dle the 'channels' for terminal i/o, disk i/o, tape i/o, an d networks. This makes it possible for a CPU components to be 'hot swappable', and is a feature in 'highly available' & 'fault tolerant' systems.
- Logical & Physical I/O
Hi-level programming languages make I/O s imple. The programmer just says 'open' a file by name. The n the program can 'read' or 'write' records in the file as needed, each of these statements supplying a file name & a variable holding the record, and 'the computer' does the rest to get data from or put it onto the hard disk.
This is the 'logical i/o'. The 'physical i/o' is transparent to us as programmers and users of computers.
The hard disk is right primitive in its programming, although today's IDE drives are genius compared to the few earlier generations. The basic things it can do are seek a cylinder, switch to a surface, & read or write a sector.
IDE drives also supply their settings for the mainboard automatically, format their selves, manage their own bad sectors, and report their health via SMART -- a bit beyond the primitive instructions for data access.
The OS code in a particular platform has 'access methods' to supply the 'primitive' instructions to disk or tape so that the application program doesn't have to be rewritten for every new device that might be added to the platform's file systems, or for every platform where the software will be deployed.
Access methods resolve a file's directory (folder) entry into the physical (cylinder, surface, sector) address for the beginning of the file being accessed. Most file operations use this address along with the 'relative record' concept introduced earlier to allow sequential & direct access to records in the file.
Beyond the starting sector, the file may be scattered across the surfaces of the disk drive. A data structure similar to a linked list is used by the disk drive so that it can always find the next record in the file. Windoze users are familiar with this concept if they understand what the 'defrag' process is doing.
- Messages & Signals are general terms that applies to many CPU
and network devices.
The 'signal' part of the essential pair is usually fluctuating (modulated) electrical current or light.
All signals used in computers & networks today take time to move thru their medium (the speed of light) and they weaken (attenuate) with distance. So the length of a circuit might be quite short where nanoseconds are important. Electrical current moves about a yard in a nanosecond, so a CPU bus length is severely limited. A signal on a copper wire becomes too distorted for an Ethernet at about 100 meters, so the local connections to equipment on a LAN are limited.
'Noise' exists in all environments. When the signal attenuates to the point where it can't be picked out from the 'ambient noise' it becomes useless. This is a good thing for most WiFi LANs, a bad thing for a company trying to link distant facilities with RF.
In computing, 'message' implies some kind of formatting or framing. This is evident in the 'header', 'body', and 'trailer' found in the messages an OS or a node in a network receives.
The text demonstrates this in a CPU at figure 4.12, where messages in a multi-processing environment are shown among the application software, the OS, and the channel control unit (probably DMA in a PC). These messages are passed between CPU and the channel's controller as electrical signals on the bus.
Several of the traces on the bus are used to carry 'interrupt' signals from various i/o controllers to the CPU. Since devices may share a limited number of interrupt traces, these signals are formatted so that the CPU knows from where the interrupt originated.
-
'Packetization' is a word that describes the way ethernets and internets
prepare data to move to another MAC or IP address.
It is the process of bundling data into packets according to a specific protocol.
There are scads of protocols, but
Ethernet and
Internet provide most of
our connections these days.
Ethernet packets and internet packets fit together like hand and glove, so well that we percieve a 'stream' of data (with occasional glitches) as a voice or video stream. Neither Ether- nor Inter-nets carry streams of data like dedicated telephone or data circuits do. They dice the stream up to fit into 'packets'; calculate a 'checksum' that describes each packet's 'load'; add the checksum to the packet with other important stuff like destination and source address; dispatch the packet to the destination over the network(s). The network OS at the destination receives the packet, calculates the checksum on the load it receives, if the checksum agrees it accepts the load else it asks for the packet to be retransmitted. (Error Correction thru Re-transmission)
On an Internet, a stream's packets may follow the same, or different, routes across the 'net. On the other end, the packets are assembled in sequenced as they arrive, error-checked, and accepted if correct or re-requested if there is an error. We expect, and get, error free transmission over our networks, or the data doesn't flow...
Here's the message format for a packet of data on an Ethernet. The data is contained in the body of a packet, addressing data in the header, and a CRC checksum used for error detection follows in the trailer. Internet is similar, but different...
- A LAN is what most people think of when 'network' is spoken.
Most PCs in a business, and many at home, are connected to a Local
Area Network. The dimensions of a typical, CAT5 wired, LAN
are governed by the length of wire (100 meters in an Ethernet), how
many repeaters can be used to connect segments of the LAN (a couple or
three), and the requirement that the ground sources of the electrical
supply for all the equipment be the same.
On a campus-styled site, individual buildings may be on different electrical services. There fiber optic, or RF is needed in place of the CAT5's copper conductors. Sometimes, lines leased from the telephone company are the best choice in a setting like this.
Where facilities are some miles, or a half-world, apart, enterprises may operate their own, private WAN (Wide Area Network) using lines leased from 'long line' carriers, bandwidth leased on satellites, or their own radios.
Today, VPNs (Virtual Private Networks) offer an economical alternative by routing portions of an enterprise's WAN over The Internet. VPN equipment & software encrypts data so they can travel over The Internet and still be private. An ISP like Global Crossing or one of the other 'backbone providers' to help ensure the privacy of the VPN.
- Check the text, or the board in class, for diagrams of the basic
network topologies: Serial, Bus, Star, Hierarchical or 'tree', Ring, and Meshed.
Network topology also applies across a wide range of physical sizes. For example, 'bus' topology can be found in the microscopic realm within a CPU chip, as a component inside a ethernet hub, or in a room-sized space where a coaxial cable (a kind of bus) wraps around the walls behind desks.
Serial cables attach one device to another -- there is no choice of route, and in many cases there is no 'address' involved except the number of port where the cable is physically attached.
Bus: For many intents and purposes, a Radio Frequency acts similar to a bus and can be global in scale. The thing they all have in common is that all the devices are attached to a common medium (copper or optical cable, or RF). In a wired bus, all the attached devices 'listen to' the signals on the media and each processes only those messages addressed to it. On a radio frequency, some devices might not be able to 'hear' other devices on the frequency due to attenuation, so extra bandwidth is required to manage the traffic on the frequency. Wired Ethernets use CSMA/CA for traffic management, where the 'CA' means 'collision avoidance'.
A typical Ethernet LAN using a hub or switch is an interesting example where bus and star topology both apply. All Ethernets use a bus, and there is a little bus, or backplane, _inside_ the hub's or switch's chassis that connects all the ports, . A switch's bus can be thought of as an 'intelligent bus', or 'switching matrix' that only carries signals to the device with the appropriate MAC address.
Variations on Busses The first Ethernets used a bus, a thick, coaxial cable (etherhose) which was run around the perimeter of the LAN. Devices were attached by driving a sharp spike through the cable's jacket into the conductor at the center of the cable & it was possible to add & remove nodes from the LAN without disrupting it. This required a high degree of skill to make a good connection.
Later attempts at making the connections easier involve installing 'barrel network connectors' in place of the sharp spike. But, these require that an entire network segment be disabled to add or remove a node. Also, the BNC connectors aren't very stable and these 'thinnet' or 'lantastic' networks were a pain to keep working.
Ethernet is a good idea, but it's best to put the bus inside the chassis of a switch or hub and use the convenient CAT5+ cables with their RJ45 connectors in a star configuration to attach nodes to the network. This way, individual nodes can be added and removed from the network without disrupting the LAN.
Star: In any star network there is a central device, like a hub, switch, or host computer and devices are attached to it by cables. In these networks, each attached device only 'hears' signals which are sent to it by the device in the center.
If the device at the center of the star is a simple 'repeater', like a hub, all messages it receives are instantly repeated throughout the star -- although any of the nodes can 'hear' the messages they ordinarily only capture those addressed to it. (Easy for a cracker to 'sniff') If the device has more intelligence, like a switch or a minicomputer, each device will only 'hear' messages dispatched to it from the center of the star, which makes network traffic easier to keep secure.
Hierarchical or Tree A hierarchical network is best demonstrated where hubs or switches may be 'uplinked' or 'cascaded' to extend the number of nodes on a LAN or its size. This is sometimes referred to as a tree structure.
Ring topology is not as common as the others today for LANs. But, some of today's fastest and most reliable networks use it. FDDI (Fiber Distributed Data Interface) rings can be physically larger than copper-based LANs, up to a hundred miles or so. Ring-based networks can guarantee that nodes get the data rate needed by using 'token passing' schemes to manage network traffic. Each node waits for the 'token' to reach it before it transmits its packet. The network can be sized so that 'quality of service' can be guaranteed for the applications on the network.
In contrast, wired Ethernet uses a 'collision detection' scheme to manage network traffic. (CSMA/CD) When a node has a packet to transmit, it 'listens' for traffic on the LAN and if it is clear it transmits. If another node (or other nodes) happens to transmit at the same time, the resulting 'collision' is detected, a 'jamming signal' is transmitted, and each of the nodes 'backs off' for a random interval before trying to transmit again. An improperly sized Ethernet will just stop working if it becomes overloaded.
'The box' that makes a token ring LAN is not called a switch or hub, it is a MAU-Media Access Unit. Similar in appearance to a switch, the MAU makes the token ring 'inside the box' and the wiring topology is a star.
Ethernets have replaced many Token Ring networks. IBM's Token Ring was very popular in terminal & PC networks connected to IBM mainframes and mid-range computers. But, today's mainframes and mid-range machines work fine with Ethernet, which has the benefit of lower cost and good performance.
Meshed network topology is common for clustering servers, storage devices, and big IT infrastructure components like Cell Phone. An example of a meshed network would be a cluster of servers and network attached storage on a fast Ethernet (bus/star) for moving data and there is also a Serial network that carries a 'heartbeat' used by the clustering software to quickly identify when a server crashes so it can start up its services on another machine in the cluster. Also, Cellphone towers are meshed: They use one network to connect to our Cellphones and another network, serial lines or highly-directional microwaves, to coordinate among the towers so that our conversation can move seamlessly between towers.
- 'Internetworking' is connecting networks to other networks.
The biggest example is, of course, The Internet. Internetworking
has never been easier or less expensive.
'Bridges' (OSI Layer 1 or 2) may be WAN equipment used to connect similar, or dissimilar, LANs even over wide distances. Or, a bridge may be used to connect different types of networks together in one location. For example, an AppleTalk/Ethernet bridge allows PCs with Ethernet NICs to share data and resources with older Apples using Apple Talk. The important thing about a bridge is that it only has one network at each end.
'Routers' (OSI Layer 3) are LAN or WAN equipment that forwards data packets among networks, which may be of different types. A small router is typically connected on the 'inside' to a port on a LAN's switch or hub and to the 'outside' to a private network provider or an ISP via a DSL or some other media. A router used this was is a 'gateway' for LAN traffic to the WAN or The Internet.
Large, 'industrial strength' routers may interconnect many networks. Many routers' interfaces are 'generic' AUI ports that accept 'media converters' so they can connect to any media, and their OS allow them to support many common networking protocols.