06.621 Data Communications and Networks 2
Week 5 – Line Utilisation Strategies

Introduction
In the early days of remote terminal access, a remote terminal used a pair of modems to communicate across a public telephone line. Initially a pair of modems enabled a single remote terminal to access a mainframe computer. The introduction of multiplexers allowed several terminals to share a single telephone line. More recently the terminal servers provide an alternate means of connecting multiple terminals to a host computer using a single telephone line.

For local terminals, the connection to the host is from the RS-232 port of the terminal through to either a front end processor on the host or a board that accept multiple RS-232 lines into the host.

Remote Terminals have a more complex arrangement and rely on the use of modems to send digital data from the host computer and the terminal across a voice grade communication line.

The diagram shows the basic hardware configuration for a data communication network, starting at the host computer end and working out to the remote end. Other hardware not shown in the diagram, include multiplexers, intelligent controllers, protocol converters and line adaptors.

The Host Mainframe Computer 
The host mainframe computer generally is considered to be the central computer or central processing system. In distributed processing, several host computers may be tied together by the data communication network. Although the host computer is not truly part of the network, it performs many network functions because these functions may be shared by the host computer and front end processor.

Front End Processor (FEP) 
The front end processor can take two forms. The first is a non-programmable, hardwired. communication control unit designed by the manufacturer to adapt specific line and terminal characteristics to the computer. The second is a front end processor that is programmable and can handle some or all communication input/output activity as well as perform some processing. The general rend today is to remove every data communication task that you can from the host computer and move it out further into the network.

The basic components of the front end processor are as follows:

• Channel Interface is the hardware interface that permits the communication processor to connect directly to the standard channel of the host computer.
• Software is the highly specialised set of stored programs that define the specific architecture/protocols of the front end. The software determines which standard protocol should be used for the front end. Some of this software is built into firmware, that is the programs are coded into the circuit chips instead of being programmed logic.
• Line interface units (also called ports) are hardware devices used to link the communication processor with the modems that terminate each communication circuit.

The functions of the front end processor include:

• Communication line control.
• Polling/selecting of individual terminals, intelligent terminal controllers, or network nodes. Polling is asking each terminal if it has a message to send, and selecting is asking each terminal whether it is in a condition to receive a message.
• Automatic answering, acknowledgment, and dialling for outgoing calls.
• Port selection, which allows several circuits to share a single port. The port is the plug or connection point where individual lines enter the front end processor.
• Ability to address messages to specific terminals
• Circuit switching, which allows one incoming circuit to be switched directly to another when it is available.
• Automatic routing of messages to a backup terminal when a specific terminal or circuit is out of order.
• Addition or deletion of communication line control codes. Line control codes are the grammar of data communication, such as the END OF BLOCK, BEGINNING OF BLOCK and the START OF MESSAGE., must be deleted before the message is passed to the host computer and must be added before the message id passed to the communication circuits.
• Protocol/Code Conversion - code conversion, that is, the software or hardware conversion from one code format to another such as ASCII to EBCDIC. Conversion from one protocol to another, which allows different machines to talk to each other if they all use different protocols such as HDLC, SDLC,and X.25.
• Assembly of Characters/Messages - Assembly and disassembly of bits into characters. Bits are transmitted in serial fashion on a communication circuit. The FEP assembles these into parallel characters for the host computer to accept.
• Assembly/disassembly to handle synchronous or asynchronous modes of transmission.
• Handling of transmission speed differences where different communication circuits transmit at different bit per second rates.
• Data and Message Editing - Control editing, which consists of adding items to a message, rerouting messages, rearranging data for further transmission, or looking for nonexistent addresses.
• Message compression or compaction, which is a methodology for transmitting meaningful data messages but through the transmission of fewer data bits.
• Signalling of abnormal occurrences to the host computer.
• Assignment of consecutive serial numbers to each message and possibly, time and date stamping of each individual message.
• Message Queuing/Buffering - Slowing the flow of messages when the host computer or the remote terminal station (node) is overburdened by traffic.
  Queuing messages into distinct input queues and output queues between the FEP and the host computer or between the FEP and the outgoing communication circuits.
• Giving different priorities to different communication circuits or automatically assigning priorities to various types of messages to speed the throughput of various message types.
• Handling of time-out facility, such as when a specific terminal station does not respond or when a circuit ceases to respond.
• Error Control - Error detection and automatic transmission for parity checks on message blocks (CRC check and others).
• Forward error correction techniques to reduce errors flowing through the communication circuits.
• Message Recording - Logging all inbound and outbound messages on magnetic tape for a historical transaction trail.
  Logging the most recent 20 minutes on a magnetic disk for immediate restart and recovery purposes.

Other Functions

• Multiplexing
• Automatic switchover to a backup host computer in the event of a primary host failure.
• Circuit concentration where a number of low speed circuits might interface to one higher speed communication circuit.
• Performance of some functions of the Packet Assembly/Disassembly (PAD), if the front end processor is also serving as a switching node in a packet switching network.
• Network Control Networks can operate in a central control mode or an interrupt mode. The central control mode involves centralised polling of each station device (terminal or node) on the network. Polling is the process of individually giving each of the terminals permission to send data one at a time.

The interrupt philosophy implies that when a terminal sends data, the incoming data stream interrupts the hosts computer and the host stops processing so it can handle the incoming data. Because this mode is wasteful of processing capacity it is not generally used on host computers, except for some microcomputer systems that have very few terminals on local area networks.

Modems 
Another essential piece of equipment is the modem at the host end of the network. A modem is the first piece of equipment encountered at the remote end. Modem is an acronym for MOdulator/DEModulator. A modem takes digital electrical pulses received from a computer, terminal, or microcomputer and converts them into a continuous analog signal that is acceptable for transmission over a voice grade circuit. Until recently all of these devices were analog modems, because the signal was sent over an analog voice grade communications circuit. The introduction of all-digital circuits required digital modems.

Multiport Modems
Multiport modem cards are devices that are designed to accomodate modems (56k), ISDN, and other analog calls through a single dial-up number. Cards are available in 4 and 8 card models and come equipped with on-board processors that offload communications tasks from the server to improve overall performance.

Multiplexers 
The diagram showing the basic hardware needed for communication networking did not include several additional hardware devices that are used to help the network to run more efficiently and faster, to be more secure and easier to use, to interconnect to other networks and so forth. The additional equipment includes multiplexers, intelligent controllers, protocol converters, hardware encryption devices and line adaptors.

To multiplex is to place two or more simultaneous transmissions on a single communication circuit. An important aspect of multiplexing is transparency. Transparent means that the hardware multiplexer box does not interrupt the flow of data. Neither the computer, nor the modem, nor the terminal/microcomputer using the modem knows the multiplexer is being used regardless of whether the transmission is on a leased or dial-up circuit. When a circuit is multiplexed at one end and demultiplexed at the other, each users terminal thinks that it has its own connection to the host mainframe computer. Multiplexing is usually done in multiples of 4, 8, 16 or 32 simultaneous transmissions over a single communications circuit. Multiplexers can be separated into major categories such as frequency division multiplexers (FDM), time division multiplexers (TDM), and statistical time division multiplexers (STDM).

Frequency Division Multiplexing (FDM) 
Frequency division multiplexing can be described as having a stack of four or more modems that operate at different frequencies so their signals can travel down a single communication circuit. Another way of looking at FDM is to think of a group of singers. There may be a combination of a bass, an alto, a tenor and a soprano. What you hear is a combination of the four singing, but sometimes you can identify clearly one or more of the individual singers.

With FDM, the frequency division multiplexer and the modem are usually combined into the single piece of hardware. In FDM, the frequency divisor multiplexer uses the 3,000 cycles of available bandwidth in a voice grade circuit, dividing it into multiple subchannels.

The multiplexer shown, is subdividing the bandwidth of the voice grade circuit into four pairs of frequencies allowing four simultaneous transmissions of 0s and 1s. The guardbands are the unused portions of bandwidth that separate each pair of frequencies from others.

Time Division Multiplexing (TDM) 
Time division multiplexing is really a type of time slicing or sharing the use of a communication channel among two or more terminals. Each terminal takes its turn.

In TDM the multiplexer takes a character from each transmitting terminal and puts them in a frame . The frames are put onto a high speed data stream for transmission to the other end of the circuit. With TDM if four terminals share a 4,800 BPS line, then each terminal can transmit at 1,200 BPS. If any three terminals are transmitting then the frame will contain a blank bit for that terminal.

TDM is generally more efficient than FDM, but it does require separate modems. It easy to expand a TDM from, let us say, 8 to 12 channels. All TDM channels usually originate from one location and all terminate at another location.

Statistical Time Division Multiplexing 
Statistical time division multiplexing allows the connection of more terminals to the circuit than the capacity of the circuit. In its simplest context, you have 12 terminals connected to a statistical time division multiplexer and each terminal can transmit at 1,200 BPS, then your total is 14,400 BPS transmitted at a given time. However, if the STDM/modem/circuit combination has a maximum speed of only 9,600 BPS, there may be a time when the system is loaded above its capacity.

The technique of statistical time division multiplexing takes into effect that there is some downtime because not all terminals transmit the maximum capacity rated for every possible available millisecond. With this in mind, you start by addressing each character in the frame or message and time division on a statistical basis.

For example assume we have a statistical time division multiplexer that multiplexes individual characters from 12 terminals. In the case the terminal address is picked up in addition to the character and is inserted into the frame. Now the multiplexer only takes a character when the terminal has a character to send.

Stat muxes, as they are called, use software and a microprocessor chip built into the multiplexer. They can support a number of different devices at different speeds, without the modem having to equal the combined speeds of all the attached devices. Although STDM can be very efficient, it can cause time delays. When traffic is very heavy there can be anything from a 1 to a 30 second delay. Some data is held in buffers when too many terminals attempt to communicate at maximum capacity for too long a period of time.

Concentrators 
In today's terminology, concentrators are special forms of statistical multiplexers. Concentrators are used for the same purpose as multiplexers. In fact they originally were intelligent multiplexers (stat muxes).

The primary use for the concentrator is to combine circuits because you have 16 low or medium speed circuits that are concentrated into one or two high speed circuits. For example, you might concentrate approximately twelve 4800 BPS communication circuits into 56,000 BPS digital communication circuit. Even though this does not work out exactly (12 x 4800 = 57,600), the statistical intelligence takes care of the small difference. Like stat muxes, concentrators can buffer or hold back data. Some concentrators can even switch messages to different communication circuits.

Terminal Servers
A terminal server is a device that accepts incoming data calls and routes request for services to the appropriate source. Examples of termial servers are the Ascend MAX 4000, the U.S. Robotics Total Control Hub, the Shiva LAN Rover, and the Computone Intelliserver.

Terminal servers generally support incoming analog as well as ISDN calls. Some terminal servers can support analog calls which are delivered over a T1 line, called a Digital Supertrunk. A channelized-T1 line provides 24 inbound data calls, one per DS0 channel. Usually one phone number is assigned for customers to call in on and software at the Telco's CO switch provides a "hunt-group" which directs the next call to the first available modem. Each of the 24 lines has a unique phone number, but 23 of these are only seen by the telco switch. This is a convienience to both ISPs and customers.

Older terminal servers like the Computone Intelliserver require discreet phone lines be connected to individual discreet analog modems. This is a less than satisfactory solution since all those modems take up enormous amounts of space in an ISP's Network Operations Center.

A terminal server also incorporates a router, which is responsible for sending requests out to the proper server, and for receiving replies from servers and sending them back out to the appropriate modem line.

Some newer terminal servers, like the Ascend MAX and the U.S. Robotics Total Control Hub accept both analog and ISDN calls. They use special devices which can handle both types of calls. They can do this because an analog phone call is actually a digital call at every point in the phone network except the local loop (the wire from your house to the Telco's Central Office (CO) switch. Modem stands for MOdulator/DEModulator. An Analog signal is modulated to become digital, sent out over the phone company network, and then demodulated (returned to an analog signal) at the other end. If the "other end" can handle digital data, then why remodulate the signal to analog? Why not just deal with it as a digital signal? That's exactly what the new terminal servers do. They can recognize the difference between an ISDN call and a digitized analog call by the signaling information sent over the D channel (in the case of an ISDN call) or the absence of it (in the case of a digitized analog call). Once they recognize the type of call, they switch it to the appropriate internal circuitry for decoding and routing.

Terminal servers also handle login authentication. Most use a RADIUS (Remote Access Dial-In User System) database directly to check that a dial-in connection has a valid login name and password before they will accept the connection.

Terminal servers are expensive devices, costing anywhere from $19,000 to $40,000.

Most vendors now refer to their terminal server products as Access Switches.

Terminal Servers Vs. Multiplexers...

As more companies find themselves replacing outdated terminals with workstations and PCs LANs, a basic architecture question often arises. Most trade journals are touting new terminal server equipment as a replacement for traditional statistical multiplexers. However, they don't analyze the realities of wide area networking in those articles, so many designers are left with insufficient information on how to determine the best architecture for a given application.This section will provide that missing information and a methodology you can use to select the best architecture.

The "traditional" method of connecting remote users to a central UNIX system is shown in figure one. This is termed a "muxed system". The system consists of remote terminals connected through a statistical multiplexer and single communications line to the host computer site. The multiplexer enables many terminals to share the same line by manipulating the data in transit so that each user seems to have a direct connection to the computer. Statistical techniques are built into the multiplexer to optimize the use of that single line. Typical systems serve four to thirty or more terminal users through one data line to the host system. The terminals may be "dumb" terminals, PCs emulating terminals, or workstations emulating terminals. The electrical interface between these devices and the multiplexer is typically RS-232. Each terminal device is connected to a specific port on the host computer. This system supports "nailed" remote printers on host computer ports configured for printers; that is, each remote printer is always tied to the same host port. The system is relatively high performance, quite bandwidth efficient, simple, and has the advantage of nailed remote printers. All configuration can be performed from the host site, and no complex equipment is installed at the remote sites.

Figure One. Multiplexer System.

An equivalent terminal server system is shown in figure two. In this architecture, a terminal server instead of a multiplexer is located at the remote site. The remote site must have a LAN and one or two routers are needed to connect the remote terminal server to the host site. User terminals (dumb terminals, PCs, workstations) are connected to the terminal server's RS-232 interfaces. If desired, some workstations can connect to the host via the routed connection using telnet over the LAN/WAN. Printers are connected to the terminal server which must use the LPR/LPD protocol to print from the host computer. Terminal connections are usually dynamic; a given terminal will connect to various logical TTY ports on the host computer... instead of always connecting to the same TTY port as in a muxed system. The terminal server communicates with the host using ethernet and TCP/IP protocols. This system is not as bandwidth efficient as a muxed system and is more complex. Its main advantage is the use of a routed connection that may be required for other applications such as a graphics UNIX X-terminal system. The terminal server can also connect to more than one host computer, so if a terminal needs to switch between hosts, a terminal server is required.

Figure Two. Terminal Server System.

What are the compromises? What tradeoffs are required and which ones are optimized? Lets compare the two methods on a point-by-point basis. We will compare an 8 port system using each method. Equipment common to both systems is ignored (DSU/CSU, cables)

1. Performance.

With equivalent recurring communications cost (the same data line between the host and remote sites), Muxed systems will always perform better than terminal server systems. Muxed systems optimize data line utilization by analyzing the data with statistical methods and customizing the data throughput on a moment-by-moment basis. There is very little overhead passed through the communications link. Terminal Servers package each keystroke (for normal typists) or small group of keystrokes( if you are blazingly fast typing) into an Ethernet packet. That adds tremendous overhead to the system. For example, when viewing a menu screen, you press the enter key to make a selection. The terminal server will send 64 characters to the host system just to transmit that enter keystroke. That's a 6400% overhead. Obviously, when sending larger blocks of data such as printing to a remote printer, the overhead goes down; but, it is still quite substantial. Overhead on a muxed system is quite low. Multiplexers are tuned for a mix of relatively slow operations such as typing on a keyboard or displaying information on a screen along with printing on a remote high speed printer. For Performance, Muxed systems win hands-down.

2. Bandwidth Requirements.

A terminal server architecture will need about five to ten times the bandwidth of a muxed system for comparable response time performance. That additional bandwidth is a recurring cost each month. For bandwidth cost, Muxed systems win by a large factor.

3. Equipment Requirements.

The Muxed system requires only two multiplexers. Terminal server requirements require a terminal server and a router (possibly two routers). If the applications is character intensive, a second router may be required to support a high-speed WAN link. Per Port cost is about the same for both systems.

4. System Complexity.

Terminal server systems are more complex than muxed systems. Terminal servers require TCP/IP configuration, remote printer LPR/LPD configuration, and on-going administration to support these configurations as well as port configuration. Other than the port configuration, there is no additional configuration for a muxed system. Troubleshooting is much easier on a muxed system. There are no protocol issues involved in the muxed system. Terminal server systems rely upon routed ethernet and TCP/IP protocols with many potential points of failure.

5. Voice and Data integration.

Using a muxed system on a 56 Kbps line allows adequate terminal response when integrated with a channel or two of voice. Routed systems require at least 56 Kbps bandwidth so any voice added to a routed system will noticeably affect performance.

Terminal Server Vs. Multiplexer Comparison

* Prices are based on DCB multiplexer, router and terminal server solutions.
Checking the chart above shows that most of design compromises favor the muxed system in average applications. So why are terminal server systems installed so often? First, they are trendy. Many people depend on the trade journals that push what their advertisers pay for, and right now, that is ethernet networking gear. Many networking professionals don't have the experience to see just how robust and responsive a medium speed data line can be. Often, those who started in the business within the last few years haven't been exposed to multiplexed systems and don't understand the concepts.

Terminal server systems are appropriate under some circumstances. If a routed network is required for some other application such as X-Window workstations, remote email systems, or other internetworking; then the incremental cost of a terminal server is relatively low. Another factor would be the need for a user to switch between several host computers. Terminal servers have this capability which isn't available on muxed systems. The caveat is still centered around bandwidth and performance. Terminal server systems require much more bandwidth than muxed systems for equivalent performance.

Which method is best for your application? The multiplexer route is normally the most cost effective and highest performing option. If using routed facilities in common with other applications such as X-windows, the terminal server may provide adequate performance. If host switching is needed, a terminal server is required. There is a hybrid method, shown in Figure Three, that works very well for a mixed system of terminal users that need some LAN type access for email. By putting the routed LAN traffic on one port of a statistical multiplexer, LAN traffic such as email is available while maintaining the performance advantage of multiplexing terminal access.

Figure Three. Hybrid Multiplexer/Router System.

Future products will be available that provide the advantages of both systems. These will use a remote multiplexer in conjunction with a device that combines multiplexing and terminal services at the host site. Until this class of products are available, select a method that provides maximum performance with minimum recurring cost.