A COMPARISON OF MESSAGE AND PACKET SWITCHING

The Uses of Message Switching In Communications Networks

A COMPARISON OF MESSAGE AND PACKET SWITCHING

form of message switching especially suited for handling data traffic. Superficially, this is indeed the case, but there are very important differences between conventional message switching and packet switching, and the latter is, in reality, an entirely separate method of communication. In fact, there are resemblances between packet switching and circuit switching also, because the packet switching principle combines those major advantages of both circuit switching and message switching that are most appropriate for handling the interchange of information between com-puter systems.

With conventional message switching systems, messages of any length are accepted in their entirety and are stored as such at each switching point during their passage from switch to switch towards their destination. The network takes full responsibility for maintaining the integrity of the message, and elaborate procedures are employed to ensure that this is achieved. The accent is on reliability, rather than speed, in the information transfer between subscribers, and the messages are held by the network until the recipient is ready to accept them, however long this may take. It is not intended that subscribers should interact rapidly with each other through a message switching network, so the type of interaction common between telephone users to overcome errors intro-duced by the network, or indeed their own mistakes, is not possible. This is why the accuracy of informa-tion transfer through a message switching network is so important.

With packet switching systems, information is ex-changed in the form of short packets, and the transit time of these messages through the network is kept low; the subscribers are expected to interact with each other by exchanging packets, in much the same way as ~hey would interact by exchanging information through a circuit-switched connection. Because the subscribers interact together, they may take part in the validation of the information exchange procedure;

this can cope with their own errors, as well as any prob-ability of corruption during transmission can be the same as for a conventional message switching net-work.

It is worthwhile pointing out here that long messages are readily handled by a packet switching network if they are broken up into short packets for transmission and reassembled again at the destination. This method of operation is often called cut-through and is shown in Figure 16. The use of cut-through has important advantages: the fixed packet structure permits efficient handling, and the absence of indefinitely long messages prevents the blocking of transmission links and keeps the queues at switching points small.

Storage plus transmission Reception begins before original message ends l - d e l a y - - 1

Packet format

1 "'T"1---"T15""'1

All packets similar

H u

Figure /6. The packet switching principle

Storage at switches is made sufficient only for a few packets, and the total amount of information stored in the network is low. The result is that the delay through a packet switching network is much smaller than through a normal message-switching network, and the rate of throughput of information can be much higher. Where message storage is required as a service to subscribers, it must be provided externally to the network rather than as an integral part of the switches in the manner usual in a conventional message-switching network. This separation of the storage in the network into two parts, that used primarily for scheduling the efficient use of communication lines and that used for providing services and storing messages, clearly confers some very important benefits for handling data traffic, and a considerable amount of attention has recently been given to the design of packet switching networks.

A COMPARISON OF MESSAGE AND PACKET SWITCHING

The difference between packet switching and message switching begins with the messages and packets

JUNE 1979

The Uses of Message Switching iii Communications l"~e'twofks

selves. Messages are the units of information rec-ognized by the users of networks. Consequently, their length must be moderately unrestricted. To make them suitable for formatting by people, messages usually have a format depending on certain markers or codes.

Packets are designed for handling by computer and so have a fixed format. To keep their transit time low, a maximum length is set, and if messages longer than this maximum have to be carried, they are split into packets for transmission. Their format is not suitable for use by people.

Message switching emphasizes the responsibility of the network for the message, because no response or feed-back is expected-the delivery of one message is a complete transaction. Because message systems were not associated with computers but with big organiza-tions, the need to replicate messages, sending identical copies to many locations, has been met. Computers would distribute information selectively, not using replication. Packet switching serves interactive sys-tems where a response is expected, so the loss of a packet is generally less troublesome than a data error introduced in transmission.

Packet switched networks aim to deliver packets with minimum delay and do not hold them for delayed delivery. If the called terminal (or computer) is busy, the conversation will probably not even begin, but when it does begin it is likely that many packets will be exchanged before it ends.

The kind of transit delay expected in a packet switched system can be estimated roughly by noting that a 2048-bit packet can be sent over a 2048 Mbps link in one millisecond. The transit delay for one node will be a small multiple of this service time, deter-mined by the queue length plus the processing time for each packet. Because of the fixed packet size, service time is constant, and it can be shown that even for 80% saturation the mean queue length is two (not including service position), so the mean queuing component of the transit time is 2 milliseconds. The processing component is likely to be comparable with this time also. In practice, a much lower occupancy is used, less than say 25%, to provide a margin for error in traffic estimation. The contribution of queuing to mean transit time for aU. K. national network with such links is less than 5 milliseconds, and 10 milli-seconds is rarely exceeded. This is in marked contrast with message switching, when: transit delays of minutes are normal.

So, although the store-and-forward principle is employed in both kinds of systems, their objectives are different, and the performance expected is very different.

SHARED PRIVATE NETWORKS

Essentially, private data communications networks with lines leased from common carriers use store-and-forward message switching techniques and, being en-tirely the concern of a single organization, may use any message format and other standards most suited to the particular needs of the organization. The Western Union INFOCOM service is typical of several other message switching networks that attempt to serve the needs of more than one organization.

This task is complicated by the problem of agreeing on standards for a wide range of users, and this is made even more difficult if the basic problem of transferring messages between subscribers is com-bined with that of offering services that are provided as an integral part of the network.

However, because a group of organizations with similar requirements is more readily able to devise common standards, there are a number of shared private networks in use or planned. There will un-doubtedly be many such similar, but incompatible, data networks in a few years time. This situation is likely to worsen unless the PITs manage to agree on really comprehensive, advanced standards and begin to implement public data ~ommunications networks.

Of the many shared networks now in existence, two have been chosen for closer study; one is the airline network established in Europe by the Societe Inter-nationale de Telecommunications Aeronautique, the other is the Advanced Research Projects Agency net-work in the U.S.A.

The SITA Network

The Societe Internationale de Telecommunications Aeronautique (SIT A) was originally established in 1949 by a group of airlines as a non-profit making organization that would provide them with a cheaper means of exchanging messages. This was needed to facilitate the sale of seats on their aeroplanes, to exchange operational information, and for the loca-tion of baggage that had gone astray. For these pur-poses, SIT A organized a common service in the form of a worldwide low-speed message switching network handling teleprinter traffic; this was well established by the time computer-based airline seat reservation systems began to be implemented. Because of these new systems, a review of the SIT A network was under-taken. In 1964, the decision was made to adapt its design to keep pace with the new patterns of traffic that were expected to arise during the next ten years.

The anticipated traffic to be handled by the modified SIT A network was of three kinds:

The Uses of Message Switching in Communications Networks I. Type A: data traffic between computer systems

requiring a rapid response~ this would comprise single address messages using various types of 5, 6 or 7 unit information codes.

2. Type B: conventional teleprinter traffic using either CCITT alphabets No.2, or No.5, with messages having single or mUltiple addresses and of lengths up to 4000 characters.

3. Type C: single address data traffic requiring a response time similar to type B, and using CCITT alphabets No.2 and No.5.

The transit times through the network for type A traffic was to be of the order of 3 seconds, while the other two categories would have three levels of priority indicated by a label-QU, 2 minutes; QN or no label, 30 minutes~ QD, 12 hours. All messages handled by the network were to be protected from loss or mutilation, and facilities were required for the repetition of type Band C messages that had already been delivered and for holding them if the destination was unable to accept them.

The computer-based seat reservation systems already in use controlled their remote terminals by polling through networks of leased lines. The redesigned SIT A network was required to provide for handling this kind of terminal. In addition, the new network would have to match into the existing telex systems operated by PTTs and into other shared networks such as its American counterpart ARIN C (Aero-nautical Radio Incorporated), which serves some 90 U.S. airlines, and any private networks operated by companies such as BOAC, PAN AM, and KLM, to name just a few.

When the SIT A network was being reappraised in 1964, it was already connected to over 100 centers, and about 100 million conventional telegrams per year were being handled. This number was increasing at a rate that would double it in less than four years. To handle this growth in existing traffic and to satisfy the anticipated new requirements, a high-level or trunk network was planned; this would provide an improved service and new facilities to a large part of Europe, and also North America, and would join to-gether areas of smaller message concentration using the existing communications methods. High-level net-work centers were chosen in New York, London, Paris, Amsterdam, Brussels, Rome, Frankfurt, and Madrid, each having a defined area for message collec-tion and delivery. These main centers were to be joined by. data links, which needed to operate at, at least, 2400 bps to achieve the short response time specified for type A traffic. This response time criterion led to a decision to use store-and-forward block trans-mission in the high-level network with blocks of

variable length up to a maximum determined by the response time and transmission efficiency. Each block would have a header giving addresses and control information to guide it through the switching centers, and the overall efficiency would, clearly, be increased by using longer blocks for a given header size. On the other hand, the shorter the block, the less likely would be the chance of its corruption by noise, with the con-sequent need for its retransmission. Also, shorter blocks would allow a more rapid response for high priority messages, because the waiting time for a lower priority block to end would be shorter. However, an overriding facto I: that influenced the choice was the average message length of the existing telegraph traffic, which was less than 200 characters. Indeed, 95% of all messages had less than 250 characters. It was, therefore, decided to use a 256 character block with a 250 character information field available for carrying message. A long message would be trans-mitted as a sequence of two or more blocks.

The requirements to handle codes of up to seven units led to the adoption of an eight-bit character with seven information bits and an eighth bit giving overall odd character parity. An alphabet similar to CCITT No.5 was chosen for control purposes, and all other characters in the information field were to be padded up to seven bits where necessary. For example, with the existing telegraph messages in alphabet No.2, the start and stop bits would be re-moved from each character, and the remaining five information bits would be augmented by a sixth bit indicating the shift case and a seventh bit which was always 'one.' This would prevent an alphabet No. 5 control character being falsely simulated by a message character.

Following feasibility studies,6 a system was designed, and a specification was prepared for the computer centers. This allowed them to be introduced into the existing network of low-speed circuits in such a way that modification to form the high-level network was readily possible later. Univac 418 II systems were chosen for New York, Frankfurt, Brussels, Rome and Madrid, while Phillips DS714 Mk II systems were installed in the larger London and Paris centers, and at Amsterdam. The high-level network shown in Figure 17 was completed in 1970 and has been in service ever since.

New York London Amsterdam

Madrid Rome

Figure 17. The SITA high-level network, 1970

JUNE 1979

The Uses of Message Switching in Communications Networks

r RIO de Janeiro

Figure 18. Some planned SIT A developments

In parallel with the development of the main network, the existing networks were progressively modernized, a process that is still continuing. At first, the manual message switching centers were replaced by multi-plexers capable of handling several low-speed devices such as teleprinters, and later satellite processors were introduced. These are able to cope with a wide variety of traffic and can control the input-output terminals designed specifically for airline automatic systems.

Figure 18 shows some possible extensions of the SIT A network likely to be made in the next few years.

The High-level Network

The SIT A high-level network was designed with three aims in view. Firstly, to improve the service for conventional traffic; this has been done and transit times have been reduced from hours to minutes.

Secondly, to provide a service for handling short data messages; this, too, has been done and has enabied British European Airways to extend its data trans-mission network into Europe. Thirdly, to offset rising cost of conventional communications facilities; this has more than been achieved, for the efficient use of circuits and the reduced number of operating staff has brought about a fall in the cost per message.

Sydney

()

Because the SIT A network has proved so successful, it is worthwhile examining the procedures devised for ensuring the integrity of the messages it handles. This is done on two levels; the individual blocks passing along each link are protected against mutilation, loss or duplication, while complete messages are similarly protected between their points of entry to and exit from, the network.

The protection of blocks in transit between centers is achieved by adding to each block an even parity check character; this augments the odd parity bit used with each character and much improves the chance of detecting errors. As long as blocks are available, they are sent continuously and are checked for correct parity at the receiving end of a iink.(If no biocks are available the transmitter sends a link message every three seconds.) An acknowledgement is generated for every received block indicating whether it was correct (ACK) or incorrect (NAK); these acknowledgements are interleaved between biocks being received on the return channel. If a full length block were being received on the return channel, the acknowledgement for several short blocks sent on the forward channel would be delayed. A block numbering scheme is, therefore, employed and the acknowledgement signals carry the number of the last block correctly received.

Basic Concepts

The Uses of Message Switching in Communications Networks The loss or duplication of blocks is prevented by

arranging for each transmitting center receiving a correct acknowledgment (ACK) to erase the as-sociated information blocks, while if an incorrect acknowledgement (NAK) is received the center repeats all blocks with numbers following that carried by the (NAK) signal. At the receiving end of a link, the detection of a faulty block inhibits reception, and all further blocks are rejected until the block previously found to be faulty has been correctly received.

If after retransmission a series of blocks is not correctly acknowledged, it is repeated a further three times.

If a correct acknowledgment still fails to appear, the communications circuit is assumed to be faulty, and traffic is diverted to another route. Meanwhile, a print-out is produced advising operators of the situation, and check messages are sent continuously into the faul.ty circuit until it is found to be operating correctly agam.

When a link between centery becomes faulty or is restored following a fault,;r status message is sent automatically to all other centers. This message causes all routing tables in the network to be modified accordingly. Should all the links to a center, or indeed the center itself, fail, the low priority messages for the center are stored at entrv points until the center becomes operational again.~However, the high priority messages must be dropped, and the polling of all remote terminals belonging to computers served by the out-of-action center has to cease.

The automatic rerouting of blocks during fault condi-tions makes it possible for some blocks of a long message to arrive out of sequence, and to overcome this problem an entry-exit block numbering scheme is used. This operates in a similar manner to that used on individual links, but a separate series of numbers is used for each type of traffic and for each pair of network centers. The entry center holds all blocks until they have been correctly acknowledged as received by the exit center. Individual blocks are not acknowledged, but there is at least one acknowledg-ment provided for every 16 blocks. The entry-exit block numbering arrangement enables the blocks of a multi-block message to be reassembled correctly and

Im Dokument COMMUNICATIONS SOLUTIONS (Seite 161-170)