by P.Bysouth, P.Hicks and J.Park April 1, 1987
Of particular commercial importance and interest to operators of large distributed computer networks is the transfer of extremely large data files. Typically these files may range in size from tens to many hundreds of Megabytes. In these days of the multinational conglomerate it is not unusual that an organization's corporate data may be required to be duplicated at several sites around the country. Hence it is immediately obvious that there is a pressing need for the error free transfer of bulk data at high speed between corporate sites.
In order to meet this need Telecom Australia is investigating the implementation of a new network for the transfer of bulk data between the corporate centres of Australia namely Melbourne and Sydney. It is possible that the network will be expanded to cover the other capital cities at a later date. This paper describes the new network which is to be known as HUMENET. The achieved data throughput is between 2400 packets/sec and 32400 packets/sec depending on the Layer 2 mechanism chosen.
Consideration has been given to the design of Humenet in view of the ever increasing needs for bulk data transmission by major corporations and government bodies. The volume of data traffic is growing exponentially worldwide and shows no signs of leveling off in the near future.
Within Australia, there is a strong demand for a facility for bulk data transfer between the major commercial centres (Melbourne and Sydney) and Humenet is aimed specifically at that market. Other centres within Australia could be addressed by the use of identical means, although perhaps at the cost of a lower quality of service or higher cost per unit of data. International traffic has not been considered because of a number of technological limitations.
The major service requirements are in accord with those of most other data communication networks. They include:
Use of OSI principles:
In view of the very wide spread interest in Open System Interconnection standards, their application for most of the modern telecommunications signalling standards, and a necessity for independence from the in-house standards of particular manufacturers, the network is to be based upon a seven layered protocol model.
The obvious protocol to use at the Application Layer for the transfer of bulk data is the OSI File Transfer And Management protocol (FTAM). This protocol has been designed with applications such as this in mind, it is an international standard and provides all of the functionality required.
At the Presentation Layer ASCII encoding is to be used.
No Session Layer protocol is seen to be necessary.
In order to provide for multiplexing and in view of the mechanisms used for the lower layers and their potential error rates, the Class 4 Transport protocol is used. This of course implies a need to retain a copy of the data currently in transit at the originating node. In addition, users who do not require as high a quality service may use the Class 2 Transport protocol.
At the Network Layer, the basic mechanism to be used is magnetic tape. It is necessary, of course, to add several features to this in order to realise the Network Layer Service and to provide a well defined interface on which to place the layer 4 protocols. These additional features are important in providing the necessary addressing information, flow control data type indications. Additional security is available to users (at a small additional cost) by the allocation of dedicated tapes. Priority is available (also at a small additional cost) by the use of specially colour coded tape holders. The additional capability of using double sided double density floppy disks can also be provided and would obviously provide a service capability for a broader spectrum of users.
Layer 2 provides the opportunity for considerable ingenuity and differing quality of service. In the best traditions of buying Australian products, layer 2 makes use of red Holden Commodores (the turbo charged 3 litre variety), with the boot space adapted to hold layer 3 modules. Whilst it may be possible to add to the prestige of the service by selecting red Ferraris or Porsches, the substantial purchase price differences and the difficulties of on-location service centres regrettably removes these possibilities from serious consideration. In addition, it is important to note that the turbo charged option can lead to reduced transit delays (although some further consideration needs to be given to the legal aspects of speed). It is possible to offer an expedited data service by maintaining a number of layer 2 transport modules on hot standby at each of the end nodes. The above selection of layer 2 transport modules is seen to have the significant advantage of ensuring that staff should be readily obtainable, indeed it may be possible to have them pay us for the privilege of using them. It may be possible to run a side business of training racing drivers, with no disadvantage to the data service, and with the potential advantage of reduced network delay! An additional service for very large users who do not mind an increased transit delay is to be provided by the use of MAC trucks at Layer 2. Of course, the trucks would also have to be painted red in order to avoid unneccessary protocol differences from other layer 2 modules and to ease the pain of conformance testing. Collision detection is not expected to be a problem as notification by telephone from the nearest network node could be expected to take place quite quickly following such an incident. On the other hand, special training and staff selection may by necessary to ensure adequate collision avoidance performance. Loss of data may occur as a result of collisions and can be expected to show a strong correlation with the delay performance objectives.
Since the majority of the traffic is expected to be between Sydney and Melbourne, the obvious choice for Layer 1 is the Hume Highway - hence the network name 'Humenet'.
At layer 1, the Hume Highway passes through quite a large number of transit nodes between the end points of Melbourne and Sydney. It is anticipated that the layer 2 vehicles will only need processing at a small number of these (perhaps the ones at Yass and Albury), although this will obviously depend upon the particular vehicle and driver, and the transit delay requirement. A particular virtue of the network structure is that processing should be possible at most nodes should the need arise.
Of special interest to network planners is the ability of the network to provide fast transit around intermediate nodes where no processing or exchange of data is deemed to be necessary. In this respect the State Governments of both Victoria and New South Wales are continuing to contribute to the future possibility of enhanced delay performance.
It is anticipated that, in common with most networks, there will be times of high traffic. There will also be times when the usual network links are not of sufficient availability (for example, due to abnormal traffic patterns, intolerable collision frequency, over-long transit delays due to water incursion at layer 1 (floods), or industrial action at the transit nodes). Consequently it is essential to ensure that alternative routes are available. In this respect, Humenet is very well served by the near proximity and interconnection of the end nodes of Princesnet (although the delay performance of Princesnet can be expected to be a little worse than the normal performance of Humenet due to the increased propagation delay and higher hop-count). It should also be noted that interconnection between the two networks at points other than the end nodes is also possible. No special interworking protocols are needed.
Assume that a 2400' magnetic tape holds 20M bytes of data and a floppy disc (double sided double density) holds 1M byte. These data densities have been chosen as they are common to a wide range of equipment. Higher densities (140M byte) are possible. Assuming a 9 byte overhead per data packet each tape can hold 145,985 128 byte packets. It should be noted that it is not necessary for the tape to hold individual acknowledgement packets, hence the utilization factor (the % of tape used for actual data approaches 93%). Assume a distance between data centres of 1000 km. Also assume that the time to transport the data storage mediums is 10 hours (ie. 100 km/hr average road speed). Assuming a car spacing of 1 minute and a minimum of 1 tape (or 20 floppy discs) per car the destination host receives 2433 packets/sec. Such a throughput rate on an X.25 type connection could only be achieved with a 3M bit/sec line. Now consider the mechanism of using a truck loaded with magnetic tapes. Given the dimensions of 30' x 8' x 8' and a volume of approximately 0.1 cubic feet for each tape the truck can carry 19200 magnetic tapes. Assuming that we only dispatch one truck per day the host receives data at a rate of approximately 32441 packets/sec. An X.25 service would require a line rate of around 40 M bits/sec (assuming that the switches could cope) to achieve a similar throughput rate.
Assuming one tape arrives each hour (minimum arrival time) it has taken 10 hours for 145985 packets to transit the network. Hence the average transit delay for a single 128 byte packet is 0.247 seconds. This compares favorably with the current performance of AUSTPAC and also meets the Recommendation. X.135 allowance for the National Portion of Transit delay.
The availability of the network is considered to be 99.998% of all time. This is based on fact that the total traffic flow is only affected by a factor of 0.002% due to petrol strikes and or road maintenance.
Bypass of Humenet is obviously possible (such services as AIRNET who would offer a premium service with very low transit delay) but, we believe, would not succeed due to high costs. As noted above, we have selected the most direct route, and a very fast mechanism for layer 2 (which is also cheaper than most of the similarly fast options). Further consideration needs to be given to the potential implications of network enhancements such as Node Bypass, Overpass, Underpass and Bridge Technology.
In comparison with other packet switched data networks (such as Telecom's Austpac network), the cost of data transfer using the Humenet network becomes competitive when large amounts of data are involved.
For example, Austpac charges on the basis of:
To these charges, the line installation and rental charges must also be added. These are typically $1000 installation fee plus a yearly rental of from $3300 for a 2,400 bit/second line to $7,260 for a 9,600 bit/second line. As an example, it would cost $9,227 to send 1 Gigabyte of data during an off peak period. The time to transfer the 1 Gigabyte at a line speed of 9,600 bit/second would be 33 hours, however if the effects of flow control are taken into consideration, then the time increases to about 36 hours.
Digital Data Network:
Unlike Austpac, the DDN network charge is based on the capacity to send data rather than the actual amount of data that is sent. The charges are based on three quantities:
An example of a 9,600 bit/second line between Melbourne and Sydney would be $9084 transmission charge, a yearly rental of $6936 plus an installation fee of $1000. This is a total yearly charge of $16020.
In comparison with the Austpac example above, if several gigabytes of data are to be sent, then the DDN network would be cheaper than using Austpac, however Humenet would be cheaper than both (and probably faster).
It can be seen that, given the strong moves toward the transfer of large quantities of bulk data, Humenet can offer a fast and cost competitive service. It is structured directly on the OSI Reference Model and meets the CCITT performance criteria (for large quantities of data). The network structure at layer 2 is novel in that it permits node bypass where processing is not necessary and allows extensive use of alternative routing in case of network congestion or loss of links. The network is not expected to be prone to attack by bypass by others. Expansion of the network to cover centres in addition to Melbourne and Sydney should be readily possible as the required layer 1 infrastructure is already in existence.