LOUIS POUZIN
Institut de Recherche d'Informatique et d'Automatique 78150
Rocquencourt (France)
Abstract
CYCLADES has been implemented on schedule, without significant change to the initial design. The packet switching network (CIGALE), host protocols, terminal access, and basic host services are operational. But CYCLADES users are still mainly a research oriented population, although no longer a majority of system programmers. A connection has also been established with the NPL network, in the London area. The workforce is mainly composed of highly motivated research oriented people, with control based on results rather than administration. Particular emphasis has been given to human communications. Critical tasks are undertaken by IRIA, with assistance from participants and software houses. Figures are given on resources spent so far. Objectives have been largely met in technical aspects, and applications are beginning to pick up. The future of CYCLADES is twofold. The present network is becoming a university service bureau, as well as a tool used to promote international cooperation, specifically in the area of distributed data bases. It will also give birth to similar systems in the private sector. A redevelopment of protocols is taking place, and the core of the present implementation is being made available to the industry. C.I.I.3 and software houses are cooperating to make CYCLADES protocols a common model in order to market a consistent set of products for building heterogeneous networks.
Introduction
A previous paper[1] has already introduced the CYCLADES network, its goals, participants, and major design points. One year later the basic services of the network were implemented and put into operation. Most of the contents of the earlier paper are still valid, and it serves little to repeat them here. The present paper is thus a companion to its predecessor. It will indicate changes in our initial plans, current developments, and the likely direction the future will take.
Participants
Participants are essentially the same as originally quoted, other than the addition of one more institute in Grenoble, two in Toulouse and one in Rennes. In other words, the network environment is composed of research oriented or university people having a computer science profile, albeit that a sizable proportion are young and have only limited experience.
Inter-connections not originally planned have been undertaken. CYCLADES and the NPL 1 network have been linked in August 1974, and another link with the ESRO 2 network should be established by mid 1975.
More than a dozen other educational or research institutions have expressed the wish to join in CYCLADES. However, we were reluctant to increase significantly the size of the network until such time as it was operational and its future better defined. Indeed, managing the whole project would have become increasingly difficult, and the additional costs not justified by contributions in new areas. However a few more participants were allowed in when they already had their telephone lines, and were involved actively in software development for the network.
Hardware
Computers are essentially the same as anticipated, except that most of the C.I.I.-10 070's have now been replaced by a new model, the IRIS 80, which is software compatible and significantly more powerful than its predecessor.
In summary, ultimately CYCLADES will have 19 host computers in 16 different locations. Currently 14 hosts are connected.
Due to budgetary restrictions, we had to reduce the number of (rented) mini-computers planned in 1975. Rather than to chop off terminal concentrators, we chose to turn some of the packet switching nodes into concentrators, in order to maintain wide access to CYCLADES hosts. Since nodes were then too few, a host interface has been implemented on concentrators. Temporarily, 2 hosts were disconnected, reducing the host population from 16 to 14.
Line speeds are as indicated on Figure I. 48 Kb/s lines were delivered late in '74 and their operation raised a few hardware and software difficulties, which delayed their use on a regular basis until March 75.
FIG. 1. - CYCLADES topology
Development
Year '72 was devoted to the general design, planning, and staffing of the project. Participating centers became active towards the end of '72.
Year '73 saw the implementation of a baby network, which was publicly demonstrated at the beginning of November '73, with 3 homogeneous hosts, and one packet switch. Facilities available were :
- Inter-operator communications
- File transfer
- Remote job entry
From that point on, the network grew rapidly.
Another official presentation in February '74 featured 4 hosts and 3 packet nodes. The packet switching network (CIGALE) was turned operational for 3 hours a day, and this immediately fostered the testing of host software on the various sites, as well as giving practical experience in the operational aspects of a packet switching network.
Software for a terminal concentrator was developed on the MITRA-15 (the same mini-computer as used for CICALE). This allows users not attached to a host to use the network and the services offered on any host. In practice, concentrators may be viewed as a specialized version of a host, on which only relevant parts of the network protocols have been implemented. In March '75, there were 4 concentrators installed. Terminals are normally provided by users.
By June '74 CIGALE had 7 nodes, and was made operational for longer sessions, up to 6 hours a day since January '75. In addition CIGALE is also available most of the time out of regular sessions, for users who want to run tests or experiments.
In February '75, CIGALE shrunk to 3 nodes, for reasons indicated previously, and 4 concentrators were equipped with host interfaces.
A control center and a measurement center were also developped on Mitra-15's. The former is located at IRIA and the latter in Rennes (in a P&T 6 research center). They monitor the network for statistics and exception conditions. The control center is used also to load programs dynamically into nodes.
Host services had to be provided, if the network were to be any useful. New developments were rather straightforward, as they were implemented as user programs interfacing through macros or sub-routine calls with CYCLADES transport stations. Existing manufacturer software, viz. time-sharing and remote batch, were made to interface with the network.
Services available in March '75 were time-sharing, remote batch, file transfer, and local terminal access, both on IRIS-80 and 360/67 computers. The same services, except time-sharing, were also available on 10 070.
No services were at that time offered on other machines, save for a few ad hoc implementations.
These facilities were demonstrated in Toronto (Canada) at the beginning of November '74. The MITRA 15 installed there worked alternately as a terminal concentrator on CYCLADES, or as a ClGALE node. The link was a 4 800 bits/s leased telephone line.
In addition, CYCLADES and the NPL network were interconnected [2] in August '74. This is considered as a first leg of the European Informatics Network [3] (EIN) of which NPL and IRIA are node centers. Initially terminals at IRIA could access services on the NPL network, principally using SCRAPBOOK, a text editor and filing system. The use of CYCLADES facilities from NPL terminals was not yet available, due to differences between present CYCLADES protocols and those agreed with NPL. Some adaptations are being worked on.
Team Organization
Controlling the work of a multitude of organizations scattered over many geographical areas would require a very powerful line of command if tasks were organized on a factory basis. In a research environment this approach is just too unrealistic. Therefore we had almost no administrative control and very few reports or large meetings (no more than quarterly).
The main idea was to develop friendship among all the participants, who hardly knew each other before, and to introduce goals, challenge, enthusiasm, as well as some redundancy so that no failure would be critical. People were encouraged to take initiative and see each other whenever necessary. Mixed teams involving persons from various places were set up, mostly to break parochialism and inhibitions. In order to make sure that no one would be left out, or retire to an ivory tower, a person from IRIA was assigned specifically to the job of traveling consultant and gossip carrier. It was assumed that information around the network would circulate much better through thousands of informal channels. In particular contributions, or lack of contribution, from individuals were more widely visible. This approach seemed to work.
However, there were unavoidably some critical paths, which were kept under the direct control of IRIA within the headquarters of the project. These tasks were:
- CIGALE, the packet switching network
- The definition of network-wide protocols
- The implementation of protocols on CII machines
(10 070 and IRIS 80), which represented 50 % of the hosts.
With these tasks under control, we were in a safer position to set up half the network on schedule.
Participants see CIGALE[4] as a black box carrying packets. Thus they were not directly concerned in its construction. It was developed mainly as a sub-project in conjunction with the French P&T, who assigned two people to this task. The CIGALE effort coincided with the first deliveries of MITRA 15's by C.I.I. This meant teething problems, mainly in technical support, documentation, and basic software. In addition our staff were mostly inexperienced programmers. It took until mid '74 to recover from these early difficulties. Well trained professionals from software houses were gradually brought in and CIGALE is now well under control.
The protocol team was initially composed of people from a variety of participants, in order to gain wider acceptance. Some individuals were assigned to IRIA for periods ranging from 3 months to one year. Others worked with us for short periods, or attended frequent small meetings. As a result, the CYCLADES protocols, were already reasonably well accepted when their specifications were made available. We experimented with a method of portable definition, intended to facilitate implementation on various operating systems, and prevent ambiguities in interpretation. This approach has been quite satisfactory, as practically no incompatibility problems have been reported from the various sites.
Participants were in charge of developing their own host software, when they could not benefit from the IRIA implementation. In addition they co-operated in various ways : adaptation of host services to the network environment, demonstrations, simulation of host protocols before implementation, preliminary studies in view of coming developments. Most of them were given contracts from IRIA to help them set up a local team, and compensate partially for the manpower they provided. On the other hand, they had to make computer time available for software testing.
A noticeable point has been the very strong cooperation with C. I. I.3 in the definition and implementation of host protocols and adaptation of host services. While basic host to host protocols were designed without the need for operating system modification, this could not apply to existing access methods designed for directly connected terminals. Furthermore a proportion of participants would not have reacted enthusiastically to the idea of having modifications, unsupported by the manufacturer. In order to anticipate these predictable difficulties, an agreement has been reached with C.I.I., whereby this company takes the responsibility of maintaining the CYCLADES software on its own computers. A similar problem could arise with other manufacturers, but its scale is so far much more limited. Modifications are presently under participant responsibility.
When CYCLADES was initially set into operation, i.e. in the last quarter of 1974, we relied on each participant initiative to be present on the network at least 2 hours a day. It did not work out, as most participants did not consider the network as really operational, and shunned sessions as a waste of time.
Through a mixture of participant volunteering and political support from the Ministry of Education, it was decided to expand the session time to 6 hours, and concentrate all effort on 2 centers, Lyon and Grenoble (2 IRIS 80, 1 360/67). Simultaneously a campaign was launched to recruit users and install terminals attached to concentrators. Monthly meetings were held to review progress, uncover hurdles, and take appropriate speedy action. All parties involved were informally invited so as to create a forum atmosphere and exchange experience.
Since January '75, the project definitely picked up new momentum, even though it was only visible within a few participants. Emphasis shifted from system to user matters: terminal keyboards, command language, manuals, session schedule, training, etc. The real concern was no longer whether the net worked, but how to make it attractive. Soon other participants got anxious to step on the bandwagon.
Personnel and Resources
The number of persons assigned to the project has been changing constantly, and not all persons were assigned full time. As it was not realistic to keep track precisely of the amount of manpower made available within each participant, we relied usually on samplings taken at a few months intervals.
The manpower spent on the project till the end of '74 is about 70 man-years. Expenses in contracts with participants, software houses, and C.I.I. totaled roughly 12 millions French Francs by the end of '74. Even without reliable figures on participant contributions, it was estimated that they would equal the amount of subsidies5 [5].
The total forecast till the end of '75 was respectively 140 man-years, and 52 MFF. In other words 50 % of the resources have been spent during the first 3 years. Since there are about 70 people currently working full time on the project, predicted costs will probably be attained over the current year.
Overall the project is both on schedule and within the predicted budget. It is however less predictable that with inflation and reduction in expected funds, we will be able to continue throughout '75 at the planned level of commitment.
It should also be recognized that the extent to which objectives have been met is not always as we expected. In the areas of basic software, results are generally beyond our initial expectations. A second iteration of host protocols should be completed by mid '75. CYCLADES and CIGALE have been accepted not only as an interesting experiment, but also as prototypes of commereial products. The transfer of know-how to the computing industry has been faster than could be predicted. On the other hand, the adaptation of existing manufacturer software is more cumbersome than we had hoped it would be.
Finally, applications got off the ground, but slower than we anticipated. It is clearer now that real users do not move as long as they cannot check the practical operability of a new tool. It appears that it could take several more years to establish live applications in a casual DP environment.
Nevertheless, prototype applications are being introduced in '75. The subject of network usage will be covered in the following chapter.
Cyclades Applications
There are basically two domains of applications for CYCLADES. One is the use of the present prototype network, in its current configuration, or with some extensions. The other is the development of independent replicas of CYCLADES for the internal use of large organizations.
We might term CYCLADES-0, the present project, initially planned through the end of '75. The main objective in terms of network usage is to turn it into a university network for the benefit of the Ministry of Education, which owns all computers within universities and some research institutes. To that effect CYCLADES should allow students and researchers (not computer specialists), to get the most cost-effective computing services available on the network. It is mainly a service bureau type of operation for unsophisticated users whose ambitions reach the level of compile-and-go Fortran jobs, or pre-packaged applications.
Therefore, what is required, is to make the usual manufacturer software services available on the larger computers of the network, and provide access through terminal concentrators. This type of use started in January '75 on the Lyon and Grenoble computers, with both local and distant customers. Once this service is considered satisfactory, other centers will be gradually added, as well as additional concentrators. IRIA is already joining the band. Existing smaller scale computers will also be used for concentrating terminals, possibly in parallel with local tasks.
Other than for the typical software and equipment problems, the main difficulties that remain are organizational and psychological. Users hate changing their habits, and it is always difficult to persuade them to try another computer or system. This may be a reason why we found better response from people who were working on new applications, and had not yet settled for specific computing services. Computer centers are not organized to deal with very distant customers, particularly in regard to documentation and consulting services. Charging practices are also quite a problem in the typical French Administration environment.
It is clear that new methods, policies and tools, will have to be devised. Our approach is to try first, and then correct from experience. There is definitely no intention of developing sophisticated and expensive tools, nor re-inventing manufacturer software. Thus, network services are going to be limited to whatever is realistic on present day systems. A few simple tools are already available and may be expanded later, e.g. network journal, standard description of services and tariffs, ready to go sets of control statements, etc.
In addition. CYCLADES-0 is intended to be used as a research tool for further network experiments : network languages, new types of terminals, new operating systems, etc. This is not likely to involve the hosts devoted to the university service bureaus.
It may be useful to list some of the applications that were at various stages of development in March '75 :
- Processors available under time-sharing or remote batch include : APL, SOCRATE (a general data base manager), MISTRAL (documentation and information retrieval).
- Using these tools, or specific programs, gives access to a chemical data bank (THERMODATA), a circuit simulator, a computer simulator, an economic modeling system, etc.
- Specific applications were under development in social sciences, linguistics, faculty management, CAI, hemophiliac data bank.
- Plans were made to set up pilot applications in the Customs Department, and in military laboratories.
CYCLADES -bis are the commercial fall-outs. Two departments of the French Administration have already opted for private CYCLADES, which will be installed by software firms. It turns out that many large organizations operate heterogeneous computers in different locations. For various reasons, not necessarily technical, they choose to keep these installations, but use them more efficiently. Among typical requirements are access to several computers from the same terminals, or file enquiry from various computers or terminals.
Some large corporations prefer building their own system, on their own design. This trend is typical when a new technology spreads over. Technical staffs are eager to acquire first hand experience and go convince their managers that only a homemade system can meet their requirements. In some cases, such an investment. is totally justified. But in general, experienced managers favour already proven systems backed by dependable vendors. Already supported on UNIDATA equipment, CYCLADES is currently the only system offered commercially with a comprehensive set of well defined protocols for heterogeneous networks. Even when their installations are homogeneous, sophisticated users balk at the prospect of getting completely trapped in the spider web of too powerful manufacturers.
Studies are under way to make CYCLADES protocols commercially available under various operating systems. Most likely, this support will be provided by software houses.
In this context no plans are yet sufficiently advanced to speak from experience. So far it seems that technical difficulties are to be found in adapting existing procedures, access methods, teleprocessing packages, into a network environment. This problem is general, not peculiar to CYCLADES. We can expect however that the careful insulation of the various logical levels of protocols will make it easier than in more ad hoc constructions.
Another potential problem arises from the commercial nature of the industry. It is a constant strategy of well established suppliers to devise products and services intended to keep customers captive, and competitors at bay. Once it is a known fact that CYCLADES opens the door to foreign computers and terminals, it is unlikely to be welcomed by large manufacturers.
Technical Changes
At the present time, the basic protocols are those described in [1], except for the "liaison" type which has not been implemented. There was no direct use for it in the conventional services we were aiming for.
During '73 extensive discussions about network standards took place on the international scene, mostly within IFIP WG 6.1. In the light of these discussions, and as a result of our experience with the present protocols, a new design was produced [6]. Among the major points which brought us to this change were :
- Letter size was dependent on packet size
- Connection set up was too costly in its implementation
- There was too much freedom in protocol variations.
This last point may sound paradoxical, since CYCLADES advocates precisely the ability to allow for any kind of user tailored protocol. A bit of explanation might be useful.
Our initial design presented the various letter services as independent and parallel implementations, on top of a basic service. In practice, independence meant redundancy and more expensive implementations. Sharing functions would have created hidden dependencies. Furthermore, future needs would have increased this undesirable dilemma, since users were tempted to invent protocols with slight variations, leading gradually to a jungle of mutant species.
The new approach tends to eliminate duplication by making protocols interdependent in a tree-structured manner. New services may be added as options, but it should be tempting to build onto the more basic ones. Furthermore, a negotiation procedure allows interworking between parties having not necessarily the same set of options. To sum up, there is freedom, and discipline. Users may start from scratch and build their own basic protocols. It is only hoped that this will not happen too often.
The new design of host protocols is taken as standard within C.I.I. for its new line of software products. It is also accepted as a first step in the European network (EIN).
Additional higher level protocols have been specified and implemented :
- A contact protocol [7], allowing a terminal to set up connections with a particular transport station.
- A virtual terminal protocol [8], intended to make compatible any terminal and any server system on a host. The CYCLADES virtual terminal is modeled after a tree structure, so as to facilitate the handling of structured data, such as files, or intelligent terminals. It handles Teletypes as well.
Again, these protocols are integrated in C.I.I. software products, so that commercial support be readily available. Several software houses have joined in a CYCLADES club, with the objective of disseminating common standards, and a consistent product line. This may be a starting point in attempting to work out nation-wide standards.
Present Developments
The CYCLADES project has mushroomed in various areas, some of which are described briefly below:
Distributed Data Bases
A long term objective is to be able to access in parallel several heterogeneous data bases, for some integrated purpose. A first step is being implemented using SOCRATE [9], a data base system available on C.I.I., SIEMENS and IBM machines. A network access method [10] has been implemented to allow accessing a homogeneous data base distributed over several computers. A second step will be to link two data base systems.
Several research projects are investigating practical schemes allowing a variety of mini-computers to offer services and resources in a consistent manner, without losing their autonomy (e.g. university of Toulouse [11]).
A particular effort is being put on the sharing and dissemination of scientific information, for which a substantial investment has been made in the past decade, in the form of specialized data banks. This is a world-wide effort. In France, coordinating agencies are BNIST 4 and BAB 5. They are involved through the European communities in the EURONET project, which intends to pool scientific information allover Europe. The interconnection of CYCLADES with the SDS-ESRO network is part of this drive. The cooperation established between France and Canada in this particular domain should result in common projects undertaken in '75.
At the worldwide level, UNESCO is conducting a long term study (UNISIST [12]), whose goals are to plan and organize the dissemination of scientific information across all countries. This is an obvious and challenging future for international cooperation.
Network languages
A research project investigates the feasibility of creating a network wide language interpreted within a portable environment [13] This work should be pursued in cooperation with EIN.
Simulation and Modeling
Host protocols are simulated before implementation, in order to check their logic and efficiency. A CIGALE model is under way [14], particularly in view of testing routing and flow control techniques. On this last aspect, co-operation has been started with the University of Waterloo (Canada).
Terminal Access
Concentrators are being refined in order to allow various configurations, and supervision through ClGALE. Studies are under way for prototypes of terminals and mini-concentrators working directly with standard network packets [15]. Radio-terminals are also being investigated.
Present Trends
In our view the network scene is rapidly changing. What was considered a few years ago as futuristic is now forming the basis of tomorrow's plans. There is noticeable pressure for less experimentation and more practical work. There is also more concern for standards [16], since it is clear that networks will have to communicate with each other.
For these reasons the CYCLADES effort is already being transferred gradually to the private sector. Both software and hardware developments and engineering are now becoming commercially marketable. It is not IRIA's mission to undertake such steps. Hence, there will be a split between research type activities and industrial developments. Specific emphasis will be put on some prototype applications, such as military, financial, or documentation systems requiring typical network technology, i.e. distributed resources.
Standards have been a definite concern from the beginning of the project [17], and it is a permanent objective to seek agreement with other "networkers" on some level of compatibility. Interconnection with NPL is one example. Another is the public vs. private network issue. CIGALE offers the type of service that the CCITT calls "datagram". Therefore, users building now private packet switching networks will be able to expand them, or to replace them later with public services, provided they use CYCLADES protocols, which only require datagrams.
The CYCLADES philosophy is geared to offer an alternative to organizations which are not prepared to sign away their future with a particular manufacturer. It seems that this is going to be less and less of an academic issue.
References
[1] POUZIN L. - Presentation and major design aspects of the CYCLADES computer network. Third Data Communications Symposium, Tampa, (Nov. 73) pp 80-85. HYPERLINK
[2] GIEN M., SCANTLEBURY R. - Interconnection of packet switching networks. Theory and practice.EUROCOMP, London, (Sep. 75), to be presented.BAD TITLE ??
[3] BARBER D.L.A. - The European computer network project. ICCC, Washington, (Oct. 72), 192-20[0] [4] POUZIN L. - CIGALE, the packet switching machine of the CYCLADES computer network. IFIP Congress, Stockholm, (Aug. 74), pp 155-159. HYPERLINK
[5] POUZIN L. - The economics of computer networks. The CYCLADES case. International Symposium on Economics of Informatics. (Sep. 74), IBI ed. Vol. II, 79-8[8] [6] ZIMMERMANN H., ELIE M. - Transport protocol.Standard host-host protocol for heterogeneous computer networks. Doc CYCLADES SCH 519.1; (Jun. 74), 31 p. Also IFIP WG6.1, doc 6].
[7] ZIMMERMANN H. - Protocole client-serveur de terminaux. Doc. Reseau CYCLADES, TER 502.2, (Nov. 74), 8 p.
[8] ZIMMERMANN H. - Terminal access in the CYCLADES computer network. International Computing Symposium, Antibes (Jun. 75), (to be presented).
[9] AERIAL J.R. et a1. - Projet SOCRATE. Nouvelles specifications (version 3). Universite de Grenoble,(Sep. 72), 625 p.
[10] CHUPIN J.C. - Control concepts of a logical network machine for data banks. IFIP Congress, Stockholm, (Aug. 74), pp 291-295
[11] BEAUFILS R., LAGASSE J.P., CAGANEL J.P. - Projet ARAMIS. Specifications d'un reseau de miniordinateurs. Universite de Toulouse, (1974), 14 p.
[12] UNESCO - UNISIST Synopsis, (Mar. 71), 92 p.
[13] DU MASLE J., FARZA M.N., SERGEANT G. - Proposed organisation of an interpreter intended for the implementation of high level procedures on a computer network. IFIP working conference on command languages, Lund, (Aug. 74), 7 p.
[14] IRLAND M. - Simulation of ClGALE. CCNG Report E-32, University of Waterloo, (Jan. 75).
[15] DENJEAN F. - Connexion de terminaux a un reseau de commutation de paquets. International meeting on mini computers and data communication, Liege, (Jan. 75), 8 p.
[16] POUZIN L. - Standards in Data Communications and computer networks. Reseau CYCLADES, Fourth Data Communications Symposium, Quebec City, Canada, October 7-9, 1975, pp 2.8-2.12
[17] POUZIN L. - A proposal for interconnecting packet switching networks. EUROCOMP, Brune1 University, (May 74), 1023-1036. Also IFIP WG6.1, doc 60
1. National Physical Laboratory, UK.
2. European Space Research Organization
4. Bureau National d'Information Scientifique et Technique.