________________________________________________________________________________________________

In recent years, several conferencing systems have been introduced and prototyped to improve the effectiveness of person-to-person collaborations. However, the emphasis of much of the work has been focused on user requirements and the effective provision of these prerequisites within a particular forum. They have not focused on the general paradigms that are required for carrying out the wide range of structured collaborations amongst individuals. This research is attempting to define these collaborations, in order to create a unifying multimedia conferencing infrastructure. It is important to note however, that this investigation is not about describing ways in which humans interact and to simulate those behavioural patterns within machines, without really understanding their properties. Instead, it is about the creation of a finely integrated telecommunications infrastructure which is capable of underpinning the required collaborative activities, in a provably optimal manner.

To this end, a detailed examination of the network conditions that such applications must face was performed. This provided the motivation for the following work, which examined application designs for two aspects of multimedia conferencing. These applications are a distributed shared editor and a distributed session directory.

The general design principle of Application Level Framing (ALF) was applied during the design of these applications. It was shown that ALF can and should be applied in a wider context than that stated in the original ALF paper from Clark and Tennenhouse, and that it results in applications that perform well and are very robust to a wide variety of conditions. However it can also lead to designs that are difficult to generalise from.

The design methodology of lightweight sessions as proposed by Van Jacobson and based on IP multicast presents a large design space in which very few points have been mapped. This thesis explored some of this design space.

In designing a shared editor the effects of designing for robustness and redundancy in shared tools were examined, and the conclusion reached that solutions resulting from such design perform well but are very specific to the design task and cannot easily be abstracted. The resulting shared editor has now been in use in frequent multicast sessions now for two years, and works well.

In designing a session directory for multicast multimedia sessions the scaling limits of lightweight sessions were examined through the example of a session directory which must scale at least as well as the sessions it describes, and examine the scaling limits of multicast address allocation schemes. This work will greatly influence the future design of multicast session directories, and we are in the process of forming an Internet Engineering Task Force working group to exploit and standardise solutions based on it.

The thesis concluded with a reflection on the contradictory design goals in conferencing applications; those of producing abstraction layers to allow reuse of code and simplify the design task, and of designing for good performance in distributed over unreliable networks, and attempted to draw some general guidelines for the design of such systems.

The GCCP gateway at the moment is binding a prototype of ITU’s H.323 type of conference and a prototype of T.120 data only conference and MMUSIC’s conference bus. It appears that the set and number of instruction to perform a kind of function in one type of conference varies a great deal with the other type of conference, although the event is exactly the same. The server holds a state table which generates a list of events and instructions. The next generation conference can participate in an existing type of conference via the gateway without too much modifications in the existing protocols. Nadia also has to investigate the best protocol suited for the reliable transmission of floor control messages for the gateway. Ripplecast, Scalable reliable multicast (SRM), Multicast transport protocol (MTP) are some of the reliable transport protocols under research at the moment.

These issues are being addressed in the EU project Prospect, which is examining the integration of telecommunications service and management software in the open service market. A particular business scenario has been adopted that is complex enough to generate realistic service integration problems. This scenario is based on a provider of educational courses that offers a Tele-Education Service to its customers. This service is an integration of three examples of a Multimedia Tele-Service (MMTS, i.e. a distributed, multimedia telecommunications service), each operated by a separate provider organisation. In addition the TES provider subscribes to a general purpose Virtual Private Network (VPN) service that it uses to provide broadband network connectivity between itself, its customers and the MMTS providers. The VPN provider further extends the service chain by using the ATM Virtual Path management service offered by a public network provider, which delivers international links between customer and provider sites. This scenario provides a basis for investigating problems in the areas of inter-domain service management including the inter-domain information flows needed to support service management functions between organisational domains and the interoperability of management functions implemented on CORBA and CMIP platforms.

In particular, UCL has been examining the CORBA-based Service Architecture specified by the Telecommunications Information Networking Architecture (TINA) Consortium and its application to IP-based multimedia services involving multiple provider organisations. Specific areas addressed are:

• The management of customer and service related subscription information distributed across multiple provider domains, and the interaction between subscription management and other service management functions such as those dealing with accounting, fault and performance.
• The application of TINA concepts of access and service session to IP-based services. This uses a web browser as the basis for user applications, with Java applets providing secure service-independent session control and a framework for running further service specific applets that exploit different service mechanisms, e.g. HTTP, RTP or CORBA based ones.
• The development of models for generic Personal Communication Support in this environment. This encompasses support for terminal, personal and role mobility as well as the personalisation of services.
• The application of component-based design techniques to telecommunications management systems to exploit the reuse of components, specifically the application of Design Patterns and the Unified Modelling Language to this problem domain.

These research themes are combined in the support of an international user trial, where service and management systems from multiple provider domains, running at distributed locations in four countries, provide end users with secure and seamless interactions with composite services via workstation- or PC-based web browsers. Services provided to the user are web services, HTTP server control services, multimedia conferencing services and service management services. The participating sites are connected through a combination of local leased lines and international ATM links, which provide the backbone of a managed Extranet. The systems involve a high level of CORBA component reuse as summarised in the

figure where the white block represent reusable, service independent components and the shaded block represent service specific components.

Subsequently in the BT funded MMN (Management of Multiservice networks) project, we have developed a network management system that controls groups of protocol stacks. A central manager uses a CORBA interface to download traffic control constraints into agents managing protocol stacks. Each protocol stack agent joins a multicast group and sends traffic monitoring information to the group; traffic constraints can then be applied to one protocol stack in the light of traffic conditions in other protocol stacks.

1.2.3. RAMS: Resource Allocation Management System

Firstly, UCL has managed to gain detailed knowledge of the traffic patterns on both the Texaco network, and the Abbotsbury web site. Secondly, we have developed and deployed in a real commercial environment a managed protocol stack for the Windows, and commercial Solaris and Linux operating systems. Thirdly, UCL has provided Prism with the knowledge and access to performance related parameters from the protocol stacks present in these operating systems, so that Prism can make use of these for the purposes of load balancing in distributed systems.

Some of the concepts developed as part of the implementation have fed into Prisms offerings to the OMG (Object Management Group) in the area of specifying additional standardised ORB services. Prism Technologies is an active member of the OMG and is plays an active part in the specification of new services within the OMG.

1.2.4. MMN: Management of Multiservice Networks

Using Orbix is a considerable change for UCL. Our development environment in the past has been based around the Internet and OSI network management protocols, SNMP and CMIP. We use SNMP mainly because it is the industry standard, most devices (ATM switches) support it and it is extremely lightweight. We have ordered an evaluation copy of Orbix and plan to spend some time familiarising ourselves with it and making it interface with SNMP. A major reason for this is so that our BT work is integrated with our other projects.

We are currently finishing off a specification that describes our thoughts on traffic management of end systems. This work largely focuses on the management of protocol stacks. There was close collaboration between the BT project and another DTI/EPSRC funded project on this. The DTI project provided a managed protocol stack environment controlled by SNMP agents. The BT project focused on how a policy based management system based on Orbix controls the protocol stacks in hosts.

In previous projects, we have developed interpreters that are integrated into compilers for network management specification languages (GDMO and ASN1) and into runtime management environments to provide mechanisms for remote programming of network management agents. In the BT project, we plan to move to a more standard interpreted environment based around Tcl and various addons. These include itcl which adds object oriented characteristics to Tcl, safe-Tcl which provides safe remote programming and scotty which adds SNMP capability to Tcl. We also looked closely at the hush and Object-Tcl C++ APIs as these appear to provide a much better interface between Tcl and C++.

We integrated such a consolidated Tcl into Orbix. We have already selected a platform (Orbix) and we are currently using this to build a prototype. During year two of the project, we concentrated on developing a prototype of a policy based traffic management system. A full description of the prototype can be found in a document entitled Policy based connection management that we wrote earlier this year.

The connection management system monitors and controls a group of hosts. Network monitoring Information is be gathered from a monitoring service measuring delays to destination networks, from the SNMP agents managing ATM switches which hold statistics about bandwidth utilisation and from protocol stack agents monitoring the performance of each connection in a host. The connection management system then uses this information to allocate and adjust the QoS requirements of applications.

A monitoring service is needed that determines the load to destination networks. We are currently writing a simple Orbix monitoring service that can determine the load to destination IP networks. We hope to collaborate with Loughborough and use their monitoring work in the future. In the section on future work, we outline various requirements of a monitoring service.

The SNMP agents managing the Fore switches that connect UCL-CS to SuperJanet can provide information about the amount of bandwidth utilised on VPs/VCs. For IP over ATM applications which run over PVCs, SNMP can be used to determine which applications are running over which PVC and thus the management system can determine whether a new TCP connection will overload the existing PVC.

Since all IP applications to the same destination run over the same PVC, it is not possible to determine from the SNMP agent managing the ATM switch how much each bandwidth each application is using. This information needs to be gathered from the protocol stack of each managed host. We developed an agent that monitors the performance of TCP-IP connections; this work involves extending RFC 1213 so that it provides per connection information. For each TCP application, the system creates QoS profiles that reflect usage at different times of day and under different network loads.

The connection management system controls the amount of traffic each application can inject into the network by managing the protocol stack. At connection time, the management system selects a QoS profile depending on network load and time of day and according to the dictates of the policies in place. During the call, the system adjusts QoS profiles as conditions change. The hooks which allow a network management system to control the protocol stack were implemented in the RAMS DTI/ESPRC project, in which UCL is involved.

Policy language statements controls how the connection management system selects and adjusts the QoS profiles that can be applied to a connection. We have and will continue to collaborate with Lancaster and OBU on the QoS models used here. We wrote a report in year 1 describing an ASN.1 SNMP MIB compiler and interpreter that we developed in order to try out various ideas about how a (policy language) interpreter can be integrated into a network management platform. The approach we have adopted is to view a policy language as an interpreter for the specification language of a particular distributed programming model; in the case of SNMP, it is an ASN.1 MIB interpreter and the variables of the language are SNMP MO variables. We have had discussions with Imperial about their policy notation; there is definitely a need for a higher level approach to expressing and managing policies. We concentrated on the lower level problems of integrating an interpreter into the network management platform (Orbix); we collaborated with Imperial in looking at the problem of how their policy specification tools can transform high level policies into executable statements.

1.2.5. IBCoBN: Integrated Broadband Communication on Broadcast Networks

IBCoBN is a three year EU funded research project looking at the way that the existing CATV (cable TV) networks might evolve to FSNs. The project involves many European cable TV providers as well as research institutes. The main part of the UCL work is based around a scenario that considers the connectivity that Internet users may have. The fixed CPN (Customer Premises Network) will use a traditional access interface such as Ethernet, while the provider network will provide the IP (Internet Protocol) service on a ATM (Asynchronous Transfer Mode) substrate. The ATM service itself will be provided on the existing CATV infrastructure.

Figure 1.2 shows the "traditional" coax-based CATV networks as they exist in much of Europe today, while Figure 1.3 is a typical example of the way the networks are evolving to HFC (hybrid fibre-coax) networks, by reusing the existing coax architecture with a digital backbone. The use of high-speed SDH (synchronous digital hierarchy) backbones allows for the provision of an integrated services broadband network based on ATM. Figure 1.4 shows that the CATV networks have the raw network capacity in their existing networks to offer symmetrical broadband data delivery to the home.

We note that the CATV network was originally designed for one-way, analogue, broadcast communication, with the users effectively connected to a shared media access network. Now, this same network must be used to offer, multi-way, digital, integrated communication services, while still offering the CATV operators a gradual migration path by allowing continuing use of the analogue broadcasts. However, there are many hurdles to be overcome:

• the CATV industry lacks the level of standardisation seen in the Telecom sector and so there are many incompatible systems.
• although ATM is considered to be the multimedia integrated services carrier for the near future, most applications are written to an IP-based interface.
• users require integrated, data transparent services, QoS (quality of service) guarantees, multicast capability and security features from the network.

To provide the required functions, the CATV industry is working hard develop new standards for use on CATV FSNs, as well as integrate existing standards into the FSN scenario. An area of particular activity at the moment is standardisation of the physical level architecture.

To demonstrate the capability of symmetrical broadband delivery to the home, IBCoBN has chosen video-telephony (VT) as its application. The VT application will be realised using the Internet protocols, so an integral part of the IBCoBN work is to provide Fast Internet services (symmetrical 2MB/s) to the home.

The work of UCL within IBCoBN is to investigate the use of IP, ATM and the CATV technologies in order to provide the Internet connectivity required. So far, UCL has produced output that shows:

• the functional requirements of home Internet users
• an assessment of the network management needs for the evolving CATV FSNs
• an analysis of the protocol architecture for Fast Internet connectivity for CATV networks

The UCL work is continuing with an investigation and analysis of the protocol interactions between IP, ATM and the CATV technologies to provide IP connectivity with QoS and multicast capability.

For more details on IBCoBN please see <http://www.cs.ucl.ac.uk/research/ibcobn/>.

All of this imposes great demands on the management systems which look after such networks. These now have to cope with a much more dynamic environment in which customers’ requirements change daily. They must be able to interact with the management systems of other suppliers in a secure way. Finally, they must be able to deal with new technologies. The MISA (Management of Integrated SDH and ATM) project, which is sponsored by the EU under the ACTS programme, is studying these problems. Both the Computer Science Department and the Electrical Engineering Department at UCL are members of the MISA consortium which also includes numbers of telecommunication operators, manufactures and R&D institutes in other European countries .

In the MISA project, an important concept is the "Global Broadband Connection Management" (GBCM) service which provides a network-technology-independent management service for end-to-end connection. The ITU Telecommunications Management Network methodology has been adopted as a system specification and modelling technology for the provision of this service.

During its first two years the project has completed: studies of the GBCM requirements; functional and informational design of the GBCM system; and system implementation design in detail. Currently the system implementation work is underway with many components ready for integration. In the Computer Science department, we have been responsible for the analysis and design of the fault management information model and the study, design and implementation of Inter-Domain-Routing in the MISA system. This latter routing work is described in Section 1.2.8. Further work is still ongoing to improve the performance of the routing algorithm and protocol.

For further information on MISA please see <http://www.cs.ucl.ac.uk/research/misa>.

1.2.7. INSERT: INtegrated SERvices Management Tools

Service management in a distributed system requires sophisticated applications enabling interoperability between network and system management services provided by multiple vendors. Management information must be integrated, modular and abstract. Integration enables different management functions to share information. Modularity enables distribution of management functions. Data abstraction enables different views of the same data.

A vital aspect of service administration is the establishment of "Service Level Agreements" (SLA) with groups of users. These specify a set of parameters concerning service availability, performance etc. which must be met. The INSERT project is developing a Service Level Agreement Monitor (SLAM) to enable users and managers to monitor their SLAs. A simplified architecture of the SLAM is shown in Figure 1.5.

There are various components (such as a parameter collector) inside the SLAM which allow monitoring of Service Level Agreements triggering of alarms etc. The SLAM interacts with the outside world via CORBA 2.0. A "monitoring adaptor" receives reports from various service monitors and makes information available to the SLAM. Users and managers may access SLAM information from Graphical User Interfaces (GUI) on WWW browsers, again via CORBA. The SLAM also has adaptors to legacy systems such as proprietary network management systems and has access to a database via SQL.

UCL is working on a general architecture for CORBA-based user interface support and this will be applied to the SLAM. Within this architecture, a system which needs a user interface structures the information it wishes to expose as a set of "user interface objects". Collectively these constitute a simple, active, object-oriented database which user interface software may access via CORBA. This scheme effectively de-couples the user-interface software from the rest of the system making it easy for the user interface to be developed or modified independently.

UCL is also developing mobile-agent-based monitoring tools to provide some of the monitoring capabilities required by INSERT. Monitoring activities change in time. By implementing monitoring tasks as mobile programmable agents, if monitoring activities or parameters change, new agents can be programmed and distributed with minimal impact on existing monitoring activities. This work is described more fully in Sections 1.2.9 and 1.2.10.

In the process of setting up a B-ISDN end-to-end connection with required QoS, the routing issue is fundamental. We need to find a path meeting various QoS constraints imposed by users and achieving good network performance. This is difficult to achieve, especially in a multi-provider and multi-domain environment.

Under the ACTS MISA project (see Section 1.2.6), which addresses the provision of a Global Broadband Connectivity(GBC) service over a hybrid ATM/SDH network, research work was started by studying the requirements for QoS inter-domain. Apart from the functional QoS constraints imposed by the user, there are also non-functional constraints in that the customer may express preferences for routes which favour or avoid certain network providers or countries. Requirements also come from the network operators’ side in that the routes chosen should make optimal use of network resources and that the routing algorithm should not require disclosure to competitors of commercially valuable information. Although B-ISDN originated within the telecommunications community it owes much to data communications practice. Therefore, following the requirements capture, a survey on network routing technologies for traditional telephone, INTERNET and ATM networks was carried out. Following this we have developed a model of inter-domain routing that can work within tight real-world constraints:

To achieve globally optimal routing requires accurate, up to date and global access to the link state of an entire network. This may not be a practical proposition due to the amount of information to be collected and the frequency at which this information becomes out of date. The problem is compounded by the unwillingness of commercial operators to give away information that reflects the internal configuration of their domains. We have proposed a model whereby the cross domain paths (calculated by the domain internally and asynchronously) are stored in a database which is updated periodically. The accuracy of this information is dependent on the frequency of internal updates, but we assume that these routes are valid until informed otherwise by the domain (e.g. if a link goes down). These pools of partial routes provided by each domain can then be used to provide path elements of global end-to-end routes.

The MISA environment offers QoS based routing as needed by many real-time applications. The inter-domain routing component achieves QoS based routing by a statistical, fuzzy rule based system that prunes network topology graphs and thereby maximises the likelihood that generated topological routes will also be QoS constrained.

A simplified version of the algorithm has been implemented and will be tested within the MISA demonstration system.

Although attractive in concept, many reported applications of mobile agents seem to perform tasks which could be done just as effectively through a conventional client-server model. We believe that network and distributed systems management is one field in which mobile agents really can make for more effective solutions:

• Monitoring activities are highly location sensitive. Moving the monitoring task closer to the monitored system reduces network traffic and improves real-time response. Moving the monitoring task close to the client systems gives the same view of the network that the client sees;
• Control activities need flexibility and autonomy. If the system to be managed can host mobile agents we achieve flexibility through our ability to send modified agents as needed. We achieve autonomy since the mobile agent can continue to act even of the network and everything else has failed.

This year we have carried out a detailed study of the requirements for a monitoring and management system based on mobile agents and have completed an initial design study for a test-bed system. This test-bed will support mobile agents written in Java. We will use the Aglet Workbench from IBM which is specifically designed to support such agents. (http://www.trl.ibm.co.jp/aglets/index.html) Aglets are genuinely mobile; execution can pause, the aglet source and its state can move, execution can then resume.

Initially we will use our test-bed to build a monitoring and alarm system for PCs and workstations. Mobile agents, in the form of aglets, will be sent to these machines in order to carry out monitoring tasks. These agents must be able to access information about resources on their hosts - for example the current processor load, free disk capacity and so on. Existing management systems (SNMP, DMTF etc.) have already defined abstract models for this kind of information. We will embody these models in a series of "resident aglets" which will live permanently on the participating systems (see Figure 1.6). Visiting aglets will communicate with these resident aglets in order to access information about the local environment.

We expect this work to be deployed within the INSERT project (see Section 1.2.7) and to make use of the theoretical work described in Section 1.2.10).

Since the beginning of the 1990s the application of Mobile Code Paradigms to Network and System Management has been investigated from different perspectives and its potential impact has been described qualitatively. In particular, Mobile Agent (MA) technologies have been looked at in this context, and it has been shown that the use of MAs to distribute and delegate management tasks is a particularly promising approach to dealing with the limitations of current centralised management systems which appear to be lacking of flexibility and scalability.

Thus, it has now become evident that this merge between mobility of code and management can affect profoundly the way management is done, enabling a dynamic distribution of management intelligence, and leading to a more rational exploitation of the newly and increasingly available technological resources.

However, despite the great deal of interest among researchers, this merge still poses several interesting, as well as hard, problems to tackle. For example, mobile code-based management systems promise to be quite complex both to design and maintain, in contrast with the relative simplicity of static centralised or distributed models. Furthermore, additional security issues have to be considered too and the actual advantages, in terms of performance and scalability, still have to be proven quantitatively.

During the last year we have been investigating a specific aspect of the above problem, that is the use of MAs to distribute and delegate monitoring tasks. Our research hypothesis is that MAs can be used for network and system monitoring and can improve the performance and scalability of a class of monitoring tasks. In order to prove that we have first adoped a dynamic hierarchical management model based on a delegation paradigm. Then, we have defined an MA-based functional architecture suitable for monitoring operations. Finally, we have considered possible uses of the proposed model and architecture for pursuing seamless and timely monitoring.

Our current activity is aimed at a more formal definition and modelling of a delegation-aware monitoringmonitorin system, that is a system in which it is possible to define a monitoring task in a "distributable" and "delegatable" way. Our goal is to show, through modelling and simulations, that performance and scalability can really benefit from delegation. Another objective it to identify the class of monitoring operations that can benefit most from mobility of code. Finally, a bench-marking methodology for a distributed monitoring system will have to be defined in order to assess our models and to carry out a comparative study between the proposed management model and the conventional centralised or statically distributed ones.

Most of the current work in the Internet community related to application QoS requirements revolves around the idea that the network will be able to provide mechanisms to fulfil the needs of the applications, e.g. by the use of resource reservation mechanisms. Such enthusiasm for moving intelligence and functionality into the network is reflected by network operators and service providers because this is a service they can charge for. This is not unreasonable as it means that the end-systems can be made less complex and so less expensive. However, there may be cases where the network cannot provide the connectivity requirements for an application, or does not have resource reservation functionality. In such cases, the application should dynamically reconfigure itself (under user control) so that it may best use the network resources that are available.

We model the application as having i discrete QoS-dependant states, s_i, that form the set S. The states in S are each defined in terms of N QoS related parameters. The corresponding network model that is being investigated can be described in its simplest terms by the following statements:

• In general, the QoSInfo (QoS-related information) that we have within our network model is represented (in part) by some vector q. The vector q includes information about N the parameters. As the N parameters change over time, we can think of q as moving through an N-dimensional space, QoSSpace. Within this space we may define some N-dimensional hyper-planes (or volumes) that represent the thresholds between the states, s_i. As q moves through QoSSpace, the application would like to know when the position of q represents a change of state for that application, i.e. when q crosses a hyper-plane (or volume surface).
• The QoSSpace is represented in the network model. Within this model the parameters relating to the QoSInfo are maintained in q and the QoSInfo itself is embodied in the expression F(q). F(q) represents the QoSInfo required for that application to make state adaptations. The QoSInfo, F(q), can be thought of as a "cooked" version of the "raw" data that relates to measurements of various parameters relating to the network, e.g. throughput, delay, delay jitter, error rates etc. The raw data may be sampled, summarised or transformed in some manner in order to provide the QoSInfo.
• The movement of q through QoSSpace is communicated to the application through a well-defined interface.
• The vector q is updated by information from the network, however the interface between QoSSpace and the network will not be researched or defined in this work. What will be investigated are the operational requirements of any mechanisms that may provide such an interface and how they affect the QoSInfo and the network model.
• For the application, the changes of state that occur form s_m to s_n, can be seen as an asynchronous process s_n = A(s_m, F(q)), where A is some function that represents the adaptation capability of the application. Hence, the primary operational goal of the network model can be considered to be the capability to inform the application as to when it needs to make changes of state. Information about where the state change thresholds occur (in terms of hyper-planes or volumes surfaces within the QoSSpace) are used to configure the operation of the model.

The work is currently at the stage of developing, simulating and testing the QoSEngine that forms the heart of the network model. The QoSEngine will use the QoSInfo to inform the application of the confidence it has that the network can currently sustain each of the states in S. Figure 1.7 shows an overview of the model. Traditionally, S, F and A, would have been realised by applying standard control theory techniques and mathematical models but, given the unpredictable and often immeasurable nature of the Internet, we are currently investigating the use of techniques based approximate reasoning and fuzzy logic.

There are problems with this architecture:

• If the manager-agent dialogue is intense or verbose then a great deal of management traffic is imposed on the network;
• Continuity of management requires the continued operation of the network itself and of a relatively few managing systems.

Thus there is need for distribution of management responsibility. Hierarchical models like the Telecommunications Management Network (TMN), distribute management functionality in layers but responsibility remains centralised.

Management by delegation was introduced in 1991 by Yemini et al., where management tasks are delegated as programs and are executed close to management data in the agent nodes. We have investigated how delegation may be added to the TMN model. We have introduced the concept of "Active Managed Objects" (AMOs). AMOs are managed objects which communicate with an interpreter which can execute management "scripts". These scripts contain management commands which act on local managed objects. In effect they perform the managing rôle at the agent site. AMOs include attributes to control the execution of scripts and the way scripts communicate with the local environment. Scripts can be changed at run-time as they are included in attributes of the AMO. Once a script is loaded and the appropriate conditions that trigger its execution are set, there need be no communication with a remote managing entity. Management traffic is, therefore, minimised and managed systems can be autonomous. The concept can be extended to allow scripts to communicate with remote managed objects via CMIP, allowing easy implementation of hierarchical management. This and other potential extensions to functionality are determined by the capabilities of the interpreter and are independent of the AMO design.

During the last twelve months we have finalised the design of an AMO system and have validated this design by the development of a simple prototype within the OSIMIS system. In particular we have investigated the execution model for scripts and the way they receive and act on local event notifications. The prototype embodies an interpreter for the Tcl language which we have extended to provide access to local managed objects. This system has been tested in several scenarios from the TMN, as well as the SNMP worlds. Results include the measurement of the amount of traffic the downloading and initialisation of the script require, as well as the time the managed system spends in executing scripts.

In the telecommunications industry management has traditionally been the subject of widespread standardisation efforts. These efforts have to date been aimed mainly at the management of network elements and networks operated under a single administrative domain. The results have been usually large complex interface specifications were only the syntax rather than the semantics of the interface are defined formally. In practice these interfaces have been found difficult to use and/or inappropriate to implement in their entirety, a situation complicated further by the number of overlapping standards and the optional aspects of those standards. This has prompted groups like the Network Management Forum working on solutions sets that take existing standards and specify the semantics of how they can be applied to specific real life problem areas.

The market for services based on open interfaces, however, will require the process of service development to be conducted at a much faster pace than current interface standardisation process allows. For service management this implies a move from management components with syntactically complex interface definitions, to smaller problem-oriented components of management functionality. The specification of these components, whether developed as a common service by an industrial forum or for a component sold for use in the development of other services, must be more self contained, with the semantics that need to be understood to use the component being expressed clearly. If large numbers of service components are being used to develop an ever growing range of further services then a large productivity gain will be gained from expressing the full semantics of the interface to service components in a common way.

This work aims to gain a better understanding of how the development of service management services can be improved in this way in the context of distributed object oriented technologies. This is being performed primarily by analysing, implementing and assessing software components for use in the Prospect service management. The expected result will be a set of abstractions and guidelines for developing modular, object-oriented service management components in a way that enables easy and rapid reuse of the components. This should support the specification of new service components consisting partially of existing interfaces, i.e. reuse of component design. An important aspect of this will therefore be a unified approach to developing management services for both existing TMN technology and emerging DPE/CORBA technology. It is anticipated that the useful specification of management service components interface semantics will involve augmenting existing mechanisms for describing interface syntax, i.e. GDMO, IDL, ODL, with some common method for expressing interface dynamics, conditional behaviour and component inter-dependencies in the form of a consistent object-oriented model.

• TCPng - Next Generation Transport Protocols (with particular attention to support for HTTP);
• IP over ATM and other committed information rate subnet technologies;
• Semi-reliable Multicast protocols, for news and similar information dissemination applications, as well as possible future application areas (e.g. multi-player systems);
• The WWW evolving systems architecture, generally.

VITAL is an ACTS project has as objective to demonstrate and validate the development, deployment, management and use of complex heterogeneous service features in an Open Distributed Telecommunication Architecture (ODTA), based on extended TINA concepts. The service features that are considered in VITAL encompass multi-media, multi-party, mobility (personal, session and terminal) and IN-like supplementary services. The service features are offered within the context of an integrated teletraining application that includes video-conferencing and distributed slide presentation facilities. The VITAL project started in November 1995 and is organised in three phases, each lasting for one year. The second phase was completed at the end of 1997, with a major demonstration of the integrated application and the supporting ODTA components. The demonstrator comprised sites in Belgium (Brussels, Belgacom), Italy (Turin, CSELT), Spain (Madrid, Telefonica) and Portugal (Aveiro, CET), interconnected over the JAMES Pan-European ATM network.

In each phase, there is a significant element of research, design and validation. Given the fact that the TINA architecture and specifications are far from being complete, at UCL we have been able to undertake interesting research activities. These have concentrated up to now in the areas of ATM Resource Configuration Management (RCM) (see Section 1.2.17) and the overall network management architecture of TINA; in issues behind re-using Telecommunications Management Network (TMN) specifications and existing TMN applications in TINA (see Section 1.2.18); in issues of both inter- and intra-domain Accounting Management (AM) (see Section 1.2.19); and in the performance evaluation of the supporting middleware which is based on OMG CORBA. In the third phase of the project, we will be expanding our activities in the areas of TINA and Internet co-existence (see Section 1.2.20), since TINA currently assumes a ubiquitous ATM infrastructure. Research activities in the context of VITAL/TINA are described in separate sections.

Realising the need for combining the functional capabilities of the control and management planes in ATM networks, the ACTS REFORM (REsource Failure and restORation Management) project concentrates into the research, specification, design, implementation and trials of an integrated system trying to optimise ATM network performance and maximise its availability. The REFORM system aims to provide a complete, network-wide solution to the problem of network availability and performance under both normal and fault conditions. Real-time control plane functions, such as route selection and self-healing mechanisms, are integrated with higher level network-wide routing and resource management functions which have a global view of the network and a longer-term perspective of operation.

The REFORM project is to some extent the continuation of the RACE Integrated Communications Management (ICM) project, whose research results have been reported in previous research reports. While ICM was based exclusively on the TMN framework, REFORM combines ATM signalling based on the emerging ATM-Forum Private Network to Network Interface (PNNI) specifications and management functionality that follows the TINA architectural guidelines that suggest a CORBA-based Distributed Processing Environment (DPE). The first phase of the project was completed at the end of 1997, with a demonstration of the features of the REFORM system operating over the experimental ACTS EXPERT testbed in Basel, Switzerland. The demonstration comprised setting-up ATM switched virtual channels through the project’s signalling solution, managing and optimising the allocation of bandwidth to ATM virtual paths to meet the targeted performance characteristics of the ATM service classes and restoring faulty connections in real-time through the back-up virtual paths.

Our contribution to REFORM concentrated mainly in the Resource Configuration Management (RCM) area, which covers the management view of both the static and dynamic resources (see Section 1.2.17). We have also provided solutions for re-using TMN object-oriented specifications and components, as described in Section 1.2.18. In the last phase of the project we will concentrate mostly on algorithms and scenarios for ATM virtual path connection bandwidth and routing management and also performance verification. Research topics related to the REFORM project are described in separate sections.

While TINA has produced relatively mature specifications for Connection Management, there is still a significant amount of research work to be undertaken for RCM before specifications and prototypes can be completed. UCL is undertaking this research activity under the auspices of the VITAL (see Section 1.2.15) and REFORM projects (see Section 1.2.16).

The Overall Architecture defined by TINA is decomposed into the Computing, Service, Network and Management Architectures, and each of these comprise Service, Resource and Element layers. Consequently there are four possible applications of RCM - one to each of the four architectures. This initial result of our research at the architectural level is a major extension to the TINA work which had never considered the new sub-area of Management RCM (MRCM). It is now possible to position some of the TINA defined components such as the Connection Management Configurator (CMC) within the scope of MRCM where previously it was always unclear where components such as these should be placed.

To date, we have concentrated on Network RCM and Management RCM building on the preliminary work done by TINA-C to produce information and computational specifications that reuse, as much as possible, experience and techniques from the more mature TMN research work performed by UCL in previous projects. A number of research topics related to TMN to TINA migration have arisen during the course of this work, and have identified many areas where Open Distributed Processing (ODP) architectures and design methodologies can benefit from the wealth of knowledge and techniques from the world of OSI management and TMN in particular.

An important part of our work has been in the area of the Network Resource Map which is responsible for maintaining a model of the managed network in terms of inventory, relationships and state information. We found the computational and engineering techniques of TINA and ODP to be inadequate for an "intelligent database" component such as this, and our solution was to use the OSI System Management approach for modelling the network. Our initial prototype provides a capability to reuse OSI-SM agents over CORBA, capturing the full expressive power of the latter for multiple object access and retrieval (see also Section 1.2.18).

Since TMN standardisation has already come a long way towards providing solutions for open interoperable management while TMN systems have started being deployed, it is of paramount importance to maintain and re-use the relevant investment. In addition, it is absolutely necessary to support the same powerful facilities provided by OSI System Management over the TINA Distributed Processing Environment (DPE) e.g. multiple object access, fine grain event reporting through filtering, naming and relationship services, generic facilities for monitoring, testing, scheduling etc.; all these are fundamental for broadband network management (SDH/SONET, ATM). The X/Open Joint Interdomain Distributed Management (JIDM) group has produced guidelines for the translation between the OSI/TMN Guidelines for the Definition of Managed Objects (GDMO) and the OMG CORBA Interface Definition Language (IDL). These allow for the re-use of existing GDMO information models in TINA and make possible the construction of generic adaptors between CORBA and OSI System Management; on the other hand, the previously described power and expressiveness of OSI System Management is lost.

As part of our research in the VITAL (Section 1.2.15) and REFORM (Section 1.2.16) projects, we have specified CORBA-based management brokers that mirror the facilities of OSI SM. These may administer clusters of other CORBA objects; they may also act as generic adaptors between CORBA clients and TMN applications in agent roles. In the former case, the TMN methodologies for producing object clusters or ensembles may be fully re-used in TINA; in fact, this is how we approached the TINA Network Resource Map (NRM) described in 1.3.14. In the latter case, existing TMN-capable elements and applications may be also re-used in TINA systems. We have implemented a first version of such a management broker using the OSIMIS platform and the Orbix implementation of CORBA. While there is plenty of ongoing research regarding TMN to TINA migration and interworking, we believe our approach retains the relevant advantages of TMN for network management while it is fully complementary to the JIDM approach. In addition, this is a viable path for gradually migrating existing TMN systems over CORBA-based DPEs. We have proposed our approach to standards bodies such as TINA, JIDM, OMG and the ITU-T Study Group 4.

In the VITAL project (see Section 1.2.15), we have investigated issues of accounting management in the TINA context, concentrating mainly on architectural issues rather than the effective ATM network usage. The TINA business model identifies various different connectivity providers (CPs) and retailers that may be involved in providing an advanced service. Federation takes place "horizontally" both in the CP domains, for end-to-end connections supporting services, and in the retailer domains for end-to-end value added services. In addition, there exists also "vertical" federation between the retailers and CPs, when the former are requesting connectivity resources. This complex decomposition of TINA systems makes the provision of flexible and efficient accounting management facilities an interesting challenge. An important target for the design of the relevant system is that it should be generic i.e. exactly the same accounting management system should be applicable to the various CP and retailer domains.

We started by considering as input the existing TMN model and specifications for usage metering, which provide a generic approach for collecting usage metering information. This generic model can be applied to the various TINA domains, with specialisations relevant to the nature of each domain. By designing the system in an object-oriented fashion, inheritance and polymorphism have been used to provide these specialisations in an extensible manner. The next key issue relates to the communication of accounting management information between domains, and we have identified two major models:

• accounting information is kept in each domain, to be combined later with cascaded management interactions for charging and billing; or
• accounting information is passed across the inter-domain reference points immediately after the relevant service instance terminates.

The first approach has the advantage of "lazy evaluation", which results in less traffic in busy times. The disadvantage is though that additional information is required to be passed to federated domains by the "master" domain in order to be able to relate the usage of supporting services, such as connectivity, to the value added service and the corresponding end-users. The second approach has exactly the inverse qualities. Having evaluated carefully the two approaches, we have opted for the second one, which results in easier bill control at the expense of additional traffic incurred by accounting management when a service instance terminates. We have produced a relevant computational and information model, which was subsequently developed and deployed in the VITAL project’s demonstrator. We intend to pursue further issues related to accounting management in the future and in particular the combination of the proposed model with effective ATM network usage.

• Communication session control in a connectionless environment.
• Communication session control in an IP/IP-multicast environment.
• Interworking between the TINA call-model and the Internet protocols for session invitation (SIP) and session announcement (SAP).
• Relationship between the Internet session description protocol (SDP) and the TINA session attributes.
• Integration of unicast<>multicast gateways in the TINA architecture for controlling the participation of IP-based terminals to TINA sessions.
• Use of existing Internet applications (vic, rat and sdr) in this environment.
• Demonstration of the synergy between TINA and the Internet.
• Justification of the added value of TINA to the Internet and vice versa.

The network architecture that is used in this activity is shown in Figure. All terminals are connected to a multicast core by access networks via a Unicast Transcoding Gateway (UTG), which is discussed in Section 2.1.1. The multicast core network is based on IP-multicast over ATM. The access networks can be ISDN connections, Ethernet or other. LEARNET may be used as the core multicast network.

There are three basic assumptions in our work on the synergy between TINA and the Internet:

• All the IP hosts are unaware of TINA control. They only contact the UTG for accessing services and creating/joining sessions by using Internet protocols.
• The TINA control covers the core network and the UTG. There is no TINA control on the IP terminals.
• A minimal number of changes can be applied to the TINA architecture. The reuse of existing TINA implementations in VITAL will be maximised.

We simplify the problem by requiring user applications to declare the price they are willing to pay for individual QoS requirements (DJLT) for each of a range of service streams (ftp, video, audio, telephone, e-mail, etc.) they may wish to receive during a session. This declaration may be done on a once only basis, for example when the mobile terminal is switched on. Thereafter a bid manager process on the wired side of the air interface uses this information to place bids for appropriate services as and when they are requested by the mobile application. At the other end of the consumer-provider pipeline is the problem of what services should be provided by the base station provider. Should each base station provide a standard range of services, merely reducing the quantity and quality of them across the board as physical limits on available bandwidth are imposed? This would certainly be a simple option. However, adaptive base station strategies which adjust the local base station 'product mix' according to which services are most in demand (and therefore bringing in the most revenue) may be more suitable.

The allocation process itself is a small, simple, single pass auction mechanism that is initiated only when supply exceeds demand, and is designed to ensure users declare the true value of QoS resources requested. There are some simple restrictions necessary to ensure that price fixing cannot occur. Microeconomic theory indicates that resources will be allocated to those most willing to pay for them. Microscopically, the information delays inherent in mobile systems are expected to precipitate a range of stable, oscillatory, and chaotic behaviours; however this may not matter. We intend to test whether small non-deterministic alterations to the bids or service requests submitted by user bid managers can smooth unstable patterns of behaviour.

At present, we are validating a set of simulation requirements to test a simple auction mechanism with a single tier network. Extensions will include multi-tier overlay networks and derivative markets. Within the coming year, we envisage using simulation results as the start points for hybrid simulations and eventual implementation within a Wavelan (TM) wireless network.

1.3.2. PIMMS: Pilot Interactive Mobile Multimedia System

The Pilot Interactive Mobile Multimedia System study has recently started and has, as one of its major aims, the extension of these tools to work within a mobile environment. The major issue raised by mobility is the sometimes dramatically reduced bandwidth between mobile hosts and the fixed network. In the coming year we expect to implement a transcoding gateway, that will use techniques such as frame dropping and image degradation to reduce the size of data sent over the link. Security concerns are also important, since the broadcast wireless environment is inherently less secure than a wired environment. These are actively being considered.

The latest version of the Internet Protocol, IPv6, addresses a range of mobility issues, including security. We hope to set up a mobile testbed using IPv6 and both radio-based and infra-red networking to assess how well our needs can be met. Further, we will assess how our transcoding gateway can best be integrated into the network and how well our security concerns can be met.

The hard disk and the CPU are among the components which consume a substantial amount of power. Previous work on power management has focused on a hardware-based approach. For example, the disk spin-down strategy tries to reduce power consumption by spinning down the hard disk during idle periods. The disk is spun up again on the next access. Alternatively, some approaches rely on an ability to reduce the CPU clock speed, since reducing the clock speed by a factor of n reduces power consumption by a factor of n2. Instead of running a process at full speed, the process is run at a lower speed, which consumes less power. The speed is chosen so that the process finishes executing before its deadline.

The power management strategy that we are developing is based on the concept of load sharing. Load sharing has traditionally been used to distribute workload among hosts in distributed networks in order to improve response time. However, the same techniques also have the potential to reduce the battery power consumed by a mobile node. In order to migrate a process from a mobile host to a host on the fixed network, it is necessary to invest battery power in transmitting the process. However, after doing this, the overall power consumption of the mobile host may be lower, since the mobile host’s CPU can be placed in doze mode and its disk can be spun down.

Previous studies on load sharing have focused on fixed network architectures. Load sharing in a wireless environment differs from this as some of the assumptions typically made in such studies are not valid in wireless networks. For example, due to the limited bandwidth and increased medium access times, communication delays can not be assumed to be negligible, as is the case in many of the existing studies.

Our load sharing algorithm decides whether a job should be transferred based on the following calculations. It calculates the amount of power consumed if the job is to be executed on the mobile host (local execution). It then calculates the amount of power consumed by transmitting and receiving if the job is to be transferred (remote execution). It also estimates the job response time of local execution vs. remote execution. If the calculation shows that transferring the job will save power, and the transfer will not degrade job response time, the mobile will send a transfer request to the base station. The job is transferred if the request is accepted, otherwise, it is executed on the mobile.

Simulation results show that load sharing can extend battery lifetime. Three factors that influence the effect of load sharing have been identified:

• Available bandwidth. As the available bandwidth increases, the cost of transmitting/receiving messages decreases and more jobs are transferred. As a result, the CPU is kept idle for longer periods. Consequently, battery lifetime improvement increases.
• Job size. While generally very few jobs can be transferred at low bandwidth, one simulation result shows that if most jobs have small job size, transferring jobs can extend battery lifetime significantly, even at low bandwidth.
• CPU utilisation. Users with high CPU utilisation are more likely to extend their battery lifetime by transferring jobs. For users with low CPU utilisation, it is cheaper to execute the jobs locally.

There have been two approaches to the deployment of wide-area mobile networks. These are cellular circuit switched networks based on assigned channel scheme and cellular data packet networks based on random access scheme. Our main research interest is on resource management technique for the cellular data packet network.

If is reasonable to expect that base stations will be bottlenecks, owing to data packets originating from sources in wired network which are sent toward mobile hosts, if the required throughput of the flows is higher than the wireless link capacity. That is, many packets from wired networks sent at high speed will be buffered at the base station that connects to wireless link with lower throughput and higher delay characteristics. On the other hand, when a mobile station sends data packets to a destination in a wired network or another cell, it has a trouble forwarding them to its base station if the current wireless link is congested. Accordingly, a cellular packet switched network should have a mechanism for settling the enequalities between wireless links and wired ones, in order to reduce the congestion. Moreover, in the context of a cellular network, mobility that complicates the matters.

To solve the problem, we are developing a new fair queuing scheme, which can simultaneously schedule three classes of flows – absolute class, weighted class and best-effort class – according to the policy for the classes under a variable-rate link. The absolute class flows are guaranteed a minimum bit rate service even when the available bandwidth fluctuates, provided that the available bandwidth is higher than the total bandwidth allocated to the absolute class flows in a busy period. The weighted class flows can receive a rate-based scheduling service within the range of residual bandwidth after serving the absolute class flows. The best-effort class can exploit any remaining bandwidth following servicing of absolute and weighted class flows.

We adopt two kinds of virtual time, absolute and weighted in order to track the bandwidth usage of the absolute and weighted classes respectively. The delay time based on the bandwidth shortage is introduced to guarantee the bandwidth for the absolute class even under conditions of fluctuating available bandwidth. A packet by packet version of this scheme is also developed. The performance is evaluated empirically using OPNET simulator. The analysis and simulation results showed this queuing method to be especially suitable for providing a guaranteed fixed bandwidth service in a cellular packet switched network since the service will be unaffected by fluctuations in the available bandwidth while the low service speed of the wireless link will bound the computational load.

The proposed scheduler can not solve the uplink bandwidth distribution. For the matter, we are planning to develop a multiple access protocol which can fairly allocate uplink bandwidth to the mobile hosts based on their share rates.

In this work, we are seeking to examine the application of user-centred design to the development of software engineering methods, tools and end-user applications, with particular reference to mobile multimedia systems. The question we are seeking to answer can be summarised as: ‘What are the particular software & usability engineering issues that arise when considering mobile communication systems, and how are these to be addressed?’ An understanding of expected user tasks will shed light on the minimum required quality of service (bandwidth, response times, latency etc.), and hence the requirements of the lower level network layers and infrastructure. The consequent lessons for software developers, will be identified.

We are currently developing a hybrid simulation, which will allow us to model and assess the effects of the physical limitations of mobile systems upon the user and to determine the extent to which the user should be troubled for guidance on dealing with heterogeneity.

A Distributed Trust Model has been developed. Its major properties are decentralisation, generalised trust information and a recommendation protocol. This allows clients to ask trusted entities for recommendations on currently untrusted entities, for a wide range of operations. Trust values range from no opinion and minimal to complete, similar to the system used by PGP. Trust relationships are always between two entities, but may be non-symmetrical. In fact, DTM does not constrain the establishment of such relationships, which can exist as a web of trust or hierarchical structure.

The recommendation protocol also allows trust queries to be forwarded to other trusted recommenders, forming a transitive trust chain. The system allows the simple refreshing and revocation of recommendations, which also occur in a distributed manner.

Further work will examine the calculation of trust values, and will aim to simulate the model to analyse its behaviour and robustness.

One of the major challenges in designing a routing protocol for the Ad Hoc networks stems from the fact that the network topology can change frequently. Furthermore, in a unique wireless environment like this, there are constraints to be considered such as the limited battery power and the scarcity of bandwidth and computing resources. Therefore, conventional routing protocols which employ periodic broadcast are unlikely to be suitable. Consequently, we are investigating how features of the link-state and distance-vector protocols can be adapted to solve problems in the wireless environment.

Our main objective is to design a routing protocol that will consider the constraints mentioned and will allow hosts in an Ad Hoc network to exchange data by dynamically setting up and maintaining a route between hosts that want to communicate. At the same time the protocol should be capable of providing efficient and high throughput communication among Ad Hoc mobile hosts. We then intend to extend our protocol to support multiple metrics for finding the best/shortest route but without compromising its convergence, and to investigate how GPS information can be used to improve the performance of the protocol and compare our protocol with existing routing protocols for Ad Hoc networks.

We are currently in the stages of implementing our proposed routing protocol using Maisie (a C-like discrete event simulation language) and we expect to produce our first results in the next month.

1. Location tracking
2. Authentication
3. Billing

Alternatives to the HLR/VLR scheme for location tracking are currently an active area of research. The proposed mechanisms have the aim of decoupling the user address from the service provider, thereby providing life long addressing for users. However, for accounting and billing to occur there still exists some fixed location where the user’s identity can be authenticated and the payment recovery processed. For example, users of GSM cellular networks are currently able to move between networks (called roaming), and still make and receive calls when away from their home network. However, this functionality is conditional. First a roaming agreement must exist between the network providers. Second, the roaming service must be enabled on the account prior to use outside of the home network. Not surprisingly the costs of roaming have to be met by the end user. This can include both a charge for enabling roaming and additional premiums for processing calls made while on another network.

It is argued that with a wide variety of users, having different service requirements, the single provider model is inadequate. Also, no longer will mobile devices be just a means to access the network service, but will be computing devices in their own right. This implies that the relationship between the device and network is de-coupled. Additionally, with the development of electronic payment mechanisms the need to have a home network for payment processing is removed. Instead of being tied into long term commitments, users should be able to choose services and networks for as long as they need them.

The work being carried out is examining an alternative to the home based model. The model is based on the notion of tickets that allow mobile users access to services without requiring advance arrangement or long term agreements. Under this model a user can arrange access to a service as needed and for the period or quantity of resource needed. It should be noted that the term "service" not only covers provision of a wireless network connection, but also application level services such as news/information services. An example is the mobile user who prints a report on a printing service without having arranged this before leaving home. Selection of which service provider to use and how long this provider will be used for, are made at the time that the requirement arises. Effectively the notion of a long-term account for network services no longer exists. Service access is arranged and paid for as required.

A paper describing the structure of the ticket and a protocol for the secure use of tickets was presented at the ACM/IEEE MobiCom’97 conference [64]. Future work is concentrating on two issues: the need to control ticket transfer between users (i.e. prevent users duplicating tickets), and the need to support the concept of service quotas (i.e. limits of the quantity of service associated with a ticket).

This work is supported by an EPSRC Industrial CASE award in collaboration with Hewlett-Packard Laboratories in Bristol.

We are taking four steps to give an answer to the above questions. The first step is to decide what mobility exactly is and what different kinds of mobility can be identified. The second step is to define an abstract network model over which an abstract service is offered. The third step is to define a set of features which the abstract service must have in order to provide certain kinds of mobility. The set of features that make a service mobile is described as an abstract mobile service architecture. The validation of this architecture will be the last step.

The Open Distributed Processing reference model (ODP-RM) was found suitable for describing this architecture in and object-oriented way and from a set different viewpoints: enterprise, information, computational and engineering.

In the past year our work was initially focused on the first and the second step of our approach. Different kinds of mobility were identified and an abstract network model as described. Later, we shifted the focus to the information and computational viewpoints of our architecture. The information viewpoint is described by using the Unified Modelling Language and it concerns allocation of unique IDs and the information needed in order to perform location management. The computational viewpoint is concerned with the computational infrastructure required to provide mobility. The computational viewpoint is described by using design patterns.

This work will continue with refining the information and computational aspects of our architecture and with addressing the enterprise and engineering viewpoints. The completion of the architecture will be followed by a case study of applying the architecture to the deployment of a specific mobile service. Apart from that, information and computational aspects of the architecture are being used in the design of a Personal Communications Support (PCS) service that will be introduced in the service environment of the European ACTS project Prospect.

A paper relating to this work [56] was presented at the IFIP/IEEE International Symposium on Integrated Network Managment ’97.

We devised a hierarchical version of Protocol Independent Multicasting (PIM) to address the Routing Protocol Core placement problem, This was fed via the Inter-Domain Multicast Routing (IDMR) WG of the Internet Engineering Task Force (IETF) into the design of baseline PIM, and into Core-based Trees (CBT) – cf. Ordered CBT – below.

We have revised the Ordered Core-Based Tree (OCBT) work from UCSC which takes ideas from H-PIM, and applies them to CBT, removing looping problems, and possibly improving scaling and core placement. This was reported in IDMR WG at Memphis.

We have implemented algorithms for building shared trees using a one-many joining mechanism. It gives a QoS choice of tree for Delay or Throughput. It keeps track of the routing trees, and prunes the states – thus reducing updates [16].

We implemented low-level traffic version of Scaleable Reliable Multicast tool – used in the Network Text Editor (NTE). [M.Handley et al: Network Text Editor, presented at SIGCOMM’97].

We have developed B-MART – a bulk transfer multicast protocol, being proposed for pre-loading web mirror sites, and VRML world databases. We have made both an implementation and simulations [125].

We completed the Session Directory, and Invitation protocols without the incorporation of security, together with releases of implementations. [M. Handley: Session Invitation Protocol, Internet Drafts draft-ietf-mmusic-sap-sec-01.txt; M. Handley: Session Directory Announcement Protocol, draft-ietf-mmusic-sap-sec-02.txt]

We completed a first draft of the Session Directory Security Addendum, which included mechanisms for using both symmetric and public key encryption for authenticating and making private announcements. We released an early implementation with symmetric encryption. [P Kirstein et al.: Specification of Security in SAP Using Public Key Algorithms, Internet Drafts draft-ietf-mmusic-sap-sec-00.txt.]

There has been an on-going activity on audio systems over Multicast network, which goes beyond the work contracted in this project into the application area. However, optimising audio over multicast networks requires considerable activity on the network interface and in the application taking note of the network performance. Hence we believe that this activity is an important aspect of the work on this project (though most of the effort actually comes from other projects). There have been several publications illustrating this activity; some are mentioned here, In addition to extensive work on the algorithms, there have also been implementations and user trials. An important aspect of the work is forward error correction (FEC) on a talk-spurt basis, where the mechanisms used take full advantage of the nature of the errors found, and characteristics of losses in packet-switched networks [114] [108].

There has been substantial activity in use of multicast multimedia over heterogeneous networks. This has included the following: implementing filtering gateways to allow users on lower performance networks like the ISDN to join high quality Mbone conferences; use of Direct Broadcast Satellites (DBS), which have the property of requiring a different back-channel for routing; use of higher performance Mbones, which can be separated from the general Mbone. Much of this work is on-going, but an overview has been written up. There are already implementations of the aspects described here. The description also covers the work done with private session announcements – mentioned above, and with the secured versions of the basic media tools. Most of the tools have been made operational both over UNIX and Windows 95/NT, though the UNIX versions have higher performance on the same platforms [109].

A paper describing the work [16] presents a new approach for building shared trees which have the capability of providing multiple routes from the joining node onto an existing tree. The approach follows a design parameter of CBT and PIM in that it operates independently of any unicast routing protocol. However, a paradigm shift is introduced such that trees are built in an on-demand basis through the use of a one-to-many joining mechanism. In addition, the paper presents optimisations of the new mechanism to help constrain its impact in the case where many receivers exist for a given multicast group.

High performance networks are moving towards a seamless integration of switching and routing technologies. There are a variety of proposed approaches including Ipsilon's flow labelling, Cisco’s tag switching, Toshiba and IBM have looked at other approaches still. Each involves dynamic creation of state and state table indexes for a flow (whether on demand, as in Ipsilon’s case, or per route as in Cisco’s proposal). Each has scaling properties, in terms of the amount of state, and the rate of change of state, as well as in the number of messages and changes to the signalling and routing protocols in the IP (RSVP) and ATM (Q.2931/Uni 4 etc) (or other) layers. These scaling properties are dependent on a number of actual traffic characteristics – e.g. location in space and time of flows (how long a flow between a source and a destination appears to last) (see activity activity).

Recent work on traffic classes and call admission shows that for some classes, aggregate flows may be admitted with higher acceptance probability, and then queued together with a single index for efficiency. The gains here are mainly in the area of switch/router memory (table sizes) and control messages rather than actual network utilisation (although there are modest gains there too).

Multicast means very different things in the Virtual Circuit world and the IP world. How switch/routers approach this is interesting, and differs widely (and depends on the underlying switch facilities to a large extent as well)

ATM and Integrated Services IP provide a variety of traffic classes. [See ftp://cs.ucl.ac.uk/darpa/rsvp3.ps for a detailed tutorial on this]. These classes do not define (prescribe) how a provider decides what percentage of the network each class is limited to. This is a matter for policy. How much we choose to allow of CBR, VBR, ABR etc, or Controlled Load, Guaranteed and Best Effort is a matter of the revenue each class generates, the demand, the service facilities/limitations etc.

Web servers are adding facilities for sizing documents, and including policies and QoS information concerning the retrieval of a document. [See <http://www.cs.ucl.ac.uk/staff/jon/hipparch/hipparch> for details of a Project involving UCL working on http ng and other aspects of this problem]. In any case, such a facility leads to the possibility for scheduling flows (retrievals for near media on demand for example), which can interact with traffic monitoring and so on.

We are interested in open service provision, QoS in routers, IP traffic QoS, measurement and estimation of traffic resource requirements, and end system requirements for multimedia. SDH, ATM and IP all have failure tolerance mechanisms based in alternate trunk routing, PNNI, and a variety of distributed dynamic IP routing schemes. Each has its appropriate timescale and performance.

The main results achieved are the design, implementation and performance evaluation of a receiver driven congestion control algorithm (rlm) suitable for one-to-many data transfer. The algorithm makes provision for the diversity of network paths in a scalable way and presents TCP-friendly bandwidth sharing among different communication instances, competing for network resources. Its evaluation has been carried out both through simulation (using the ‘ns’ simulator) and with a performance evaluation of an implementation of the algorithm on the Internet.

‘rlm’ has been used as a building block for a FEC based bulk-data transfer protocol and as a transport layer for multimedia streams. Reliable multicast data transfer has been subject of studies too, in this context FEC techniques using ‘redundancy block codes’ has been used, which allow to address the issues of ‘feedback implosion’ and ‘different repair needs’ at the same time.

The most tangible results of this work are the ‘rlc’ protocol implementation (available as a C library, running on Unix in the user space), and the code used for the protocol simulation in the ‘ns’ simulator. Prototypes of a reliable multicast data distribution protocol and a congestion controlled video distribution has been implemented as well. Furthermore the research activity has also produced several papers [125] [120] [173].

Finally, side results of the activity are the debugging of ‘mrouted-3.9’ (popular implementation of DVMRP multicast routing protocol), where a problem in the IGMP implementation of the router was found and fixed, and some bug fixes in the ‘ns’ implementation of routing.

The heterogeneous connections involved ATM links to CAIRN and various European sites, a DBS satellite link to INRIA and also ISDN lines. In the CAIRN link case the deployment and testing of an ATM driver for FreeBSD which was modified incrementally to support rate limiting, Berkeley packet filter, native multicast and pseudo interfaces.

There have been a number of important demonstrations; a surgical workshop, demonstration at JENC 8th, the multicast of SIGCOMM’97, and a demonstration for the European Comission. Also installation and deployment of HP NetMetrix for network monitoring and analysis on Ethernet LAN segments.

1. Novel ways of using multicast technology in the delivery of web pages.
2. The development of dynamic resource reservation protocols to optimise bandwidth efficiency.
3. Reliable multicast techniques that minimise retransimssion bandwidth consumption.

Papers describing the work include [127] [128].

1.6.1. QEDANA: A QoSEngine Enabling Dynamically Adaptable Network Applications

QEDANA is just beginning and is an extension of existing current work that is being carried out within the department (see Section 1.2.11). QEDANA has recently received funding from HP Laboratories Bristol.

1.6.2. LINGO: Laboratory for IP Next Generation Orienteering

The measurement facility will be complemented by a distributed simulation system for network teaching and research based on the ns simulator (http://www- mash.cs.berkeley.edu/ns/) and the Berkeley Network Animator (nam), both of which are being used very successfully in the department. These systems are being developed in a DARPA project on advanced Internet technology, with which we are also concerned in the CAIRN programme. A networked resource based on this software (a web site with common case simulations available via Java applets) and an authoring system both running on hi-end multi-Pentium servers will be set up. User access to these facilities will be through 5 HP Vectra multimedia workstations. The workstations will run the UCL Multimedia applications, and both IPv4, and IPv6 with RSVP etc.

We will ensure that these systems have access to the BT-provided LEARNET (see Section 1.4.3) and the DARPA-provided CAIRN IPv6/ATM facilities (See Section 1.4.1 through our UCL-US ATM link) for wide area experiments. The measurement system will be integrated with the simulation system, which will use the same source code. The wide-area network system will be connected to this laboratory, to allow detailed statistics of system performance to be gathered on real links or logged to be used as trace files for driving simulations.

The laboratory will allow easy set-up of real world experiments, and symbiosis between work on existing network systems (Internet, SuperJANET, LEARNET, CAIRN test beds etc.) and simulations. It will provide an excellent vehicle for projects to be done under an advanced Data Communications Networks and Distributed Systems MSc course.

Delay jitter or simply jitter, is the variation in the overall delay experienced by data or multimedia traffic while traversing a network. It may be described as the non-regular and bursty arrival of traffic patterns at a receiver. Jitter has generally been known to occur at three different places: at the sender, in the network, or at the receiver. Most multimedia applications perform some processing on the information before transmitting. This processing, which may include the collection and compression of media streams, introduces variable delays at the sender. At the receiver, jitter can occur in decompressing, combining, and in the playback times of the multimedia streams. Access to the network interface can also cause jitter. This is usually the case in local area networks where a host must contend with other hosts sharing the link for access to the media. However, upon successful transmission, the streams can still experience jitter in the intermediate nodes. When a busy node receives network traffic, the streams may further be delayed because of buffering. Packets may also become lost as buffer space for all the incoming traffic in the node may be insufficient.

Fragmentation at the sender and intermediate nodes have also been known to cause jitter. If packets presented at a node or network interface are larger than the maximum transmission unit, it is first fragmented before it is forwarded. When it finally reaches its destination, the packets may have been fragmented and reassembled several times along the way.

For data applications, jitter is just a side issue compared to loss and sequencing of data. However, in supporting real-time communication, the quality of a network connection is often evaluated in terms of its ability to deliver the source traffic to the receiver with as little jitter possible. Real-time multimedia communication is characterised by the exchange of isochronous traffic – traffic that is dispatched at fixed intervals. The temporal relationships associated with real-time application traffic must be preserved in order to be useful to the receiving host. Jitter disrupts these temporal relationships causing problems not only at the end receiver but at the intermediate nodes as well.

The effects of jitter can be reduced by reserving sufficient resources like buffer space, at the receiver and at the intermediate nodes. This allows jitter to be smoothed out at the expense of an additional delay. Along with buffering, a service discipline must also be in place at the intermediate nodes to discriminate between packets. In this scenario, when a network accepts a real-time stream, it assumes the responsibility of providing the resources to meet the flow’s jitter requirements and other specifications. In cases of high network resource demand and when the requirements of the flow are too stringent, the network rejects the request.

However, before any of these schemes are implemnted and used, the main problem should first be addressed. That is – the causes of delay jitter must be determined and its impact be quantitatively assessed. To answer this, we are developing a simulation model that considers and evaluates the effects of different network components and end-to end protocols to a real-time flow.

Firstly, I completed previous work on the Link-Level-Resource-Management Protocol (LLRMP) prototype according to the Internet-Draft submitted in December 1996 (draft-kim-llrmp-01.ps). This prototype was implemented in the kernel of a LAN switch to: a) prove the basic LLRMP protocol concept, and b) to estimate the amount of protocol status information required in a LAN switch for supporting a reservation setup with active admission control. The latter was an issue because existing LAN switches forward data packets in hardware and often do not have a sophisticated processor (sometimes even one without floating point support) and much memory to run a complex signalling protocol. The concept and some results are described in an OPENSIG paper, which however was already published in October 1996.

Secondly, I continued to work on the admission control conditions required for supporting deterministic service guarantees in shared 802.12 networks. I now analyzed cascaded (multi-hub) topologies and computed the relevant network parameters. When used with these results, the admission control conditions previously derived for the single-hub network also apply to cascaded topologies. Theoretical results received for network parameters and the quality of service e.g. the end-to-end delay were experimentally tested using the LLRMP implementation with integrated admission control in an 802.12 test network. The results will be published [168] [169].

I also began looking at the admission control conditions required in shared and half-duplex switched 802.12 networks for supporting a Controlled-Load type service. This started with an analysis of the available bandwidth and of the high priority access time on half-duplex links. Measurements were taken to confirm the results for both parameters. I further defined parameter based admission control conditions consisting of a bandwidth- and a buffer space test which allow the allocation of average rates. Time was also spent to look at traffic-models –sampling and measurement issues. There is however nothing written up yet. Future work includes the verification of service parameters (packet-loss and average-delay) in single- and multi-hop switched 802.12 networks. This is of particular importance since, following the ISSLL model, our test LAN switches can not isolate single Controlled Load flows, as always assumed for routers in the WAN.

Genetic Programming (GP) is a relatively new technique that uses the principle of Darwinian evolution to automatically program computers. GP has many advantages over other techniques that have been used in an attempt to automatically program agents such as neural networks and conventional genetic algorithms. The most important is that the representation used by GP is a computer program. It is therefore possible for humans to understand generated solutions and therefore predict their behaviour (compare with neural networks). Furthermore, the generated programs can be easily executed on current computer architectures and do not require any specialised environments. A second advantage is that GP has been shown to be very general and has been successfully applied to a wide variety of problems. Further details of GP can found in our GP survey.

Our goal is to show that GP can be used to evolve a society of agents that communicate and interact to solve a given problem. By this approach we hope to evolve programs which simultaneously define:-

1. each agents’ behaviour;

1. why, what and when agents communicate;

1. the "language"' that they use for communicating;

1. how sentences in the language are formulated;

1. how sentences in the language are interpreted;

1. how the agents interact and coordinate with each other to provide a solutionto the overall the problem (i.e. formulate a distributed algorithm).

Results demonstrating a subset of the above goals, in an experiment called the two agent problem was presented in the GP’96 conference and is available in the proceedings. The results demonstrated evolved agents communicating to solve a very simple global problem.

Due to the lack of scalability of existing GP systems, we developed a next generation GP system called GPsys. GPsys is a scalable GP system written in the language Java. It is portable, efficient, easy to use, strongly typed, and supports ADFs and demes. The latter is a powerful feature allowing GP runs to distributed across a network of computers (any computer capable of running Java) and allows very difficult problems to be solved. We have made GPsys, complete with source code and detailed documentation publicly available via the WWW and FTP. So far, there have been over 700 downloads of our software.