CASE 0 ------ Title: Student: Anna Bouch, supervised by Angela Sasse & Jon Crowcroft Where: UCL-CS ------------- Statement of Proposed Work: --------------------------- A key problem with the current standards for ATM and IP based networks is the absence of mechanisms for charging for network services. This project will investigate users' perception of, and attitude to, charges for network services which support a range of applications across a range of circumstances. The results of this work will help to predict what charges and billing mechanisms will be acceptable to certain user groups for certain tasks; this provides the basis for implementing user interfaces which communicate service options and associated charges for network services to users, and allow them to select services accordingly. The work will be closely linked (via CASE 1) to the charging mechanisms developed for ATM and IP networks at CUCL. There is a rapidly growing literature on electronic payment and economic models for Internet services. This builds on a very substantial literature on Quality of Service (QoS) requirements from a technical point of view, and mechanisms which may be implemented within networks to ensure that these requirements are met. There is, however, considerable uncertainty as to how these technical QoS requirements relate to end-users requirements, and which charging and billing mechanisms are likely to be acceptable to end-users. User acceptance of charging mechanisms is closely linked to the perception of the speed and quality at which media are delivered. The level of quality required for different media - and the level at which increasing quality yields diminishing return - depend on the type of interaction and real-world task which the user is trying to perform. There is currently no straightforward way of predicting perceived quality from objective measures (delay, packet loss, video framerate, etc.). Ongoing work on user perception of audio and video quality in multimedia conferencing at UCL (by BT CASE student Anna Watson) has shown (a) that perceived quality can be measured through continuous rating scales for quality and effort, and (b) provided a basic framework for describing tasks. The proposed project will build on this work and investigate the relationship between task, perceived quality, and acceptable charge. We will start by adding questions about acceptable charges to the video quality assessments in videoconferences and interactions with videoservers, to be performed as part of the BT/JISC HIGHVIEW project (with Essex Univ., starting October 1997). Other applications to be investigated are high-quality audio (provided by the BT/JISC JAVIC project), audio/shared workspace centric collaboration (e.g for remote design collaboration, WWW, Web-TV, Internet "chat lines", distributed games and distributed virtual reality applications (virtual travel agency, virtual business meetings - provided by the ACTS COVEN project). Data to be collected and analysed will include - system logs of quality-related user actions (e.g. adjusting framerate, audio quality), - semi-structured interviews about applications and charges - web-based market research questionnaires (UCL is currently experimenting with combined sliders for playing out audio and video at different quality, and indicating respective charges) - focus groups (both face-to-face and over the net). We will use grounded theory (a social science approach which we have adapted and applied successfully (e.g. on complex issues such as user perceptions of passwords and privacy) to combine the findings based on quantitative and qualitative data into a model which will allow us to predict which charges users will find acceptable for certain tasks and quality. The model (and associated guidelines) provide the basis for deciding what QoS information should be relayed to the user, in which form. We will implement user interfaces (or add to end-user application interface if possible) which communicate to end-users information about - available QoS options for the media employed by the application; and - the charges associated with those options, and evaluate those through experiments and field trials with end-users of a range of applications (see above). CASE 1 ------ Title: Network Support for Internet Charging and Quality of Service ------ Student: Richard Mortier -------- Where: CUCL ------ Statement of Proposed Work: ---------------------------- A key problem with the current standards for ATM and IP based networks is the absence of mechanisms for charging for network services. This project will engineer network controls in LEARNET which will enable charging for ATM and IP services to be implemented. The charging regime will reflect the network price of transmission of user data, and will be closely tied (via CASE 0) to the actual billing mechanism provided to users. In ATM networks it is possible for users to request quality of service (QoS) for their traffic according to the needs of the service they support, and given an appropriate traffic profile the network can ensure that this is delivered, however to date few realistic charging mechanisms for multiservice traffic have been proposed. Recent work in the ACTS CA$HMAN Project (Kelly, 1997), has developed specific mechanisms which attempt to price resources in the network according to the effects of the traffic on the network and the cost to the network of transporting the user traffic. Other groups are also developing charging mechanisms, and their work addresses security and resource allocation issues. The ESPRIT MEASURE project at Cambridge proposes the use of measurement based resource allocation and charging for ATM networks, and has developed several practical algorithms of use in this area. In IP based Internets/Intranets the charging problem is even more severe: there is an absence of deployed mechanisms for delivery of QoS and the security and ability of the network to signal congestion and pricing information to users is severely lacking. The Internet as currently engineered and provided to users is unable to provide anything other than best-effort service, and consequently there are few incentives for users to place realistic values on the traffic they transmit and receive. The result is a heavily congested network with poor QoS, upon which business critical applications cannot rely. There is a rapidly growing literature on electronic payment and economic models for Internet services. This builds on a very substantial literature on QoS and mechanisms which may be implemented within networks to ensure that traffic flows receive the delay and loss bounds which they require. As an example, the RSVP protocol has been proposed by the IETF as a signalling mechanism which can be used to reserve resources within an IP network according to the ability of the user's end system to deliver the required QoS and the user's preparedness to pay for the resources which they use. Several problems arise with the use of RSVP, however, including its inability to signal charging information, its inappropriate nature for inter-operator signalling, and the mapping of RSVP onto ATM. The LEARNET project will build an ATM network which is fundamentally different from standards based ATM networks in that it will support open control and the provision of dynamically sized networks running various control architectures, on demand. The enormous flexibility which this offers will allow us to engineer into the control architectures which support both standard ATM signalling and IP routing, mechanisms to enable provision of QoS to flows/connections with different QoS requirements, and signal information which can be used to charge for QoS sensitive services. This research will build on recent theoretical work by Kelly (Statistical Laboratory, Cambridge) and Key (BT Laboratories) on congestion pricing and resource allocation in networks, and in collaboration with partners at UCL, to implement specific control signals within LEARNET which can be used to control user traffic according to a pricing regime. The protocols will be able to modify resource allocations within the IP and ATM virtual networks running on an ATM switch, to ensure that users receive the QoS for which they are charged. In response to network congestion and measured user activity appropriate congestion signals which relate to network costs will be transmitted back to charging regime, and thence to the users. The project will interact closely with work at UCL which will implement the user-end of the congestion based charging mechanism. That work will build a demonstrator which uses the network charging signals to modify QoS delivered to a user in response to user feedback and payment. ---------------------------------------------------------------------- CASE 2 ------ Title: Connection Closures for Network Performance Management ------ Student: TBF -------- Where: CUCL ------ Statement of Proposed Work: ---------------------------- Software components are essential constituents of telecommunications systems ranging from: embedded, tightly engineered systems for routing traffic through switches and establishing connections using signalling protocols; through management components for configuring networks, measuring performance and managing faults; up to service operation and management systems for controlling advanced service sessions and maintaining personalised customer profiles. A range of architectures and technologies have been proposed and (sometimes) deployed for the various telecommunications software systems including IN, TMN, ODP, TINA, SS#7, INAP, Open signalling, SNMP, PNNI, CMIP, CORBA. The TINA initiative was one of the first major initiatives to consider a unifying framework for telecommunications software based on distributed object oriented systems using a ubiquitous distributed processing environment. Developing management systems for large networks containing equipment from multiple vendors, and possibly delivering services with different performance requirements, is an extremely complex task. To begin with, typical standardised MIBs in switches and routers lack sufficient detail to be unambiguously understood/implemented. Vendors implement vendor-specific additions to switch MIBs in order to differentiate themselves from their competitors. It is thus often impossible to ensure that a large network presents a common interface to network management tools. For the network operator, this can be a huge problem, not only because of the additional complexity this introduces into the network management, but also because it may be impossible for the operator to extract the appropriate information for performance management or other management functions from all of their equipment in a standard way. With the advent of agent technologies and the opportunities provided by mechanisms for mobile code, it may be possible to devise an advanced architecture for all telecommunications software based on a very lightweight framework, rather than prescribing a rigid structure as is the case in current signalling and embedded control systems, IN, TMN and TINA. The LEARNet project will build an ATM network which is fundamentally different from standards based ATM networks in that it will support open control and the provision of dynamically sized networks running various control architectures, on demand. The technique uses a concept called "switchlets" which in effect partitions the resources of a single physical switch between multiple control architectures. It will also be possible to download into the network small executable programs which we call connection closures. A connection closure can provide some of the application/service specific control functions which new services will require. The connection closure concept was developed by Sean Rooney at the University of Cambridge Computer Laboratory. A possible scenario is one where network elements themselves are shipped with very little native functionality embedded at manufacture time, instead software components may be deployed more dynamically through the use of agent technologies and mobile code. Furthermore, the management plane could be similarly structured, with management host computers providing a lightweight environment to allow management software to be dynamically deployed according to the current policies of the operators and users of the networks and according to dynamically changing operating conditions. From this basic, lightweight environment for control, management and services it would be possible to impose some structure through the gradual addition of objects to the network elements and the management hosts, increasing the "resident" functionality at these locations. In this way this very general framework can be enhanced to mimic TMN operations systems or TINA computational objects. It may be possible to comply with all previous telecommunications software architectures and standards as well as providing the opportunities for less organised interactions through the use of agent technologies. Depending on the lifetime of particular agents this approach can either be considered as a software deployment mechanism with permanent, resident agents, or as a true, dynamic agent environment with relatively short-lived agents performing specific, focused tasks. The advantages of the architecture will be most obvious when a combination of "resident" and "visiting" agents with varying lifetimes cooperate. While this framework offers enormous flexibility careful consideration must be given to issues of access rights, prioritisation of conflicting actions, etc. The success of this work will depend very heavily on the ability of the framework to accommodate flexibly defined policies on security and priority issues according to the varied demands of different stakeholders of future telecommunications networks based on this technology and architecture. Task 1: Overall Framework Development ===================================== This task will design and prototype an advanced architecture for performing network and service management on LEARNet. This architecture will be based around the used of distributed and/or mobile agents (MA), consisting of both code and data, interacting through a thin layer Framework context. The functionality of both the management system/platform and its devices and environment is made available through collections of MA communicating through the framework. The MAs are not fixed, but are defined by their functionality, role within the system and behaviour (API+). Thus a particular function may be enhance or specialised for some particular user requirement or task by the appropriate selection of a (class) of entity. The base line for the Framework is thus to intermediate communications between these MAs (see 1.A). The Framework is, essentially, a local environment. It may exist either on a general purpose platform or in specialised devices (switches, routers, SSF, SCF, CCF). The functionality of the environment is made accessible through local fixed agents (see 1.B). Equally, system functionality such as resource allocation or communication between such frameworks - or MAs working within several frameworks - may be performed by other specialised MAs (see 1.C). Some amount of information is common to all entities which is used to determine their functionality, security, priority and such like (see 1.D) 1) A study of existing distributed network and service management architectures should be made in order to: A) Determine basic and enhanced functional requirement for the Framework environment. B) Determine classes of local service elements C) Determine classes of Framework or platform elements. D) Determine the core identity attributes of the MAs Task 2: Adaptation to Switchlets ================================ This task will build on the connection closure concept to apply it to performance management of broadband networks. By engineering into the switch control systems the ability to dynamically load a control regime and suitable performance management constraints, this work will allow multiple virtual networks with different performance constraints and management requirements to be simultaneously operational on a single ATM switch infrastructure from multiple vendors. It will facilitate the performance management of multiple network services with different performance constraints by enabling the management system to download mobile code to each switchlet-based virtual network. The "management closure" which this mobile code represents, will encode the performance management requirements of the virtual network and will be interpreted by the switch controller or the switch itself. Further, by having the facility to download code onto each switch controller, it will be possible to ensure that each physical swit Task 3: Application to Performance Management ======================================== This architecture is to be tested in the context of the LEARNet by focusing on the issues of performance management; and possibly configuration and fault management. Components and frameworks need to be designed to construct the network and service management platform, to interface into the switchlets and to interface at the user/provider ends. In this context the system extends the architectures of TINA & TMN (2.D). Equally, end-user & service elements need to be defined to provide end to end application support in a complex distributed environment (e.g. to support the QUILT project). In this second context the IN & WWW environments are extended .. or brought together (2.E). 2) Design Architecture with; A) Inter-entity communications semantics B) Baseline Framework functionality C) Baseline set of platform specific MAs D) Elements for (TINA) like N&SM. E) Elements for (IN/WWW) like user services