A. Gill Waters

Using UML to specify QoS constraints in ODP

This paper is concerned with Quality of Service (QoS) specification in distributed system design. The specification and implementation of QoS is increasingly important in distributed systems due to the need to address questions of performance, particularly for systems involving multimedia. To ensure correct implementation of QoS requirements, statements of QoS need to be introduced early in the design process, and in terms of design we consider the use of the Unified Modelling Language (UML), which has quickly become the de facto standard for object-based designs.The framework we use for distributed system construction is that provided by the open distributed processing reference model, and we focus in particular on its computational viewpoint. The aim of this paper is to construct a UML model of the computational viewpoint focusing on the description of QoS within that viewpoint. To specify the QoS aspects of computational objects in the UML model, we use a real-time logic called QL. In order to express further constraints on the UML model of the computational viewpoint, we use the Object Constraint Language to express invariants that each instance of our model must satisfy. The purpose of our UML model of the computational viewpoint is to act as a template via which specific distributed system designs can be constructed and we illustrate this with the specification of a lip synchronisation mechanism.

A performance comparison of self-organising application layer multicast overlay construction techniques

Application layer multicast (ALM) uses overlays built on top of existing network infrastructure for rapid deployment of multicast applications. Key to the efficiency of this technique is the structure of the overlay tree used. This work reviews and compares various self-organising techniques that strive to build low cost, and low delay trees using extensive simulations. Protocols investigated include HMTP, HostCast, switch-trees, DCMALTP, NICE, TBCP and Narada which encompass a wide spectrum of overlay construction, optimisation and maintenance techniques. The protocols are evaluated based on their ability to achieve their objectives, overlay path penalties, protocol convergence and overhead. We also conduct detailed analysis of two main components in building an overlay: initial construction and the overhead of periodical improvement. Based on the observed results, we identify strengths and weaknesses of various approaches, and provide suggestions for future work on ALM overlay optimisation.

A multiple shared trees approach for application layer multicasting

The data delivery mechanisms used for the many-to-many multicast applications have been based on two extreme approaches: a single shared tree and source rooted trees. These two approaches complement each other in terms of performance, protocol overhead and robustness. In this paper, we propose and evaluate a multiple shared trees protocol for application layer multicasting. The protocol constructs an application layer overlay with m data delivery trees, where m>1 and m /spl Lt/ the number of data sources. Our simulation results reveal that the multiple shared trees design is a promising application layer overlay construction technique.

Permabase: predicting the performance of distributed systems at the design stage

The paper describes the Permabase project, funded by BT, which produced prototypes to predict software performance automatically at the systems design stage. It discusses the Permabase rationale and describes the architecture and details of the prototype systems and its validation using case studies. It discusses the use of UML as a mechanism for capturing the information needed for performance prediction modelling, and shows how translation makes it possible to produce simulation models. It reviews the achievements of the project and looks forward to ways in which the work could be enhanced and built on to encompass a greater range of systems.

A UML Approach to the Design of Open Distributed Systems

The design of distributed systems is a highly complicated and non-trivial task. Introduction of multiple types of media into distributed systems causes a dramatic increase in the complexity of design. To deal with the inherent complexity of systems, two approaches have received considerable attention; ODP and UML. Open Distributed Processing (ODP) is a joint ITU/ISO standardisation frameworkfor constructing distributed systems. Unified Modelling Language (UML) is a de facto standard for visualising, specifying, designing, and documenting object-oriented systems.This paper presents a case study using a UML approach for the design and specification of distributed systems based on ODP. The purpose of the case study is to try this approach on a large system containing multiple types of media. The case study is carried out on an Interactive Multimedia Kiosk (IMK) example. IMKs integrate different types of media such as text, graphics, audio, video, animation and sound in the form of a large system; this provides an ideal subject for case study.

Design and Verification of Distributed Multi-media Systems

Performance analysis of computing systems, in particular distributed computing systems, is a complex process. Analysing the complex flows and interactions between a set of distributed processing nodes is a non-trivial task. The problem is exacerbated by the addition of continuous system functions that are time dependent, such as communication between components in the form of multimedia streams of video and audio data. Quality-of-Service (QoS) specifications define constraints on such communications and describe the required patterns of data transfer. By making use of these specifications as part of the performance analysis process it is possible to add significant confidence to predictions about the correct (required) operation of a distributed system. This paper presents a method for designing distributed multimedia systems, including the specification of QoS, using the ODP framework and UML and describes a technique for verifying the QoS specification against the designed functional behaviour of the system using Timed Automata.

MeshTree: reliable low delay degree-bounded multicast overlays

We study decentralised low delay degree-constrained overlay multicast tree construction for single source real-time applications. This optimisation problem is NP-hard even if computed centrally. We identify two problems in traditional distributed solutions, namely the greedy problem and delay-cost trade-off. By offering solutions to these problems, we propose a new self-organising distributed tree building protocol called MeshTree. The main idea is to embed the delivery tree in a degree-bounded mesh containing many low cost links. Our simulation results show that MeshTree is comparable to the centralised compact tree algorithm, and always outperforms existing distributed solutions in delay optimisation. In addition, it generally yields trees with lower cost and traffic redundancy.

Low-Cost ATM Multicast Routing with Constrained Delays

An increasing number of networking applications involve multiple participants and are therefore best supported by multicasting. Where a multicast application consumes high bandwidth, it is important to minimise the effect on the network by offering economical multicast routes. Many new applications made possible by networks based on ATM involve real-time components and are therefore also delay-sensitive. This paper discusses reasonably simple techniques for multicast routing which tackle both of these constraints, that is: first, the route makes efficient use of network resources and, secondly, delays to all recipients are kept within a bound. The problem is NP-complete, so we present heuristics which build up a directed graph containing potential routing solutions and use a greedy approach to select a solution from that graph. The heuristics are discussed and evaluated and are shown to offer good results for a variety of situations including both large and small multicast groups. Our approach is also compared with a previous solution to this problem, which has a greater time complexity.

ATM Multicast Routing

Several multicast routing heuristics have been proposed to support multimedia services, both interactive and distribution, in high speed networks such as B-ISDN/ATM. Since such services may have large numbers of members and have real-time constraints, the objective of the heuristics is to minimise the multicast tree cost while maintaining a bound on delay. They should also be fast to compute and may need to be suitable for dynamic groups.