Distributed Systems and Networking discipline Seminar Series 2005

Click here for Seminar series in 2004

Overview

DSN seminars run fortnightly on Fridays during the semester from 3:00-5:00pm.

These AU seminars are run in an informal way, and therfore they different from the "conventional ones" (e.g. our school seminars). These seminars are there to help our students (i.e. PhD, Master and Honours) in

It is also an opportunity for our research staff to present some of the research work they are involved in. Guests from outside of the AU will be invited regularly to give talks at these seminars to expose our staff and students to different areas of research conducted by groups from other organisations.

Schedule

Date Location Speaker 1 Speaker 2 Speaker 3 Speaker 4
11/02/2005 14.10.06a Vincent Chau Fabio Zambetta      
25/02/2005 14.10.06a Peter Dimopolous Panu Phinjaroenphan      
18/03/2005 14.10.06a Sandy Citro Damien Phillips   Gregory Craske    
25/03/2005 14.10.06a Good Friday - Holiday      
08/04/2005 14.10.06a Guest Speakers - Andras Varga, Steve Woon, Eric Wu and Dr Ahmet Sekercioglu, Monash University      
22/04/2005 14.10.06a Jens Bin Rong      
13/05/2005 14.10.06a Guest Speaker - Dr. Dinesh Kant Kumar, School of Electrical and Computer Engineering, RMIT University      
20/05/2005 14.10.06a Guest Speakers - Richard Smith and Paul Robinson, Sun Microsystems      
10/06/2005 14.10.06a Mikhail Perepletchikov James Broberg      
24/06/2005 14.10.06a Guest Speakers - Laurence Plant, Andrew Brockfield and James Kelly from IBM      
08/07/2005 14.10.06a Pablo Rossi Sakib Kazi      
22/07/2005 14.10.06a Nalaka Gooneratne alice wang      
05/08/2005 14.10.06a Guest Speaker - Henry Wu, School of Electrical and Computer Engineering, RMIT University      
16/09/2005 14.10.06a Sandy Citro Vinh Dao   Gregory Craske    
14/10/2005 14.10.06a Bin Rong Peter Dimopolous      
28/10/2005 14.10.06a Nalaka Gooneratne James Broberg      
11/11/2005 14.10.06a No seminar due to school seminar by our au member      
25/11/2005 14.10.06a No Seminar due to Research student conference      
09/12/2005 14.10.06a Guest speaker - Professor Bill Appelbe, VPAC Melbourne      
# confirmed * guest speaker to be confirmed


11/02/2005

Speaker : Vincent Chau

Title : Distributed Rendering Architectures for Interactive Tiled Display Applications

Slides : none

Abstract: We briefly introduce tiled displays and their motivations, as well as the 3D graphics rendering process. We then explore in detail, the two approaches currently used for the distribution of rendering information across a tiled display cluster. Our discussion will focus on the performance issues including network traffic and ease of implementation. We also introduce the problem of load balancing and potential solutions.

11/02/2005

Speaker : Fabio Zambetta

Title : Introduction

Slides : none

Abstract: Dr.Fabio our new graphics staff will give a short introduction about himself and his work.

25/02/2005

Speaker : Peter Dimopolous

Title : Reducing the delay of interactive TCP packets in a QoS network.

Slides : none

Abstract: Applications like ssh and small web applications are considered to be interactive TCP connections. If these types of applications suffer from packet loss, the packet retransmission time will increase the user perceived delay more than the queueing delay. This work introduces a Dynamic Priority RED Queue (DPRQ) algorithm which reduces the user perceived delay by reducing packet loss in interactive TCP connections.

25/02/2005

Speaker : Panu Phinjaroenphan

Title : An Algorithm to Predict the Reliability of a Grid Node

Slides : none

Abstract: In this talk, I will present an algorithm to predict the reliability of a node in a computational grid. For more information, please come to the seminar.

18/03/2005

Speaker : Sandy Citro

Title : Evaluating and Extending Consistency Management Algorithm for use in Mobile Real Time Collaboration

Slides : none

Abstract: Mobile real time collaborations allow two or more co-workers to work on a shared document at the same time on separate mobile devices. The replicated architecture is attractive for this application since it does not need a central server to operate and if a device is disconnected, it can continue to work on its local replica while the others continues the collaboration. Thus, a consistency management algorithm is important to ensure every local replica is consistent with each other. Several algorithms have been proposed for this purpose, however they do not take resource usage into consideration in their algorithm design. They assume the available resources are unlimited while this is not true in mobile environment. Mobile environments have very limited bandwidth and mobile devices are limited in memory and processing power. Therefore, there is a need to design an algorithm that uses minimum resources to operate. Some existing techniques is proposed to be used in the algorithm design as well as a novel technique called Partial History Copy. The combinations of those techniques are then tested using simulation to determine which algorithm design combination is the most suitable for in mobile network environments.

18/03/2005

Speaker : Damien Phillips

Title : Fair Highspeed TCP and larger packet sizes

Slides : none

Abstract: Fair Highspeed TCP is a proposal to introduce RTT fairness into the existing Highspeed TCP proposal. Its aim is to bring competing connections' increase rates dependant on perceived throughput rather than the general packets-per-RTT increase inherited from standard TCP. However, under some conditions this new proposal can generate increases of an unduely large number of packets per RTT. In fact, it has been proposed that slow start limit its increase rate to 100 packets per RTT. This is to prevent a large number of packets being dropped when the bottle neck link and (droptail) queue reaches saturation. We aim to investigate whether larger packet sizes with equivalent queue memory can alleviate TCP's problems from recovering from a loss of 10s to 100s of packets.

18/03/2005

Speaker : Gregory John Craske

Title : Incorporating negotiation into service communication protocols.

Slides : none

Abstract: This talk is a position statement and discussion of preliminary research into dynamic communication protocols in SOA. Service infrastructures, like Web services, rely on rigidly defined functional specifications for services to allow integration.Selection of services includes the ability to satisfy functionalconstraints, which is limiting to services that are able to be flexible in its interactions. The aim is to allow a degree of flexibility in how services can interact, that is, their communication protocols. Such flexibility can give a service some market advantage by allowing a more personalised client interaction.

08/04/2005

Speaker : Andras Varga, Steve Woon, Eric Wu and Dr Ahmet Sekercioglu, Monash University.

Title : Network simulation with OMNeT++

Slides : none

Abstract: The seminar will introduce OMNeT++ (www.omnetpp.org) as a network simulation tool. OMNeT++ is a cross-platfrom, public-source, component-based, modular and open-architecture simulation framework with a well-defined API and graphical animation and debugging capabilities, and support for exotic use cases like embedding, real-time, hardware-in-the-loop and parallel distributed simulation. OMNeT++ is primarily designed to simulate computer networks, multi-processors and other distributed systems. OMNeT++'s use is increasing in wireless, IPv4 and IPv6-related research in academia, and also in commercial environments for a wide spectrum of projects and products. The seminar will focus on the major network simulation suites for OMNeT++ (INET Framework, IPv6Suite, Mobility Framework), and include special topics such as queueing, parallel distributed simulation, real-time and hardware-in-the-loop simulation. Special attention will be paid to IPv6, Mobile IPv6 and wireless LAN (802.11) simulations, and these parts will be presented by the (co)authors of the respective simulation models in IPv6Suite.

22/04/2005

Speaker : Bin Rong

Title : An Adaptive Gossip-based Application Layer Multicasting Membership Management Algorithm

Slides : none

Abstract: Membership management is very important to the success of application layer multicast. In particular, a scalable and reliable membership management algorithm will enable the fast deployment of application layer multicast related services on the Internet. The overlay network built atop Internet is always under high dynamics, because the users can join and leave at will. How to capture these changes and communicate among the remaining users in a timely and efficient fashion, and how to balance the network overhead, computational complexity and the network performance become major concerns of the application layer multicast schemes. In this paper, a new adaptive gossip-based membership management algorithm is proposed. Its adaptive nature enables it can confine the control overhead to local ranges and, adapt to the ever-changing network traffic conditions and group membership at the same time. The random nature of the algorithm ensures it can cope with random failures and offer proactive measures to maintain service at a certain level. Mathematical analysis is used to prove the correctness of the algorithm. Simulation results indicate more than 90% of the nodes can work properly even under high network dynamics (with a short half-life time of 50 seconds). At the same time, a low overhead is used to achieve this desirability.

22/04/2005

Speaker : Jens Anderson

Title : Service Level Agreements and traffic shaping at application level

Slides : none

Abstract: When considering guarantees and QoS at application level in Web Service architectures, it is an important issue what kind of constraints that are included in the Service Level Agreement. Today it is common that the only shaping parameter included, is expressed as a max number of requests per second (max/s). This method means that it is possible to send requests in quite large bursts without violating the constraint. I^Òm investigating what we can gain by expressing the shaping constraints as peak rate variables instead. In the case of having hard time constraints it is no doubt about the advantageous with peak rates, but when considering soft time constraints the performance is totally dependent of the arrival process. The more burstiness in the arrivals, the more gain with the peak rate constraint. In some cases the max/s shaping constraint will even perform better than a peak rate constraint.

13/05/2005

Speaker : Dr. Dinesh Kant Kumar, School of Electrical and Computer Engineering, RMIT University

Title : TBA

Slides : none

Abstract: With the growth of IT, and sense of insecurity, there is an urgent need for identifying people, who they are, and what they represent. The difficulty with the current biomtrics trends is that these systems are based on the anatomical details of the people, and hence, cannot be hidden. There are number of conditions when these maybe copied- such as the finger print from a cup, etc. DNA is still slow and clumsy. Our team has come up with a novel idea of using human gestures for this application. This maybe changed by the authentic user, and requires extremely small memory space.

20/05/2005

Speaker : Richard Smith and Paul Robinson, Sun Microsystems

Title : TBA

Slides : none

Abstract: Richard Smith, a senior engineer will talk about what's new in Solaris 10, and a bit about clustering technologies with Solaris Paul Robinson will talk about ways of doing grants, etc with Sun.

10/06/2005

Speaker : Mikhail Perepletchikov

Title : A methodology for designing and implementing Service-Oriented Architectures

Slides : none

Abstract: Service-Oriented Architecture (SOA) is a promising approach for developing integrated enterprise applications. Although the architectural aspects of SOA have been investigated in research and industry literature, the actual process of designing and implementing services in SOA is not well understood. The main goal of this research is to develop a systematic methodology for the design and implementation of SOA-based systems including detailed processes, guidelines, and activities. This includes deriving and validating a set of software metrics for objective evaluation of such methodology. The secondary aim is to integrate the developed methodology into one of the mature software development process frameworks, namely Rational Unified Process (RUP) and OPEN Process Framework (OPF). This presentation will discuss the recently submitted PhD proposal in terms of rationale, research questions and research methodology, and will also briefly overview a paper submitted to the International Workshop on Services Engineering.

10/06/2005

Speaker : James Broberg

Title : A Distributed Algorithm for Multi-commodity streaming topologies.

Slides : none

Abstract: Consider a network of servers, each server has finite amount of computing resource (e.g. CPU), the physical links between servers have finite bandwidth. Each server contains multiple programs that can be executed in parallel by sharing the com- puting resource. Different programs may perform different functions, which may require different amount of computing resource, process different input streams, and produce different output streams for different downstream servers. Flow conservation may not hold in this type of stream processing - different programs may have shrinking or expansion effects on flow. Each server is then faced with two problems: first, it has to decide how much computing resource to spend on each program for each of its downstream nodes; second, it has to decide the bandwidth sharing among the multiple output flows. We aim to find the best allocation of available computing resources and bandwidth in such a scenario.

24/06/2005

Speaker : Laurence Plant, Andrew Brockfield and James Kelly, IBM

Title : The Grid and Games

Slides : none

Abstract: TBA

08/07/2005

Speaker : Pablo Rossi

Title : Software, Performance and Resource Utilisation Metrics for Context-Aware Mobile Applications

Slides : none

Abstract: As mobile applications become more pervasive, the need for assessing their quality, particularly in terms of efficiency (i.e. performance and resource utilisation), increases. Although there is a rich body of research and practice in developing metrics for traditional software, there has been little study on how these relate to mobile context-aware applications. Therefore, this paper defines and empirically evaluates metrics to capture software, resource utilisation and performance attributes, for the purpose of modelling their impact in context-aware mobile applications. To begin, a critical analysis of the problem domain identifies a number of specific software, resource utilisation and performance attributes. For each attribute, a concrete metric and technique of measurement is defined. A series of hypotheses are then proposed, and tested empirically using linear correlation analysis. The results support the hypotheses thus demonstrating the impact of software code attributes on the efficiency of mobile applications. As such, a more formal model in the form of mathematical equations is proposed in order to facilitate runtime decisions regarding the efficient placement of mobile objects in a context-aware mobile application framework. Finally, a preliminary empirical evaluation of the model is carried out using a typical application and an existing mobile application framework.

08/07/2005

Speaker : Sakib Kazi

Title : Energy Balancing in Self-configuring Sensor Networks

Slides : none

Abstract: Like other distributed systems, unattended sensor networks need to be balanced not only by load but also in terms of energy. Sensor networks are expected to live long. Lifetime can be maximized if and only if a balanced network can be formed. In this research work we first identify the tasks, that a sensor node accomplishes in its life time. Based on those, we define the sensor node life-cycle. Then we characterize the nodes based on their residual energy levels as, SEN - nodes having sufficient energy to perform additional responsibilities other than its own sensing task, or NEN - nodes having only necessary energy to perform its own task. Every node starts as SEN, and at some point it becomes NEN based on a predefined threshold value. Thus energy scared nodes are protected from the early exhaustion. Finally, we developed a virtual clustering technique which is self-configurable and scalable. We apply our approach to a simulation environment and the simulation results justify our assertion.

22/07/2005

Speaker : Alice wang

Title : An Enhanced Gradient-Based Algorithm for Estimation of Fingerprint Orientation Fields

Slides : none

Abstract: The estimation of fingerprint orientation fields is an essential model in fingerprint recognition systems. Conventional gradient based approaches are popular but very sensitive to noise. In this talk, we are going to present a novel implementation that can improve the performance of the gradient-based methods. The enhanced algorithm estimates the dominant orientation field of each image block from its four overlapping neighborhoods. A voting scheme for the best estimate is used based on the associated reliability measures. Our experiment results suggest that the enhanced algorithm is more robust against noise and the required computation time is modest in comparison with other gradient-based methods.

22/07/2005

Speaker : Nalaka Gooneratne

Title : Automatic discovery and validation of composite web services

Slides : none

Abstract: The vision of the Semantic Web is to convert the World Wide Web to a universe of computer based services that automatically coordinate and collaborate to satisfy user requests. For this vision to be realized web service descriptions have to be automatically interpreted and techniques to discover, compose, verify, invoke, and monitor web services have to be defined. In the proposed project we will define a web service description framework, a composite matching technique, and a verification technique. The service description framework will represent the functional semantics of a web service using constraints, parameters, and communication protocols. Existing service description and verification techniques are only able to represent and validate communicative actions that take place at time points. The proposed framework and verification technique will enable the representation and validation of communicative actions that take place at time points and during time intervals. Existing composite matching techniques, which are used in web service discovery, are only able to consider constraints that apply to individual web services. The proposed matching technique will extend on the above notion by considering constraints that apply to multiple web services of a composition.

05/08/2005

Speaker : Henry Wu

Title : Visual Information Processing and Communications

Slides : none

Abstract: This talk will present an over view of recent research work by the speaker and his co-workers in areas of digital video image processing and enhancement, digital picture quality assessment and metrics, and vision model based picture coding and compression. The video processing and enhancement work is based on a number of his journal and conference publications and has been developed into a software package. Vision model based picture quality metrics have been extended to design of perceptual coders for images and video sequences. In particular, the perceptual lossless coding of medical images is based on recent research work jointly conducted with Faculty of Medicine at Monash University and Department of Diagnostic Imaging, Southern Health Monash Medical Centre. It discusses a vision model based digital image coder which produces compressed bit stream compliant with the state-of-the-art JPEG 2000 International Standard. The subjective evaluation conducted at Southern Health shows that the perceptually lossless coder achieves significant compression gain over the information lossless coder (such as JPEG Lossless coder) with visual quality which is indistinguishable from that of the originals. The medical images used in evaluation were selected to include different modalities such as CR, CT and MRI. The perceptually lossless coder has been applied to ultrasound images as well. A number of interesting topics in image processing and compression for medical imaging, archive and communication applications will also be highlighted in the talk for possible future collaborative projects.

16/09/2005

Speaker : Vinh Dao

Title : IDENTIFYING REGIONS OF VISUAL ACTIVITY IN A VIDEO REQUIRING DETAILED CAPTIONING

Slides : none

Abstract: This research introduces a metric that measures the amount of visual caption congestion contained in a given video. The metric^\uffffs method of measurement is primarily based on the temporal Gray Level Cooccurrence Matrix (GLCM), which compares a reference video frame to a fixed number of neighboring video frames. A correct combination of features obtained from this GLCM optimally determines the temporal amount of visual activity in the video. This amount of visual activity is then compared to the temporally coinciding captions in the video, resulting in the identification of regions containing potential amount of visual caption congestions present in the video.

16/09/2005

Speaker : Sandy Citro

Title : A Membership Management Algorithm for Real Time Mobile Collaboration

Slides : none

Abstract: Real time mobile collaboration systems allow two or more users to work on a shared document at the same time regardless of their geographical position. Users can participate in the collaboration as long as they are connected to each other. Mobile networks are characterised by frequent disconnection and ad hoc network creation. Sites might get disconnected from each other from time to time. A site may decide to join a currently running session and demanding to be able to participate in the session. A proper membership management is needed to ensure all the sites are able to continuously participate in the collaboration and the collaboration session has to be able to continue transparently and smoothly regardless of those events. This paper aims to introduce a membership management algorithm that is able to handle the membership events in the real time mobile collaboration system. The proposed algorithm works in a replicated architecture which means that it does not rely on any dedicated server or any leader, thus avoiding the single point of failure. The proposed algorithm also incorporate a novel technique, called State Map, to handle the event where multiple sites join the session at the same time. We built the algorithm on top of an existing consistency management algorithm so the proposed algorithm is able to manage membership events while maintaining the document consistency.

16/09/2005

Speaker : Gregory Craske

Title : Fine grained execution of business processes.

Slides : none

Abstract: Business process execution is traditionally handled by a centralised coordinator (execution engine). There are well-known performance and availability issues with such centralised approaches. One approach is the parallel execution approach, which distributes the coordination amongst statically allocated engines. However, the parallel execution model is not distributed to the finest grain and so there is room for improvement on availability and performance. This talk proposes a new distributed execution model based on mobile `flows', which are autonomous, self-propagating units of coordination that communicate on a peer-to-peer basis. A distributed monitoring scheme and associated failure recovery techniques are also proposed for this execution model. Early results, which are promising, will be shown to demonstrate the effectiveness of this model.

14/10/2005

Speaker : Bin Rong

Title : Enhance reliability for application layer multicast

Slides : none

Abstract: Reliability is very important for multicast, espeically for application layer multicast. Our algorithm explores the power law distribution of the lifetime of the participating nodes to enhance reliability for application layer multicast.

14/10/2005

Speaker : Peter Dimopolous

Title : Reducing the User Perceived Delay of Interactive TCP Connections Using a Dynamic Priority Approach

Slides : none

Abstract: Many interactive applications require continuous user interaction, for example ssh and many web applications. The TCP connections created by these applications are therefore in a class called interactive. When interactive applications suffer from packet loss the packet retransmission time severely increases the user perceived delay. This presentation introduces a Dynamic Priority RED Queue (DPRQ) algorithm that dynamically changes the priority of queues instead of dropping packets when the queue is overloaded. The algorithm reduces the user perceived delay by reducing packet loss in interactive TCP connections. The DPRQ is compared to an existing Class-Based Queue which incorporates RED (RCBQ) as would typically be used for Assured Forwarding. An analytical model of the DPRQ and RCBQ is presented with both experimental and analytical results. The DPRQ is found by simulation to decrease packet loss by up to eight times and therefore provide a lower user perceived delay even though queuing delay is increased by up to five times.

28/10/2005

Speaker : Nalaka Gooneratne

Title : Verifying Functional and Behavioural Descriptions of Web Services

Slides : none

Abstract: The vision of the semantic web is to create an environment in which web services are automatically discovered, composed, executed, and monitored. The techniques that are used to perform automatic service discovery and composition require the functional and behavioural aspects of web services to be described. Currently, the functional and behavioural descriptions are created using unrelated frameworks and formalisms. Hence, the functional description and the behavioural description created for a web service can be inconsistent. We propose two formalisms, WS-ALUE and WS-pi-calculus for describing the functional and behavioural aspects of web services. WS-ALUE extends the Description Logic based language ALUE so that the components that are required to represent the functional aspects of web services can be described. WS-pi-calculus describes the behavioural aspects of a web service by modeling its interaction protocols. It extends pi-calculus by enabling the interaction protocols and their effects to be accurately modeled. The interaction protocols are accurately modeled by supporting the representation of all four types of communicative actions; send, receive, invoke and service. Finally we propose a technique for verifying the consistency between the WS-ALUE and WS-pi-calculus descriptions.

28/10/2005

Speaker : James Broberg

Title : Adaptive Task Assignment

Slides : none

Abstract: There are significant problems faced when analysing static M/G/c queuing models. As the problems get larger they can become intractable. This makes creating adaptive models nearly impossible under M/G/c assumptions. We are currently examining existing ways to numerically simplify our existing task assignment policy for M/G/c queues, TAPTF. We are first looking to fit a series of exponential distributions to our General (Pareto) distribution, in order to make the problem more tractable. This will enable us to do two key things: 1. Solve larger models (we can only optimally solve 3 host models currently) 2. Make an adaptive TAPTF model possible, that recomputes its parameters based on incoming traffic.