|
|
||||
|
|||||
|
|||||
![The University of Melbourne [logo]](/template_images/logos/logo_crest_comb.gif){kind=logo}
|
Research ProjectsThe CUBIN Core Research Program comprises five inter-linked projects:
Below are brief outlines of these projects and their sub-projects follow, including lists of the staff, students and collaborators involved. Signal Processing
The Signal Processing project develops sophisticated signal processing algorithms for wire-line and wireless broadband networks. Recent activities have been focused on wireless networks such as Code Division Multiple Access (CDMA) based ad hoc networks and promising wireless communication technologies such as Orthogonal Frequency Division Multiplexing (OFDM). The traditional layered communication network structure can be modified by cross-layer optimization to make wireless networks more useful. For example, the physical layer in a wireless communication system (namely, the wireless channel) has a critical impact on the design of various higher layer algorithms and protocols. Existing signal processing algorithms for CDMA based ad hoc networks take such cross-layer issues into account. Similarly, OFDM systems employing multi-carrier modulation must implement efficient and accurate synchronization techniques, intelligent power allocation algorithms and state of the art optimization methods. The broad aims of the Signal Processing project in 2004 have been to:
Cross-Layer Design in Wireless Ad-Hoc NetworksAd hoc wireless networks lack centralized coordinated control such as exists in cellular mobile telephony networks where dedicated network infrastructure, e.g. a base-station, controls and co-ordinates nodes. Examples of ad-hoc networks include those based on the popular Bluetooth® protocol, where devices such as phones, PDAs, computers and other peripherals connect together in order to exchange information wirelessly. These devices negotiate the network structure amongst themselves in a peer-to-peer fashion. Ad hoc networks are often deployed in environments where power is limited and where the network configuration can change very quickly due to mobility or node failures, etc. In addition to the usual difficulties imposed by wireless channels such as random channel variations and interference from other users, ad hoc networks pose additional challenges for optimal resource allocation. Recent work appearing in the literature has shown that in a wireless multi-hop network, significant end-to-end gains on throughput can be achieved if aspects of the physical layer are jointly optimized with aspects of higher layers, e.g. the transport layer where network congestion control is performed. The challenges in such ad-hoc networks are optimizing the layers in a distributed manner, where each wireless network node implements an algorithm whose actions contribute to the global good of the network. This is harder in a fast-fading wireless environment where circumstances change quickly. In 2004, we looked at how to jointly optimize the power allocation (power control) and data-rates (congestion control) when the wireless channels between nodes are subject to fast-fading.
We applied our previous work on outage-constrained power allocation to this new research problem of cross-layer design. Initial results suggest that an outage-constrained problem, both in fading-induced signal-to-interference ratio (SIR) outage and congestion induced outage, may be posed as a convex optimization problem with some approximations and simplifications. When modelled in this way we can solve the mathematical optimization problem using the rich literature available on duality theory. Ultimately we will create distributed algorithms that solve the global network-wide problem with limited message passing. Optimal Resource Allocation for OFDM SystemsOrthogonal Frequency Division Multiplexing (OFDM) is a promising technique for future high-bit-rate wireless communication systems. It splits the information to be transmitted amongst multiple radio channels and offers distinct advantages over traditional single-carrier systems in multi-path fading immunity and in bandwidth efficiency. A good example of an OFDM system is the current ADSL standard using Discrete Multi-tone (DMT) modulation. Optimally allocating resources such as data rates and transmission power to maintain guaranteed quality of service poses difficult challenges to designers of practical OFDM systems. In addition, OFDM systems do have drawbacks in terms of high peak-to-average-power ratio (PAPR) as well as high sensitivity to time and frequency synchronization errors. In 2004, we developed a framework to allocate resources (data rate and power) in wireless OFDM systems. This uses channel fading statistics measured over multiple symbol periods, rather than the instantaneous channel conditions that are difficult to track and respond to in real-time. Channel prediction has also been employed to overcome latency in estimating the channel and feeding it back to the transmitter. We are investigating a few models for channel prediction, including Gauss-Markov and Hidden Markov models (HMMs), and will test our framework with real channel data. Initial work has begun on resource allocation for multiple-input multiple-output (MIMO) wireless systems (see below) to deal with various power constraint scenarios. Caption: Block Diagram of an OFDM System In addition, we developed novel optimal peak-to-average-power ratio (PAPR) reduction techniques using tone reservation methods. We completed a project with Lund University and Ericsson which extends previous work with ADSL2 standards (ITU G.992.3). Our new technique maintains the computational efficiency of its predecessor and achieves very close to optimal PAPR reductions. Adaptive Coded Modulation and Power Control for Optimizing Spectral Efficiency in Wireless NetworksThe capacity of a communication channel is defined as the maximum transmission rate at which reliable data recovery can be guaranteed with an arbitrarily low probability of error. This concept can be applied to wireless channels in many ways. One way, known as the ergodic capacity defines the long term average capacity where the average is taken over the statistical distribution of the randomly varying channel. Mathematical expressions for the ergodic capacity of fading wireless channels have been derived. Achieving these capacities in practice is still a significant challenge, even in the single user case. In theory, capacity can be achieved by using a code with long block lengths spanning many fading realizations, however this introduces long delays. Another proposed method is known as adaptive coded modulation. Here shorter block lengths are used, but the rate of transmission, and hence the codes used, will have to be varied whenever the channel changes. To achieve the maximum capacity, the number of different codes must increase to infinity. This is of course impractical. Research is focusing on how close to capacity we can get when the number of codes is finite. In previous work by other authors, information theoretic upper bounds on the spectral efficiencies achievable using a finite number of codes have been obtained, though they did not consider any power adaptation. In this project we investigated the situation where we can adapt both the allocated power and the code rates. We used a partitioned channel inversion power allocation scheme, in which the code rate is selected based on the fading inside a certain region. Upper bounds on the achievable average spectral efficiencies were compared with the ergodic capacity, as well as the average spectral efficiencies for the scheme with rate adaptation only. We found that a very small number of codes, eg. 1 or 2, would still get us very close to capacity. There were also considerable performance gains over the rate adaptation only case. We also considered a simple multi-user extension of our formulation, and proved some asymptotic (high average SNR and large number of users) results when the fading is Rayleigh. Advanced Algorithms for MIMO-OFDM ReceiversMultiple-Input-Multiple-Output (MIMO) systems use multiple antennas at the transmitter and/or receiver. This creates spacial diversity amongst the different radio paths between the transmitters and receivers. MIMO systems are becoming increasingly popular for high-bit-rate wireless communication systems as they achieve a higher channel capacity than single antenna systems with the same radio spectrum. The throughput of a MIMO system can be further increased by incorporating broadband signals (larger frequency range). An efficient way of doing this is to incorporate orthogonal-frequency-division-multiplexing (OFDM) into the MIMO structure, i.e. MIMO-OFDM (multiple antennas and multiple radio channels). This project's specific objectives are:
We investigated advanced receiver algorithms for MIMO-OFDM systems. The major receiver signal processing issues such as time and frequency synchronization, channel estimation and tracking, and data detection have been studied with specific emphasis on low complexity algorithms suitable for practical implementation. We have developed novel low-complexity time and frequency synchronization and data detection algorithms for OFDM based systems with multiple transmit and receive antennas. We have used optimal peak-to-average-power ratio reduction methods with power spectral density constraints on reserved tones conforming to ADSL2 standards. In addition, new power and rate control algorithms have been developed for OFDM based systems for wireless fading channels with outage probability guarantees. We have also designed novel power and rate control algorithms for CDMA based ad hoc networks with data rate outage and route failure probability constraints. These results have been extended to include state of the art receivers employing statistical channel prediction methods to avoid latency issues. Time-synchronization for OFDM Systems with Dispersive ChannelsIn 2004, we further developed approaches for OFDM time synchronization and carrier offset recoveries that are specifically tailored for dispersive channels. More specifically, we have been dealing with fast-fading channels and utilizing averaged signal correlation statistics of these dispersive channels as opposed to the instantaneous conditions themselves. This work utilizes a cyclic prefix, which is inserted to guard against interference over time and to allow for perfect equalization at the OFDM receiver. Previous work led to a new technique which asymptotically finds the optimal timing to minimize the combined interference over time and between channels. Future Plans (Signal Processing)There are many open and interesting research problems that have arisen or have been formulated due to the research carried out in CUBIN in the Signal Processing project in 2004. The following research directions are by no means exhaustive; however, they give some indication of the many interesting projects we will undertake in 2005 and 2006. These projects include:
Optical Networking
Despite the recent telecommunication industry downturn and the consequent delay in the deployment of optical telecommunication technologies, growth in use of the Internet continues unabated. The driving forces behind optical networking continue - namely the demands for increased network capacity, greater operational flexibility, and improved cost-effectiveness. The broad aims of the Optical Networking project are to:
Monitoring and Buffering in Optical NetworksThe Optical network to support next generation internet is envisioned to be all-optical where the communication channels remain in the optical domain throughout. The traditional approach of relying on SONET/SDH to monitor and maintain the network needs to be revised for such a network. We anticipate that various optical performance monitors will be placed at ingress/egress ports of the optical cross connects at either the core or the edge of the network. These monitors will observe important parameters of optical channels, such as wavelength, power, optical signal-to-noise ratio (OSNR), bit-error-ratio and chromatic dispersion.
Caption: Optical performance monitoring in optical network
We have built demonstrations of an optical dispersion monitor, based on the RF-fading technique. Our monitor uses two time-multiplexed in-band sub-carrier tones to provide increased range and resolution, compared to previous methods. We have also demonstrated a new method for measuring optical signal-to-noise ratio using signal-spontaneous beat noise. In our new technique, the modulated signal together with the amplified spontaneous emission (ASE) noise is split into two paths. Using optical and electrical signal processing, the signal component is removed by RF subtraction and uncorrelated beat noise is measured to monitor optical signal-to-noise ratio. This is the first time that balanced RF subtraction has been used to monitor optical signal-to-noise ratio. We have shown experimentally that our monitor provides good performance.
Caption: A new technique for measuring optical signal-to-noise ratio
In collaboration with researchers at the University of California, Berkeley, we have investigated the application of so-called "slow light" to the storage of optical data. One outcome has been the development of a new theory of slow light optical buffers for optical signal processing and optical packet switching. In addition, we have proposed a number of new structures for delay line buffers based on slow light, including improved architectures for first-in-first-out (FIFO) and random access memory (RAM) buffers. This work provides a framework for work we are planning in 2005 and 2006 on optical buffering technologies and optical packet switching for the all-optical Internet. Passive Optical NetworksTo bring optical networking to the customer, we are investigating a number of possible architectures for passive optical networks (PON's). Because of their high bandwidth requirements, business customers are the first adopters of fibre in the local loop so we have developed a novel architecture well suited to them. This wavelength division multiplexed passive optical network (WDM PON) consists of an optical network terminal (OLT) at the central office and a number of optical network units (ONU's) at customer premises. It also includes optical cable plant with an arrayed waveguide grating (AWG), an optically passive wavelength division multiplexer, located in the network.
Caption: A new architecture for Passive Optical Networks (PONs)
One optical network unit (ONU) can serve a business building or one business entity in a business park with a data rate of up to 10 Gbit/s. Business customers can access to the external world using their own routers that are linked to the central office router at a speed they choose. In addition, a closed group of business customers can create their own private network that is overlaid on the common physical fibre cable network. This optical virtual private network (OVPN) reduces costs by sharing the cost of the AWG among customers without compromising the information handling capacity of the system. The OVPN concept has been experimentally demonstrated at 10 Gbit/s and its flexibility in supporting an arbitrary topology has also been confirmed. The protection of networks is critical for business customers. We have demonstrated that such high speed OVPNs can be protected against cable cuts or equipment failure. We have also experimentally verified that this basic architecture can be extended to a metropolitan scale by the use of modular arrayed waveguide gratings. Eventually, as costs continue to drop; this architecture will migrate to residential customers. Since broadcasting of broadband services such as CATV, IPTV, and high quality music are the major demand in the residential marketplace, we have developed a novel technique allowing such services to be broadcast, multicast, or unicast to the customers. This technique has been experimentally demonstrated and research into the optimum operational conditions is still underway. This work will be carried out in collaboration with the Victoria Laboratory of NICTA. Optical Burst SwitchingOptical Burst Switching (OBS) is a new approach to optical switching that can potentially eliminate the need for optical buffering. A number of researchers have postulated that optical burst switching may one day become attractive but there have been few studies of the cost benefits. Using the costing framework that we developed in 2003 for optical circuit-switched networks, we have compared, for the first time, the economics of optical burst-switching and optical circuit-switching for a national network covering Australia, with 8 million users, each receiving a broadband bi-directional service at an average bit rate of 10 Mbit/s. Our results shed light on the prospects of optical burst switching. The outcome of our analysis appears to defy conventional wisdom as it shows that, for the network we considered, burst-switching may not provide an economical approach to building high-capacity networks compared to circuit-switched networks. Our work in this area is preliminary, but forms the basis of what we plan to be a major study of the economical advantages of different network technologies for networks of different sizes and different capacities. While demonstrating that currently proposed optical burst-switching technology will find it hard to beat its circuit switched counterpart, we have developed a novel method that significantly reduces the blocking probability of optical burst switching networks. The idea is based on two concepts. The first is a well known approach called "deflection routing" where a burst of data that is blocked due to congestion is deflected to other routes and the second involves giving a deflected burst lower priority. Although preliminary simulation studies have shown a significant benefit to this new approach in terms of reduction in blocking probability, a thorough evaluation the economic advantages of the new method is a matter for future study. A fundamental design issue associated with optical burst switching is the so-called signalling offset. Capacity reservation requests are sent in advance of each burst. The advance notice is called the offset time. Processing of reservation requests takes a non-negligible amount of time that results in a reduction of the offset time as the burst traverses the network. If the reduction is significant enough, it can increase the chance of data loss as the burst gets closer to its destination. We have performed detailed simulation experiments to help design engineers determine how fast they must make the electronic controllers at each cross-connect in order to minimise this deleterious effect. Another important question associated with optical burst-switching technology is how it will perform in conjunction with TCP. In collaboration with the Optical Internet Research Center (OIRC) in Korea, we have developed a mathematical model that approximates the performance of TCP over optical burst switching. We plan to evaluate this mathematical model through both a simulation and an experimental test-bed available at the OIRC. Preliminary analyses, for the scenarios considered, show that TCP over optical burst-switching will achieve network utilization of about 55%. Designing OBS networks to deliver customers' data with guaranteed service levels requires a good understanding of network performance. We have obtained such an understanding through an analysis of different carrier scenarios and have derived simplified models of network behaviour. Using well-known techniques from classical circuit-switching theory and operations research, we have developed techniques to minimise OBS network cost and required capacity whilst still meeting customer needs. These OBS network analysis and design tools empower operators to make business decisions about competing network architectures and future upgrade paths by comparing their respective network design costs and revenue models. ASON Research ProjectAn exciting initiative in 2004 was the ASON research project which is part of a larger collaborative research agreement between CUBIN and Telstra. The Automatically Switched Optical Network (ASON) includes a set of software modules that has been standardised by the International Telecommunication Union (ITU). ASON provides the capability to automatically discover the optical network topology and to dynamically control network connections for carrier-grade backbone optical networks. ASON promises to deliver reduced operating costs as well as differentiated levels of quality of service. The aim of this project is to design and develop a simulator to model an ASON. We will investigate various ASON architectures and perform tests to determine the applicability and limitations of ASON in carrier grade backbone optical networks. We will focus on connection determination algorithms, link and node failure cases and re-routing scenarios to determine if the desired quality of service standards can be achieved and maintained. Significant progress was made on the Automatically Switched Optical Network simulator in 2004. A modular approach has been taken so that the simulator can be used in a wide range of network architectures. Several of the ASON features, such as the automatic detection of link and node failure, failure recovery and the automatic detection of new nodes in the network have already been demonstrated. The network routing and dynamic re-routing algorithms are under development.
Performance Assurance
The performance assurance project is developing performance evaluation models for modern and future telecommunications networks. These models enable the development of new network design methods for wireless and wired based networks and systems of both the present and the future. Over 50% of this project's resources are devoted to research on wireless networks where customers are connected to the network via radio links. These networks include mobile cellular technologies such as those based on Global System for Mobile communications (GSM) and Code Division Multiple Access (CDMA), and wireless local access networks that provide wireless Internet connections to customers, such as those based on the IEEE 802.11 standards. Wireless communication is increasing in importance in our daily life and, despite the downturn in the telecommunication and information technology industries, wireless networking continues to grow at a rapid pace. Wireless is an important access medium to the broadband internet, as it allows users more freedom, and the ability to set up networks at any desired location in a building, or outdoors. There has been a huge growth in the deployment of wireless local area networking (WLAN) over the past 5 years. At the same time, outdoor cellular networks have continued to grow, and in 2004, the first 3G networks became available. These networks offer increased data rates, and new services such as mobile pictures, and video streaming. Other Performance Assurance research focuses on the Internet. Using traffic measurements, we develop models that enhance our understanding of Internet traffic statistics. We also study how Internet content providers can better store their content on servers to provide more efficient and reliable service to customers. New methods and tools for analysis of network measurements analysis and simulation are being developed, that enable more reliable networks and better quality of service to customers at lower cost. The broad aims of this project are to:
Performance of Wireless LANsIn recent years, demand for wireless Internet connectivity has led to a proliferation of wireless local area networks (WLANs). These have been deployed as an extension of the wired Internet in offices, homes, factories and as hotspots in public venues such as airport terminals and restaurants. Products based on the IEEE 802.11 family of standards have captured the lion's share of the burgeoning wireless local area network market. A typical 802.11 wireless local area network consists of a base station, or access point, and a population of wireless clients within radio range (usually < 50 m) of the access point. As in the wider Internet, the majority of network traffic carried on a typical IEEE 802.11 WLAN today consists of non-real-time applications such as web browsing and email that are carried over the TCP transport protocol. However, in the near future, it is expected that a significant proportion of the traffic on WLANs will consist of new real-time services such as voice over IP and video.
A number of important models have been developed to predict the performance of WLAN's using the 802.11 MAC protocol. We have developed a methodology for evaluating packet delay performance of the IEEE 802.11 MAC protocol. This will help in understanding the potential for IEEE 802.11 WLANs to support real-time applications. Another important research direction is to extend the models to non-real-time applications, such as file transfers and web browsing. The model must include the flow control aspect of the TCP protocol which controls when these traffic sources can actually inject data into the network. We have developed a new performance model that is extremely accurate in predicting throughput and delay for traffic that is controlled by the ubiquitous TCP protocol. It captures the key control aspects of TCP, and the random back-off and collision avoidance of the IEEE 802.11 MAC protocol. One novel feature is the incorporation of TCP, allowing the model to describe the performance of a large class of web traffic, including web browsing, and file transfers. Another novel feature is the ability to handle both uplink and downlink traffic (not allowed in prior models for non-TCP traffic). This model allows a network administrator to calculate the number of access points required to handle the traffic in a given network to a given grade of service (transaction latency). The model can therefore be used for network dimensioning. The IEEE 802.11 protocol is not always efficient, especially when many users share the WLAN simultaneously. The collision avoidance strategy of the protocol means that two users cannot transmit simultaneously, even if a small amount of extra coding and redundancy would allow this to occur. Moreover, the protocol does not take full advantage of wireless channel conditions, when they are favourable. We are developing a new protocol that allows multiple transmissions to be simultaneously successful, and which also adapts to the wireless channel conditions, not just to the amount of contention between users. At present, we are using a theoretical model that provides a rigorous testing ground for the control aspects of our new protocol. In future work, we will conduct experiments to test the protocol in real-world situations. Our new protocol requires a feedback rate of 1 bit per timeslot, irrespective of the number of nodes in the network. We have shown that having only 1 bit per slot does not reduce the throughput of the protocol, as compared to similar protocols with much higher rates of feedback. In fact, the throughput is not dependent on the feedback rate. Our protocol requires about the same amount of feedback as the distributed mode of 802.11, yet it can achieve far greater throughput, due to its ability to have multiple simultaneously successful packets. It requires far less feedback than 802.11 in centralised mode, and provides superior throughput even for that case. On the other hand, the work makes a number of assumptions, and is not yet in a form for practical deployment. Its main value lies in the insight it provides for the development of new, practical MAC algorithms for the next generation wireless LANs, and data carrying cellular networks. Combined Power Control and Packet SchedulingUsually, power control is considered at the physical layer, and packet scheduling at the MAC layer. We have developed a joint power control and packet scheduling algorithm that updates both the transmit power and the proportion of bandwidth allocated to each user in the network. The natural application is to a base station in a cellular network, transmitting to a set of users on the downlink. The objective of the power control algorithm is to find the right allocation of power for the transmitting node, and the right bandwidth allocation to each link, to satisfy the rate requirements of each user in the network. A key assumption is that the power allocated to each timeslot is constant. Thus, the base station will send at a constant power, and does not need to adapt its power on a packet by packet basis, as would be needed in current proposals. This has attractive implications for the total energy consumption of the amplifier at the base station. These implications may be more important in other applications, where the transmitting node is energy-limited, such as occurs in a multi-hop, ad-hoc network. We have shown that the algorithm converges if the rate constraints are feasible under the assumption of constant transmit power at the base station node. We have observed that the set of achievable rates appears to be almost identical to that provided by schemes in which the bandwidth per receive node is first fixed, and then power is allowed to vary on a packet by packet basis, in a more traditional manner, as studied in prior work on power control. Evaluation of Handoff Algorithms in Mobile NetworksThe coverage area of a cellular mobile network is divided into sub-areas called cells. Each cell has a base station that receives and transmits the communication signals to users. When the user is too far from its base station, the quality of the signal is reduced and the call may drop out in the middle of a conversation. Handoff is a procedure of transferring an ongoing call from one cell (base station) to another as a user moves through the mobile network crossing cell boundaries. As the demand for mobile services increases, it leads to reduction in cell size in congested areas, so the number of handoffs grows. As a result, the handoff mechanism will have an increasingly important impact on the overall performance of a mobile network. This makes managing handoffs efficiently an important research topic. A good handoff scheme is efficient in terms of usage of radio transmission resource, and capable of meeting required quality of service as expected by the users. During 2004 we have started a development of a framework for performance evaluation and comparison between exiting handoff algorithms. The key development was to define a comprehensive call quality signal level measure in our framework. With this new proposed measure we are now able to obtain a benchmark for handoff mechanism in mobile networks. This is particularly important because it can be used to compare existing and future handoff procedures, so that telecommunications providers can choose the best algorithm to optimize handoff management functions. Understanding and Modelling Internet Traffic StatisticsLinks in large networks such as the Internet carry packets from many thousands of different connections simultaneously. Modelling this aggregate traffic by random processes involves considering the many potential sources of dependency and complexity in the data, including the dynamics of restless users browsing the Web. Based on many gigabytes of high precision packet level data collected on large Internet links and operational routers, our traffic modelling work seeks to understand, both in statistical and physical network terms, the origins of key statistical features of packet data. It then seeks to utilise the insights from the modelling to find solutions to traffic measurement and engineering issues such as traffic sampling, high speed generation, and router performance monitoring. An example of the statistics obtained is given in the Figure below, which displays a density plot giving insight into the variety of packet flows present in a large trace. The colorized density provides a breakdown based on two criteria: the rate of packet arrivals within a flow, and the number of packets the flow contains. The densities are on a log scale, so that each increase in `darkness' corresponds to a significantly higher density. What we see from the figure is that although flows are diverse, there is sufficient concentration in a narrow region to allow the definition of a representative value of flow rate and size, which can be used to simplify reality, leading to a compact model describing the packet arrival process.
Building on an understanding of packet traffic provided by our previous work with colleagues at Sprint ATL in California, a traffic `replay' system was designed which allowed stored traces from very high speed networks to be accurately replayed using commodity PCs. This allowed an on-line measurement system to be tested under realistic conditions in the laboratory for later implementation in the network. We also extended our previous traffic modelling work, from a link, to a network node. We showed that the crucial merging and splitting properties required to move from the link to the node level are present in the model, and also apply to real data, as collected at an operational router at Sprint ATL. This opens up the possibility of a network level traffic calculus to supersede the traditional one based on Poisson streams, and the associated independent Poisson splitting and merging properties. The new network model would be capable of describing burstiness over a wide range of time scales, including accounting for long-range dependence. Performance Evaluation and Optimization of Video-on-demand SystemsVideo-on-demand (VOD) services to a large population of users could be a major driver of future broadband traffic. In such a system, a large collection of movies needs to be managed and stored, possibly by a service system made of a large cluster of on-line disks. Because of the enormous storage space and I/O (Input/Output) bandwidth required to store and deliver movie contents, it's important to manage the resources efficiently while at the same time meeting user Quality of Service (QoS) requirements. The aim of this project is to investigate optimal file assignment in a large scale VOD system. The file assignment problem is concerned with how to replicate and allocate the movie files to the disks so as to minimize the blocking probability. To ensure that movies are truly available on demand, it's necessary to develop scalable and accurate analytical solutions for the blocking probability of the system, and to design efficient heuristic file allocation algorithms and search techniques that work within reasonable computing time. In 2004, we have been studying how load balancing can be used to attain efficient operation and minimum blocking probability. We have found that a certain ideal file allocation instance always yields the lower bound blocking probability of the system. This led us to design an efficient numerical index to quantify how evenly a file allocation instance distributes the traffic loads on multi-copy files. Using this, we have devised an efficient heuristic file allocation algorithm that can achieve near-optimal blocking performance of the system.
Caption: A video-on-demand system
Active Networks
Active or programmable networking is a new approach to networking in which packets not only carry data but also information for network control. The control information can be as simple as Quality of Service (QoS) parameters for a flow control algorithm or as complex as a small program to be executed at a router or switch. This enables a network to be restructured very quickly and allows new services to be provided simply and flexibly. Such flexibility will be critical in future networks. The Active Networking project aims to improve network control and performance by closely linking software and hardware aspects of active networking. The broad aims of this project are to:
Project: Active Networks for Controlling Denial-of-Service AttacksThe aim of this project is to use active networks to identify and control denial-of-service attacks and other types of malicious traffic on the Internet. By using active and programmable networking, each router can selectively activate monitoring in response to a problem, and cooperate with neighbouring routers to isolate the cause of the problem. For example, denial-of-service attacks can severely disrupt the Internet, but can be extremely difficult to trace due to the attacker's ability to fake the source address in IP packets. Previously, we developed a practical scheme to defend against Distributed Denial-of-Service (DDoS) attacks based on IP source address filtering. This approach enabled us to filter attack traffic when it reached the victim. We have now finalized an agreement with a local company (IntelliGuard IT Pty Ltd) to commercialise this defence scheme. We will jointly develop our DDoS defence scheme as a commercial product to protect local networks or web servers that are connected to the Internet. This project with IntelliGuard IT won the 2004 Melbourne University Entrepreneur's Challenge (MUEC). MUEC is an annual competition to recognize innovative ideas and business plans. It's run by the Melbourne Business School and is sponsored by Hewlett Packard. The competition is open to all Victorian tertiary institutions. Our winning bid led a field of six finalists, including projects in the fields of biomedical science, telecommunication and electronic commerce. Improving the Efficiency of Web Search EnginesThe aim of this project is to improve the quality of information retrieval from search engines. It's an ongoing challenge in large information networks to find relevant documents when searching for information. Today's search engines rely on word frequencies to rank documents, and ignore any structure or word flow throughout the documents. Previously, we developed a search technique called spectral-based document retrieval to overcome this problem. The spectral-based technique records not only the number of times each word appears in a document but also its position throughout the document. This allows us to find documents that have matched-phase keywords (implying that the words appear near each other). We also developed novel techniques to improve the efficiency and accuracy of our spectral-based approach by applying Discrete Cosine Transformations. Recently we developed a new scheme by analyzing the spectra of query terms via certain wavelet transforms; we were able to capture the position and frequency of each of the terms in a compact manner. This information can be used to retrieve lists of documents with high accuracy (greater than those using the Fourier or Cosine Transform), while significantly reducing the amount of data that needs to be stored by the search engine. We followed this work by applying latent semantic analysis. Normally these techniques are very slow in practice. However, we have developed a new method which makes this scheme very efficient both in terms of storage and processing. We also developed a new ranking algorithm which has given the best precision of all the known schemes. We have not published this work as we intend to patent the algorithms.
Design Automation
The Design Automation project develops tools for the design and analysis of ultra-broadband networks. All CUBIN staff and students contribute to this project through high quality contributions to the fundamental theory and practice of design, implementation and management of ultra-broadband networks. Much of the intellectual property developed in the Centre is in the form of algorithms and design tools. There were a number of developments in this project in 2004 including additions to the CUBIN Software Repository and the expansion of the CUBIN Laboratory (CUBINlab). CUBINlab: The CUBIN LaboratoryThe CUBIN Laboratory is a vital facility that supports the IT needs of CUBIN staff and students. CUBINlab is a UNIX based integrated network environment providing and supporting:
The CUBINlab core network includes a firewall, a server with redundant power supply and comprehensive tape backups, a Global Positioning Satellite (GPS) synchronised time source, and both wired and wireless LAN connectivity.
Caption: CUBINlab - The CUBIN Laboratory
Highlights for 2004 include:
More details on each of these satellite entities and the research that they support is available on the CUBINlab web page: www.cubinlab.ee.unimelb.edu.au/. The CUBIN Software RepositorySoftware is an increasingly important part of today's communication networks. The complex management and signal processing tasks are generally performed by sophisticated algorithms running either on general purpose processors, or on digital signal processors. Communication research relies heavily on software simulation, both to verify theoretical calculations and to guide researchers in promising directions. When the subject of research is an algorithm, the distinction between "simulation" and "implementation" becomes blurred, and simulations play the role that prototypes play in traditional engineering research. A key to the rapid implementation of reliable and efficient software is the ability to reuse software. To this end, CUBIN is compiling a collection of reusable software modules. In 2004, the software repository expanded to include source code for real protocol implementations. In particular, it now includes a free implementation of an IEEE802.1x authenticator, written using FreeBSD's NetGraph technology. This software was contributed by a department alumnus Philip Crumpler. It improves the security of wireless LANs, by allowing "unauthenticated" traffic to be discarded, and by allowing cryptographic keys to be exchanged and updated. Another major addition to the software repository is a module for the popular NS2 simulation package to implement Caltech's FAST TCP congestion control protocol. This module was written at the request of Caltech researchers, and has the potential to become the "official" FAST module for NS2. The initial development of this software was undertaken by CUBIN student Tony Cui and the development is being continued outside CUBIN. Distribution of the module is still supported by CUBIN's Software Repository infrastructure.
Other Research ActivitiesCUBIN Staff and students are involved in a number of other research activities that are outside the core research program. A selection of these non-core projects is listed below.
Efficient Macro-mobility Management for GPRS IP Networks (joint collaboration between CUBIN and OIRC researchers) |
CUBIN
SS
Caption: Ad-Hoc wireless network showing the logical interconnection of a section of the network. A number of sources are transmitting information to their next-hop neighbour along a set of logical links.
Caption: A typical wireless local area network
Melbourne School of Engineering | Disclaimer & Copyright | Privacy |
Date Created: 17 August 2004
|
© The University of Melbourne 1994-2008
|