Prof. Ariel Orda, Dept. of Electrical Engineering, Technion - IIT Title: Over Two Decades of Research on Networking Games Date: 12/02/2004 Time: 2:00 PM - 3:00 PM Location: Hawthorne 1S-F40 Host: Giovanni Pacifici

Abstract:
The application of game theoretic models and tools in the fields of networks and distributed systems has been gaining increasing attention in recent years. From what was a rather "esoteric" area of research not many years ago, it has grown to be well within the mainstream of the networking and distributed systems research agenda. In this talk we will review the major stages in the evolution of this area. First, we will move back 4-5 decades and examine some of the early game theoretic advances in related fields such as queuing systems and transportation networks. Then, we will move on to about 2 decades ago, and examine the early steps of game theory in communication networks. We will consider the early motivation, contrast it with current motivation, and overview some of the fundamental results in flow-control and routing games. Moving on to more recent studies, we will observe the main issues at stake, namely characterization of the network properties and performance in a noncooperative environment, and an attempt to improve the inherent inefficiency that stems from unregulated behavior, via proper design and management tools (in particular, pricing). We will conclude by outlining several current challenges and opportunities for future work.

Biography:
Ariel Orda received the B.Sc. (summa cum laude), M.Sc., and D.Sc. degrees in Electrical Engineering from the Technion––Israel Institute of Technology, Haifa, Israel, in 1983, 1985, and 1991, respectively. Since 1994, he has been with the Department of Electrical Engineering at the Technion. He was a visitor at IBM Watson, Columbia University and Bell Labs. He has held several consulting positions with Israeli industry. Prof. Orda received several research and teaching awards, most recently for ICNP'2004 Best Paper. He served as Program co-chair of IEEE Infocom'2002 and is an Editor of the IEEE/ACM Transactions on Networking and of the Journal of Computer Networks. His current research interests include network routing, survivability, QoS provisioning, wireless networks, the application of game theory to computer networking and network pricing.

Matei Ripeanu, University of Chicago Title: What Globus and PlanetLab can learn from each other? (Resource Management in Federated Distributed Systems) Date: 11/19/2004 Time: 1:00-2:30 PM Location: Hawthorne GN-F15 Host: Asit Dan

Abstract:
PlanetLab and Globus Toolkit are gaining widespread adoption in their respective communities. Although designed to solve different problems (PlanetLab is deploying a worldwide infrastructure testbed for experimenting with network services, while Globus is offering general, standards-based, software for running distributed applications over shared resources) both build infrastructures that enable federated, extensible, and secure resource sharing across trust domains. Thus, it is instructive to compare their resource management solutions. To this end, I'll review the approaches taken in the two systems, attempt to trace back to starting assumptions the differences in these approaches, and explore scenarios where the two platforms can cooperate to the benefit of both user communities. I believe that this is a key first step to identifying pieces that could be shared by the two communities, pieces that are complementary, and how Globus and PlanetLab might ultimately evolve together. This evaluation is based on first hand knowledge and experimentation with Globus and PlanetLab software. I will conclude with an overview of two Grid-related projects active at The University of Chicago. Firstly, a Replica Location Service targeted for environments where replica location queries are prevalent but the dynamic component of the system (e.g., node failures, add/delete operations) cannot be neglected. The solution I propose is based on three mechanisms: probabilistic representations of location information, soft-state protocols, and a flat overlay network to disseminate state information; and offers important benefits: low query latency, support for finding collocated sets of replicas, and adaptability to dynamic settings.  (This project is part of my thesis work.) Secondly, the larger Diaspora project, sponsored in part by IBM, aimed at building systems that are as deployable (i.e. scalable and supporting incremental deployment) as today’s P2P systems and but are capable to offer the complex services provided by today’s Grids.
Bio: http://people.cs.uchicago.edu/~matei/bio.html
For more information:  http://people.cs.uchicago.edu/~matei

Carsten Ernemann, University Dortmund, Germany Title: Design and Evaluation of Online-Scheduling Algorithms with Computational Intelligence Date: 11/10/2004 Time: 1:00-2:00 AM Location: Hawthorne 4S-K21 Host: Vijay K. Naik

Abstract:
This talk presents a methodology for automatically generating online scheduling algorithms for a complex criterion defined by a machine owner. This research is focused on online scheduling with independent parallel jobs, multiple identical machines and a small user community. First, evolutionary algorithms are used to create a 7-dimensional solution space of feasible schedules of a given workload trace. Within this step no preferences between different basic criteria need to be defined. This solution spaces enables the resource providers to define a complex evaluation criterion based on their specific preferences. Second, optimized scheduling algorithms are generated by using two different approaches. On the one hand, an adaptation of a Greedy-Scheduling algorithm is generated which uses weights to create an order of jobs. These job weights are extracted again from workload traces with the help of evolutionary algorithms. On the other hand, a Fuzzy rule based scheduling system will be applied. Here we classify a scheduling situation which consists of many parameters like the day time, the week day, the time, the waiting queue length etc. Depending on this classification, a Fuzzy rule system chooses an appropriate sorting method for the waiting job queue and a suitable scheduling method. Finally both approaches, the Greedy scheduling and the Fuzzy rule based scheduling system, will be compared by using again workload traces.

Matt Welsh, Harvard University Title: Market-Based Programming Paradigms for Sensor Networks Date: 10/20/2004 Time: 2:00 -3:30 PM Location: Hawthorne 1S-F40 Host: Fred Douglis

Abstract:
Sensor networks present a novel programming challenge: that of achieving robust global behavior despite limited resources, varying node locations and capabilities, and changing network conditions. Current programming models typically require that global behavior be specified in terms of the low-level actions of individual nodes. This approach makes it extremely difficult to tune the operation of the sensor network as a whole.Ideally, sensor nodes should self-schedule to determine the set of operations that maximizes that node's contribution to the network-wide task. In this talk, we present market-based macroprogramming (MBM), a new approach for achieving efficient resource allocation in sensor networks. Rather than programming individual sensor nodes, MBM defines a virtual market in which nodes sell goods (such as sensor readings or data aggregates) in response to prices that are established by the programmer. Nodes take actions to maximize their profit, subject to energy budget constraints. The behavior of the network is determined by adjusting the price vectors for each good, rather than by directly specifying local node programs. Nodes individually specialize their operation in response to feedback from payments. Market-based macroprogramming provides a useful set of primitives for controlling the aggregate behavior of sensor networks despite variance of individual nodes. We present the MBM paradigm and a sensor network vehicle tracking application based on this design, as well as a number of experiments demonstrating that MBM allows nodes to operate efficiently under limited energy budgets, while adapting to changing network conditions. This project is in collaboration with Geoff Mainland and David Parkes.

Biography:
Matt Welsh is an assistant professor of Computer Science at Harvard University. Prior to joining Harvard, he received his Ph.D. from UC Berkeley, and spent one year as a visiting researcher at Intel Research Berkeley. His research interests span many aspects of complex systems, including Internet services, distributed systems, and sensor networks.

Prof. Carlos Varela, Rensselaer Polytechnic Institute Title: Middleware for Dynamic Decentralized Distributed Computing Date: 09/30/2004 Time: 11:00 - 12:00 PM Location: Hawthorne Cafeteria AnnexHost: Alan Bivens

Abstract:
The Internet is becoming a ubiquitous platform for high-performance distributed computing. In this talk, we present a new software framework for distributed computing over large scale dynamic and heterogeneous networks. Our framework wraps computation into autonomous actors, self organizing computing entities, which freely roam over the network to find optimal target execution environments. We introduce an actor-oriented programming language (SALSA), a distributed run-time environment (WWC), and a middleware infrastructure for autonomous reconfiguration and load balancing (IOS). Load balancing is completely transparent to application programmers. The middleware triggers actor migration based on profiling resources in a completely decentralized manner. Our infrastructure also considers the dynamic addition and removal of nodes from the computation, while continuously balancing the load given the changing resources. To balance computational load, we introduce peer-to-peer random work stealing (RS). We also introduce two more informed variations thereof: application topology-sensitive (ATS), and network topology-sensitive (NTS) work stealing. We evaluated RS and ATS with several actor interconnection topologies in a local area network. While RS performed worse than static round-robin (RR) actor placement, ARS outperformed both RS and RR in the sparse connectivity and hypercube connectivity tests, by a full order of magnitude. In essence, the presented middleware is a decentralized virtual network of agents. The framework modularity provides a testbed to evaluate distributed systems performance and scalability under different middleware agent topologies, application and resource profiling mechanisms, peer-to-peer communication protocols, and application reconfiguration decision functions.

Biography:
Carlos Varela is an Assistant Professor in the Computer Science Department at Rensselaer Polytechnic Institute, where he leads the Worldwide Computing Research Group. His main areas of current research interest include web-based and internet-based computing, middleware for adaptive distributed systems, concurrent programming models and languages, and software development environments and tools. Carlos received his B.S. with honors, M.S., and Ph.D. in Computer Science, all from the University of Illinois at Urbana-Champaign. He was previously a research staff member at IBM T.J. Watson Research Labs and he helped start up a software development company in Tokyo, Japan: International Systems Research. Carlos has been an active participant in software development efforts leading to public domain releases of major software packages, including: NCSA Shape2D, NCSA httpd, NCSA Mosaic, and the SALSA programming language. Carlos has given over two dozen lectures about his research nationally and internationally, including seminars in Switzerland, Japan, Colombia, Canada, Germany, Australia, Netherlands, France, and the Czech Republic. For access to publications and more information on his group, please visit: http://www.cs.rpi.edu/wwc/

Prof. Bobby Bhattacharjee, University of Maryland Title: Replication and Search in Distributed Namespaces Date: 09/09/2004 Time: 11:00 AM - 12:00 PM Location: Hawthorne 4S-K21 Host: Sumeer Bhola

Abstract:
Peer-to-peer systems can be used to form low-latency decentralized, distributed namespaces, such as Distributed Hash Tables (DHTs) and Distributed Directories. Beyond efficient data lookup, there are two fundamental problems all of these systems have to address: how to avoid load imbalances amongst peers, and how to provide efficient and useful search capabilities? Structured systems (such as DHTs) provide both low latency and excellent load balance with uniform query and data distributions. Under the more common skewed access distributions, however, individual nodes are easily overloaded, resulting in poor global performance. In the first part of the talk, I will describe a lightweight, adaptive, and system-neutral replication protocol, called LAR, that maintains low access latencies and good load balance even under highly skewed demand. Searching P2P namespaces is a topic of much current interest. I believe large scale efficient searching will necessarily entail maintaining distributed indexes. In the next part of the talk, I will focus on identifying and addressing the structural challenges posed by such large-scale distributed indexes. I will discuss strategies for storing potentially extremely large indexes, techniques for efficient caching/reuse of search results, and mechanisms for updating indexes as data and the set of participating servers change.

Biography:
Bobby Bhattacharjee is an assistant professor in the Computer Science department at the University of Maryland, College Park. His research interests are in the design and implementation of scalable systems, protocol security, and peer-to-peer systems. He is a member of the ACM, and a fellow of the Alfred P. Sloan Foundation. For more information please visit http://www.cs.umd.edu/~bobby/

Prof. Arno Jacobsen, University of Toronto Title: Toward Aspect Oriented Middleware: About Concepts, Experiments, and Tools Date: August 17, 2004 Time: 10am -11am Location: Hawthorne 1S-F40 Host: Stefan Tai

Abstract:
Over the past decade, middleware platforms such as CORBA, J2EE, and Web Services have become increasingly popular, addressing software engineering problems for distributed application development in the most diverse application domains and under greatly varying requirements. For many of these platforms, these proliferations have led to complicated architectures, complex abstractions, and bloated implementations. We think, this is mainly due to the interaction of too many concerns, imposed from a wide range of requirements, and the lack of being able to adequately modularize these concerns. For example, it is not possible to easily customize a given middleware implementation or flexibly adapt it to address ever changing domain requirements. In this talk we will present results drawn from our on-going Aspect Oriented Middleware research project. We will first present examples illustrating the phenomenon of logic tangling and implementation convolution, which we believe to be central for the above outlined problems. This phenomenon is not due to bad design but rather due to the limitations of conventional architectural decomposition methodologies. We will show how aspect oriented programming (AoP) techniques can remedy this problem. We will present results on quantifying logic tangling in several existing middleware implementations through aspect mining. We will briefly review our mining methodology and, if time permits, the Prism mining tool we have developed to aid in the discovery of aspects in large code bases. Our aspect mining results clearly indicate the inherent presence of aspects in (legacy) platform implementations. Moreover, our extensive aspect oriented re-factoring efforts constructively prove that AoP can improve the customization, configurability, modularity, and computational efficiency of an aspect oriented middleware implementation. This research is joint work with Charles Zhang and will draw from our following publications: C. Zhang and H.-A. Jacobsen. Resolving Feature Convolution in Middleware Systems. ACM OOPSLA 2004. C. Zhang, H.-A. Jacobsen. Refactoring Middleware Platforms. IEEE Trans. on Parallel and Distributed Systems, November 2003. (Extended version of our AOSD 2003 paper.) C. Zhang, H.-A. Jacobsen. Refactoring Middleware Platforms: A Case Study. International DOA Conference, Italy, November, 2003. The Prism Tool Make-shift Web site: http://www.eecg.toronto.edu/~jacobsen/prism/

Biography:
Hans-Arno Jacobsen holds a faculty position with the Department of Electrical and Computer Engineering and the Department of Computer Science at the University of Toronto, where he leads the Middleware Systems Research Group. His principal areas of research include middleware systems, distributed systems, and information systems. Arno's current research focuses on distributed event-based processing and aspect oriented software development. Arno received his Ph.D. degree from Humboldt University, Berlin in 1999 and his M.A.Sc. degree from the University of Karlsruhe, Germany in 1994. Between 1992 and 2000 Arno traveled around the world, working at various research institutes. including LIFIA in Grenoble, France, ICSI in Berkeley, U.S., LBNL in Berkeley, U.S., and INRIA in Rocquencourt, France. Arno has served as program committee member of various international conferences, including ICDCS, OOPSLA, Middleware, and VLDB. He is the Program Chair of the 5th International Middleware Conference in Toronto. For more information please visit http://www.eecg.toronto.edu/~jacobsen

Prof. Arno Jacobsen, University of Toronto Title: Scalable Matching for Publish/Subscribe over XML Date: August 16, 2004 Time: 10am -11am Location: Hawthorne 1S-F40 Host: Stefan Tai

Abstract:
We propose X-ToPSS, a new algorithm for filtering XML documents against large numbers of XPath expressions. This algorithm leverages existing publish/subscribe algorithms by casting the XML filtering problem into a content-based publish/subscribe matching problem. In our algorithm XML documents are translated into sets of attribute-value pairs that are evaluated over conjunctions of Boolean predicates, representing XPath expressions. The strength of this approach is that the predicates resulting from overlapping XPath expressions are stored and processed only once. We experimentally evaluate the performance of our algorithm and validate its scalability to millions of XPath expressions. Our experiments demonstrate trade-offs between our algorithm and alternative XML filtering algorithms. In this talk, we will first present the details of the X-ToPSS matching algorithm and summarize our experimental evaluations. We will then provide a high-level overview of the Toronto Publish/Subscribe System (ToPSS) project, including A-ToPSS (approximate matching), S-ToPSS (semantic matching), L-ToPSS (location-aware matching), M-ToPSS (mobility supporting publish/subscribe) and other current research efforts in this context.

Prof. Dr. eva Kühn (Prof. at TU Vienna and CTO of TECCO AG (university spin-off)) Title: CORSO Space Based Computing: An Enabling Infra-structure for Enterprise GRID Architectures Date: May 17, 2004 Time: 1-2 pm Location: Hawthorne 1S-F40 Host: Sumeer Bhola

Abstract:
CORSO (Coordinated Shared Objects) is an implementation of the Space Based Computing paradigm based on research work carried out at the Vienna University of Technology. The basic differences to Java Spaces are that CORSO implements genuinely distributed spaces. Technically speaking, end-to-end argument replication protocols and a distributed transaction manager have been implemented to administrate the distributed data objects.

With CORSO the software application layer is virtualised: The space realizes a complete separation between application and physical infra-structure (HW, network, topology) thus virtualizing the software application layer. Each participant has a unified and consistent view of the state of the entire distributed system. Based on this information each participant may: optimize its own behavior, manage itself, and join and leave dynamically. There is no need for a central coordinator which enables self-organisation and self-healing of software components.

Language bindings are supported for .NET, Java and C++ on UNIX, Linux, Windows, z/OS and Mobile platforms. The intergration into existing frameworks like J2EE and .NET is possible. As a patented basic infra-structure, CORSO enables the implementation of advanced enterprise GRID architectures with clear responsibilities providing the advantages of mobility, scaleability and fail-over.

Biography:
Dr. eva Kühn holds the titles graduated engineer of computer sciences (Dipl. Ing.), Ph.D. (Dr. techn.) and Venia Docendi (Univ. Doz.) from the University of Technology, Vienna. She received the Heinz-Zemanek research award for the her Ph.D. work on “Multi Database Systems” and a Kurt-Gödel Research Grant from the Austrian Government for a sabbatical at the Indiana Center for Databases at Purdue University, Indiana, USA. Dr. Kühn is employed as a.o. Univ. Prof. at the University of Technology Vienna, Institute of Computer Languages, since 1984.
Dr. Kühn has published more than 40 international articles in the areas of multi-database integration, heterogeneous transaction processing, parallel and distributed programming and coordination languages. Her current teaching topics are parallel processing and coordination tools. She has served as conference
chair, program committee member and local arrangement coordinator for many international conferences, was member of the governing board of the Austrian and European UNIX systems user group, member of the ISO working group for the standardization of Prolog, project coordinator of two research projects
funded by the FWF (Fonds zur Förderung der wissenschaftlichen Forschung) and national technical coordinator of an ESPRIT IV project.
In 1997, an Austrian patent was registered for her research work on a new “coordination system” – the European patent was granted in 2001, the US patent is pending. Based on these patents, Dr. Kühn founded the company tecco Coordination Systems in April 1997. tecco develops and markets the middleware system
CORSO, which is a lean, peer-to-peer based, virtual shared memory system and CORSO based products. Main CORSO application areas are enterprise application integration and replication, collaboration of mobile computers and workflow management. Since 1997, Dr. Kühn is managing director and chief technical officer of tecco.

Baruch Awerbuch, Johns Hopkins University Title: Learning Network paths Adaptive Routing with End-to-End feedback: Distributed Learning and Geometric Approaches Date: January 19, 2004 Time: 10:00 AM - 12:00AM Location: Hawthorne 1S-F40 Host: Alina Beygelzimer

Abstract:
Minimal delay routing is a fundamental task in networks. Since delays depend on the (potentially unpredictable) traffic distribution, online delay optimization can be quite challenging. While the uncertainty about the current network delays may make the current routing choices sub-optimum, the routing algorithm can nevertheless try to learn the traffic patterns and keep adapting its choice of routing paths. Naturally, if the delays of the edge are known, or are changing very slowly, the problem is trivial. The problem is much more challenging if the changes in the delays are occurring at the same time granularity as the routing decisions, and if the delays schedule is adversarial w.r.t. routing decisions. Surprisingly, adversarial model of delays changes appears to be more appropriate than commonly used stochastic models, even for the ``common’’ case where no adversary is actually present in the network.
This adversarial model can be captured quite elegantly in the framework of minimizing regret, which was pioneered in the learning theory community. In essence, the algorithm may make a bunch of mistakes in the transient, but eventually converges to near-optimal path. The attempt is to bound the overall performance degradation of the dynamic online algorithm w.r.t. static optimal path. In this process, the feedback received about the state of the network is crucial. We can distinguish following basic feedback models. In the completely transparent model, we allow feedback on delays of all the edges involved in the current path. In this case learning can be per efficiently using the ``experts'' framework, that will converge to the optimal static path in time proportional to the length of the path. The completely opaque model only allows end-to-end feedback on total end-to-end delay of the path being used. Since Internet is best described as ``opaque'', we focus on this model in the current paper. In this case, learning can be performed using the ``bandit'' framework in ``gambling in rigged casino” result. Applying this result directly, the convergence bound is much weaker; it takes time linear in the number of options, i.e. source-destination paths. However the number of such paths is exponential, leading to a completely unacceptable exponential convergence bound.
We show, for the first time how to accomplish learning in the opaque model with Polynomial time, thus improving the ``gambling in rigged casino” result for network setting. We also extend the results from shortest paths to more general geometric optimization setting. This part of the work involves concept of ``barycentric spanners'' which is a special linear basis of for a compact space of the constraints. We prove bounds on the resulting regret and also provide polynomial time constructions of barycentric spanners for the special case of the shortest paths problem. Joint work with Robert D. Kleinberg

Biography:
Prof. Awerbuch's research is focused on theory online and distributed decisions making and application of this theory to networks, and distributed systems. Prof. Awerbuch is an Associate Editor of the Journal of Algorithms. He has been a co-chair of the 1995 ACM Mobile Conference on Mobile Computing (MobiCom 95), and has been a member of program committees for ACM PODC (1989) and ACM STOC (1990, 1991).
As a consultant to IBM's T.J. Watson Networking Group, he has participated in the design of protocols for IBM's novel high-speed network (PARIS/plaNET), which is a part of the AURORA gigabit testbed. The work of Awerbuch, Cidon, and Kutten (1990) has been recognized with the Outstanding Innovation Award of IBM's Research Division and has been adopted as standard for IBM's wide-area and metropolitan-area networks.

Professor Christof Weinhardt, University of Karlsruhes Title: Market Engineering in the World of Web and Grid Services - Towards a structured procedure Date: Nov 18, 2003 Time: 10:00 am Location: Hawthorne 1S-F40 Host: Heiko Ludwig

Abstract:
“Economists are increasingly being called on to give advice about how to design markets” (Varian 2002). Accordingly market designers have to apply principles of economic analysis to design markets for the private sector (e.g., Ebay) as well as in highly sophisticated domains like financial trading platforms (e.g. Nasdaq, Xetra). Recently, the demand for market-like coordination in decentralized systems has increased - in deregulated industries (e.g. FCC spectrum auctions) as well as in supply networks or in the world of Web or Grid Services. The basic coordination principles are the same for almost all of these scenarios. In both areas (private or corporate), even small design mistakes decide over the success or failure of an electronic marketplace. These mistakes do not only concern the microstructure but just as well the ICT and the business structure. Hence, Market Engineering focuses on the “structured, systematic and theoretically founded procedure of analyzing, designing, introducing and also quality assuring and advancing electronic market platforms as well as their legal framework - regarding their microstructure, ICT infrastructure, and business structure simultaneously”(Weinhardt, Holtmann, Neumann 2003), to guide market engineers in their task and to support them with a preferably integrated tool set. After introducing the Market Engineering as a comprehensive approach the talk will focus on two different perspectives to Web and Grid services: the first is market engineering as a generic methodology for building Web Service based market like co-ordination systems for various areas including Grid computing; the second is related to market engineering as a useful mechanism for an efficient coordination through “trading” Web or Grid Services.

Biography:
Christof Weinhardt is a Full Professor for Information Management and Systems; Department of Economics and Business Engineering, at the Universität Karlsruhe (TH). He is the cofounder of the Institute of “Informationswirtschaft und -management”. Christof Weinhardt is currently Assistant Dean of the Department of Economics and Business Engineering. Besides, he is Chairman of the German Scientific Commission of Information Systems, Founder and Chairman of the research group "e-Commerce" (FG 5.5) in the German Society of Computer Science (GI) and Member of the German Scientific Commissions of Finance & Banking, and Logistics. He is member of the editorial boards of “Wirtschaftsinformatik” and “BIT – Banking and Information Technology”, guest editor and reviewer of a variety of international journals and co-author of two Industry Studies for the German Bundestag. Christof Weinhardt has received his Diploma in Industrial Engineering and Economics and his Ph.D. in Economics from the Universität Karlsruhe (TH). Prior to joining the Universität Karlsruhe (TH) again in 2000 he was Head of the Chair for Quantitative Methods in Business Administration at University of Bielefeld and Head of the Chair for Information Systems at University of Giessen. His research focuses on the “structured, systematic and theoretically founded procedure of analysing, designing and also quality assuring and advancing electronic market platforms (Market Engineering)”. Market engineering is not only the basis for the design process but also in the refinement and quality assurance in the life cycle of electronic markets. In this context, his research concentrates on the branches of financial services, energy industry, logistics, and eLearning. Within recent years, he has initiated numerous innovative research projects in co-operation with corporate partners and government e.g. on Agent-based Markets, Corporate (Financial) Portals, eBrokerage, eExchanges, eLearning, eNegotiations and Meta-Mediation.

Alan Fekete, University of Sidney Title: Consistency for virtual applications in a service-oriented architecture Date: Nov 7, 2003 Time: 10:00 am - 11:30 am Location: Hawthorne 1S-F40 Host: Isabelle Rouvellou

Abstract:
Virtual applications are being increasingly built by loosely coupling otherwise independent and autonomous computing systems which each export services to the outside world. Executing a virtual application involves long-running stateful peer-to-peer interaction between parties that are not within a single trust domain. Applications built this way suffer from potentially serious information consistency problems, as the transaction-based techniques used within organisations are no longer appropriate. A number of standards are being proposed to deal with aspects of this issue: for example BPEL4WS allows an orchestration to describe compensation-based failure handling for one party in an interaction, and WS-Transaction allows parties to coordinate the termination of activities.

A research project at the CSIRO and the University of Sydney is exploring ways to enable architects and developers to be able to systematically design and implement loosely coupled distributed systems thorough a service-oriented architecture, particularly for B2B (Business-to-Business) and EAI (Enterprise Application Integration) applications. As a first step, we have evaluated the strengths and limitations of the existing approaches to consistency maintainance in such systems. A substantial e-procurement scenario is used as a test-bed. The scenario involves long-running interactions between a customer, a merchant, a warehouse, shippers, and a bank. We identify (and classify) a variety of different situations that might lead to inconsistencies between the parties. We show how the compensation-based exception handling (typical of BPEL4WS) deals with some of these cases, and we have found other situations which do not fit easily into that model. We similarly have looked at the coordination issues, to see where the WS-Transaction proposals can be used, and where additional support is needed.

This is joint work by Alan Fekete (U of Sydney), and Paul Greenfield, Julian Jang, Dean Kuo and Surya Nepal (CSIRO).

Biography:
Alan Fekete is Associate Professor in the School of Information Technologies at the University of Sydney. He completed a PhD at Harvard, and has held visiting positions at Cornell and MIT. His research has applied "formal method" techniques to reason about key algorithms in distributed systems, such as replica managment, process group membership, and concurrency control. He is also active in the Computer Science Education community.

Armando Fox, Stanford University Title: Crash-Only Software Date: Oct. 22, 2003 Time: 1:30 PM - 3:00 PM Location: Hawthorne GN-F15 Host: Gautam Kar

Abstract:
Crash-only programs crash safely and recover quickly. There is only one way to stop such software -- by crashing it -- and only one way to bring it up -- by initiating recovery. Crash-only systems are built from crash-only components, and the use of transparent component-level retries hides intra-system component crashes from end users. Crash-only is a design paradigm that can be thought of by analogy to transactions: whereas transactions simplify programming by providing an easy-to-understand model for data integrity, crash-only simplifies failure detection and recovery management by providing an easy-to-understand model for recovery from transient failures. In particular, if recovery can be made predictably safe and predictably inexpensive (in terms of performance penalty), fault detection and diagnosis is simplified as well because false positives become less of a problem. This clears the way for the use of powerful yet generic statistical anomaly based failure detection techniques, and for simple models of recovery management. I will talk about our progress on both formalizing the crash-only model and building a crash-only framework for hosting a variety of interactive Internet services, including providing a near-crash-only application server, crash-only state stores, and a statistical-anomaly-based failure detection toolkit. Crash-only software is part of the Stanford/Berkeley Recovery-Oriented Computing (ROC) project, at http://roc.stanford.edu or http://roc.cs.berkeley.edu.

Jon MacLaren, University of Manchester Title: Service Level Agreement based Scheduling Date: 10 October 2003, Friday Time: 10:00 AM - 11:00 Noon Location: Hawthorne 1S-F40 Host: Gautam Kar

Abstract:
Many Grid Computing Use Cases require complex workflows to be mapped onto distributed computational resources. Naturally, consumers of this technology expect assurances regarding the completion time of such jobs. Unfortunately, the scheduling systems controlling the underlying resources do not provide any information on the expected run time of jobs, unless the run time has been fixed by the use of an advance reservation. Unfortunately, advance reservation is restrictive to the scheduler, and harms machine utilisation, which is often linked to the resource owner's revenue. While sufficient, advance reservation is often not required; some bounds on soonest start and latest end time could be sufficient. Based on these observations, a UK project is developing alternative scheduling algorithms.

We will capture the acceptable start/end information as part of a Service Level Agreement, one of which will be negotiated for every job in the system. The scheduling problem now starts to look more like a "traditional scheduling" problem, like timetable optimisation or hotel room booking. Strategies such as overbooking become important, etc. During the talk, I will describe the research problem in more depth, and explain how the project team, combining Grid Scheduling experts, and "traditional scheduling" experts aims to solve it.

Biography:
Jon MacLaren graduated from University of York (UK) with 1st class honours in Maths and Computer Science in 1993. After two years in industry, he returned to academia and gained his MPhil and PhD from the University of Manchester, completing in 2001. Since then, he has worked in the e-Science Team at the University of Manchester. During his time there, Jon has worked on a number of projects, including the EUROGRID project, where he developed a Resource Broker for the UNICORE Grid Middleware. He is currently works on the Market for Computational Services project, again specialising in Resource Brokering.

Ugur Cetintemel, Department of Computer Science, Brown University Title: Aurora: A Processing Engine for Stream-based Applications Date: 9 October 2003, Thursday Time: 11:00 AM - 12:00 Noon Location: Hawthorne GN-K35 Host: Chitra Dorai

Abstract:
There is a host of existing or newly-emerging applications that require sophisticated and timely processing of fast, high-volume data streams - these stream-based applications typically track data from numerous continuous data streams (coming from such sources as sensor networks and stock feeds), filtering them for signs of abnormal activity, and processing them in a timely fashion for purposes of reduction, aggregation, and correlation. Example applications include environmental monitoring, asset tracking, portfolio management, and network monitoring. This talk will describe Aurora, a general purpose stream processing engine that strives to meet the processing and performance requirements of real-time stream-based applications. The talk will cover Aurora's unique features, including a data-flow orientation, a novel set of stream-oriented operators, and Quality-of-Service-driven algorithms for resource allocation and overload management. The talk will also present preliminary work on Aurora, an overlay network of Aurora nodes that achieves high availability and scalability by distributing stream processing across multiple machines. Aurora is currently under development at Brown University, Brandeis University, and M.I.T. Details of the Aurora project can be found at: http://www.cs.brown.edu/research/aurora/

Biography:
Ugur Cetintemel received the PhD degree in computer science from the University of Maryland, College Park in 2001. He is currently an assistant professor at the department of Computer Science, Brown University. His research focuses on the architecture and performance of networked databases and information systems.

Dr. K. Mani Chandy, Simon Ramo Professor of Computer Science, California Institute of Technology Title: Event Driven Architectures for Crisis Management Date: 14 July 2003, Monday Time: 10:30 AM - 12:00 AM Location: Hawthorne 1S- F40 Host: Roman Ginis

Abstract:
Crises happen. No matter how much we plan, there is often an element of the unexpected in crises such as hurricanes, earthquakes, SARS and terrorist attacks, and even supply chain disruptions and exceptions in financial operations. The Caltech Infospheres Project studies information infrastructures for crisis management. A central problem is that of responding appropriately and in a timely fashion when the system enters a critical state. The system state in a crisis is distributed across multiple institutions in a coalition and also spans potential enemies, competitors, and the general public. The system state is truly global and changes rapidly. I'll discuss methods for specifying, detecting and responding to critical states by a distributed coalition of institutions and individuals that cannot share information completely. The software architecture for crisis management is an example of an event-driven architecture in which multiple components in a distributed system assimilate information from diverse streams of events to form different pictures of global states and respond to a variety of conditions. I'll talk about implementation and performance issues using Java and XML. http://www.infospheres.caltech.edu/crisis_web/executive-summary.html

Biography:
K. Mani Chandy has been the Executive Officer of the Computer Science Department, and a professor at Caltech since 1989. Dr. Chandy is a member of the National Academy of Engineering. He received the IEEE Koji Kobayashi Award for Computers and Communication in 1987 and the A.A. Michelson Award from the Computer Measurement Group in 1985. He was a Professor of Computer Science at the University of Texas at Austin from 1970 to 1989. Dr. Chandy does research in distributed computing. He has published extensively on distributed computing, verification of concurrent programs, parallel programming languages and performance models of computing and communication systems. His book with Jayadev Misra on "Parallel Program Design: A Foundation" is the 89th most cited computer science work ever.

Professor Morris Sloman, Imperial College, Department of Computing, London, UK Title: Ponder: A Language for Specifying Security and Management Policy for Large-scale Distributed SystemsDate: 10 April, 2003 Time:  09:00 AM - 11:00 AM   Location: Hawthorne GN-F15 Host: Alexander Keller

Abstract:
Security policies define who can access resources, what operations they can perform and when they are permitted access. In order to cater for complex inter-organisational systems it must be possible to group objects to which policies apply and to specify policies in terms of roles related to organisational positions. Security management requires policies defining the actions to be performed when security violations occur. In Large-scale Pervasive systems, there are issues of trust which have to be taken into consideration for defining what resources or services can be accessed. Obligation policies or event-condition-action rules can be used for specifying policies relating to adaptibility to cater for user mobility, current context and for self-organising (autonomic) network and application systems. This seminar describes Ponder - a declarative, object-oriented language for specifying policies for security and management. As well as supporting basic authorisation and obligation policies, it includes concepts for structuring policies into configurations of roles to reflect rights and duties for automated agent or human role relationships. The talk will also briefly cover some of the new research activities starting at Imperial College relating to Ubiquitous Computing for Healthcare in the Community (http://www.ubicare.org)

Biography:
Professor Morris Sloman leads the Distributed Software Engineering Group in the Department of Computing, Imperial College London. He has managed a number of research projects funded by the UK Engineering and Physical Science Research Council (EPSRC), European Union and various industries on management, security and design of distributed systems, multimedia systems and mobility. He has many journal and conference publications on these topics and is editor of a reference book on Management of Network and Distributed Systems published by Addison Wesley. He is a member of the editorial board of the Journal of Network and Systems Management. Professor Sloman was a member of the 2001 Research Assessment Panel for Computing Departments in the UK and a panel assessing Irish ICT research in 2002. He is on the UK Engineering and Physical Science Research Council ICT Strategic Advisory Team and a member of the executive panel of the UK Computing Research committee. He was program co-chair for Mobile Data Management (MDM) conference in 2003 and is on the steering committee for the workshops on Policies for Distributed Systems and Networks.

Prof. Joerg Kienzle, McGill University, Canada Title: Separating Concurrency and Failure Concerns using Aspect-Oriented Programming Techniques Date: 28 March, 2003  Time: 10:00 AM - 11:00 AM Location: Hawthorne 1S-F40 Host: Stefan Tai

Abstract:
Concurrency and failures are fundamental problems in distributed computing. One likes to think that the mechanisms needed to address these problems can be separated from the rest of the distributed application and reused in several different contexts: in modern words, these mechanisms could be "aspectized". Does this however make sense? The talk will present an experience that uses transactions as a mechanism to handle concurrency and failures, and AspectJ as representative of aspect-oriented programming environments. We tried to achieve separation and reuse at 3 different levels: aspectizing transactions as a whole, aspectizing the interface to the transaction support, and aspectizing transaction mechanisms. We demonstrate that aspectizing transactions as a whole is not achievable, argue that aspectizing the interface to the transaction support is potentially dangerous, and finally show that AOP can be applied in an elegant way to achieve aspectization of transaction mechanisms.

Biography:
Jörg Kienzle is assistant professor in computer science at McGill University, Montreal, Canada, where he is leading the software engineering laboratory. He holds a Ph.D. and engineering diploma from the Swiss Federal Institute of Technology (EPFL) in Lausanne. His current research interests include fault tolerance, software development methods, and aspect-orientation.

Alain Mayer, CTO, CenterRun, Inc.  Title: Application Aware Management of Internet Data Centers OR How to wean your Operations Staff from using obscure Scripts Location: Hawthorne 1S-E53/F53 Date:11 November, 2002 Time:10:30 AM - 12:00 PM Host: Alexander Keller

Abstract:
We have built a comprehensive solution to address the management aspects of deployment and analysis of web-based applications in Internet Data Centers. Our work was motivated by the high total cost of ownership of operating such centers, largely due to the variety of applications and their distinctive management requirements. We have chosen an approach that encapsulates application specific knowledge (is application aware) and deployed it in a number of corporate Internet Data Centers. Operations staff found substantial cost reduction in managing applications using our approach.

Biography:
Alain Mayer has a PhD in Computer Science from Columbia University. After over 4 years at Bell Labs, Lucent Technologies, he has joined the start-up world. He is currently Chief Technology Officer at CenterRun, Inc. (www.centerrun.com), where he guides the research and development of data center management software.

Jeff Chase and Amin Vahdat, Department of Computer Science Duke University Title: Research Challenges for Opus: An Overlay Peer Utility Service Date: October 18th, 2002 Time: 10:30-11:30 Location: Hawthorne 1S-F40 Host: Giovanni Pacifici

Abstract:
As consumer computing devices become more diverse and specialized, the user experience of dependable computing increasingly relies on the network services that tie them together. The goal of the Opus project is to enable automated management of a large-scale wide-area server infrastructure to deliver target levels of performance and availability for Web services and other network services under dynamic conditions. A key premise of Opus is that a shared pool of servers and storage resources under coordinated, dynamic control can yield the best service performance and dependability at the lowest cost. The Opus approach applies to ``grids'' pooling server resources at many autonomous sites, or to distributed computing utilities under common administration.
This talk outlines some principles of the Opus approach and current research work supporting the Opus vision. It focuses on adaptive provisioning within an Opus site, service placement across multiple Opus sites, and evaluation of Opus prototypes using ModelNet, a system for scalable Internet emulation.

Valerie Issarny, INRIA, France Title: Software Architectures of Dependable Systems: From Closed to Open Systems Date: Thursday, September 19, 2002 Time: 9:30AM-10:30AM Location: Yorktown, 20-059 Host: Kamal Bhattacharya

Abstract:
Work in the software architecture domain primarily focuses on the standard (as opposed to exceptional) behavior of the software system. However, it is crucial from the perspective of software system robustness to also account for failure occurrences. In this talk, I will give an overview of the work done within the ARLES research team at INRIA, towards assisting architecting of dependable distributed systems, which relates to providing systematic aid in the development of middleware architectures for dependable systems, and supporting architecture-based exception handling. I will then discuss our ongoing research work aimed at addressing dependability requirements of open, distributed systems, considering more specifically systems based on the Web Service Architecture.

Biography:
Valerie Issarny holds a "Directeur de Recherche" position at INRIA where she is leading the ARLES research team at the Rocquencourt research unit. The ARLES research team investigates solutions to architecture-based development of distributed systems, in order to support the composition of systems offering quality properties to users. In that context, the ARLES teams studies the definition of languages, methods, tools and supporting middleware infrastructures, which ease the development of distributed systems by offering solutions to automated design, analysis and construction of systems. Current research work within the ARLES team is more specifically centered around the development of distributed systems enabling ambient intelligence applications. Valerie Issarny is further currently vice-chair of the ACM SIGOPS.

Erik Hendriks, Sung-Eun Choi Los Alamos National Labs Title: "Life with Ed: A Case Study of a LinuxBIOS/BProc Cluster" Date: Friday, June 21 Time: 1:00 TO 2:30 Location: 4S-K21 Host: Ron Mraz

Abstract:
In this talk, we describe experiences with our 127-node/161-processor Alpha cluster testbed, Ed. Ed is unique for two distinct reasons. First, we have replaced the standard BIOS on the cluster nodes with the LinuxBIOS which loads Linux directly from non-volatile memory (Flash RAM). Second, the operating system provides a single-system image of the entire cluster, much like a traditional supercomputer. We will discuss the advantages of such a cluster, including time to boot (101 seconds for 100 nodes), upgrade (same as time to boot), and start processes (2.4 seconds for 15,000 processes). Additionally, we have discovered that certain predictions about the nature of terascale clusters, such as the need for hierarchical structure, are false. Finally, we argue that to achieve true scalability, terascale clusters must be built in the way of Ed. For additional information see, http://www.acl.lanl.gov/cluster

Biography:
Erik Hendriks is a staff member in the Cluster Research Lab at Los Alamos National Laboratory. Erik was part of the original Beowulf team at NASA Goddard, where he began his work on BProc, Linux Single System Image for cluster management. Erik's research interests includes operating systems, networks, and parallel and distributed systems.

Prof. Priya Narasimhan, CMU Dependability for Web Services Date: Wednesday, June 12, 2002 Time: 11:00 am - 12:30 pm Location: Hawthorne, 1S-F40 Host: Isabelle Rouvellou

Abstract:
Web Services represent the next generation of enterprise infrastructures for the exchange of information and services across the Internet. The power of Web Services lies in their ability to abstract away platform- and language-specific details of service implementations, and to expose service interfaces that allow for interoperable dynamic registration, publishing and discovery of services across the Internet. With this rich set of capabilities, along with protocols like Simple Object Access Protocol (SOAP), Web Services promise to be the "glue" for business-to-business enterprise interactions of the future.
The Web Services of today do not as yet incorporate quality-of-service (QoS) system properties, or "-ilities", such as reliability and security. Furthermore, the Internet, in its role as the underlying communication medium for Web Service interactions, is inherently unreliable, and has been known to exhibit system failures due to crashes and security attacks. With downtime being prohibitive in terms of cost and loss of reputation, and with faults and security breaches becoming increasingly unacceptable, the widespread deployment of Web Services, in the absence of security and reliability guarantees, will not be realized.
Building systems that must provide both reliability and security simultaneously to the application poses distinct technical challenges. In most cases, from the perspectives of both the application and the infrastructure, security and reliability do not always make a good "marriage". Furthermore, the intrinsically dynamic element (e.g., run-time discovery and late binding) of Web Services make the problems of dependability challenging, and far more interesting than conventional client-server enterprise systems.
In this talk, I will focus on the research issues surrounding quality-of-service for Web Services, with emphasis on developing infrastructures that provide both security and reliability. In particular, I will address some of the trade-offs and compromises that are inevitable when combining reliability and security. The talk will also discuss the difficulty of evaluating the success of such research efforts, given the current lack of metrics or benchmarks for the objective measurement of dependability.

Biography:
Priya Narasimhan is currently a faculty member with the Institute of Software Research International, which is a part of the School of Computer Science at Carnegie Mellon University. She has been involved extensively in building dependable systems, and in incorporating her results into industrial practice through her contributions to the Fault-Tolerant CORBA standard. Her current research interests include all aspects of dependability, including reliability for large-scale enterprise middleware systems, metrics for quantifying dependability, and the composition of reliability with other "-ilities" (security, real-time) in order to build useful systems.

Prof. Zvi Kedem and Prof. Vijay Karamcheti, NYU Title: Computing Communities: Adaptable Distributed Applications Date: Tuesday May 7, 2002 Time: 11:00 am - 12:30 pm Location: Hawthorne, GN-K35 Host: Jakka Sairamesh

Abstract:
Current day distributed applications are required to execute in diverse network environments with widely varying resource and security characteristics, and need to cater to multiple usage scenarios. To avoid having to construct different application configurations for each scenario, one would ideally like to rely upon a software system infrastructure that allows applications to automatically adapt to their execution environments.
Computing Communities is an umbrella project, which has as its goal the seamless aggregation of resources belonging to multiple owners for use by applications belonging to any of them. A core aspect of the project is support for adaptation: how should an application make use of additional resources when they become available or cope with the taking away of some resources. Other aspects of the project include virtualization and resource sharing.
This talk will present an overview of ongoing research activities in the NYU Parallel and Distributed Systems Group, which is investigating different ways of constructing such infrastructures. The talk will focus on three frameworks - Application Tunability, CANS, and Mutable Services - which support adaptation at the level of a single component, at the level of data streams flowing between components, and at the inter-component level respectively.
Using application studies spanning image visualization, web access with weak devices over low-bandwidth networks, and clients accessing wide-area network services, we discuss how application flexibility is exposed in each of the frameworks, and describe the system support required to exploit this flexibility for adaptation purposes.

Prof. Maurice Herlihy, Brown University, Title: Distributed Queuing Date: Friday May 3, 2002 Time: 10:30 am - 12:00pm Location: 20-001 Host: Maged Michael

Abstract:
Distributed Queuing is a distributed coordination problem that lies at the heart of a variety of distributed applications, such as managing concurrent access to mobile data objects, and scalable ordered multicast. The Arrow protocol is a simple distributed queuing protocol based on path reversal. In this talk, we discuss our experience using Arrow as the basis for tracking and synchronizing access to mobile shared objects, including an explanation of apparently inexplicable performance results. Time permitting, we will discuss an ongoing project to use Arrow as the communication backbone of a scalable peer-to-peer multiplayer game.

Biography:
Maurice Herlihy is a Professor of Computer Science at Brown University. He has an A.B. in Mathematics from Harvard University and a Ph.D. in Computer Science from MIT. He has been a faculty member of the Computer Science Department at Carnegie Mellon University and a member of research staff at Digital Equipment Corporation's Cambridge Research Lab. His research interests concern practical and theoretical aspects of distributed computing.

Alastair Green, CEO, Choreology Ltd. Co-author of OASIS Business Transaction Protocol draft specification. Choreology is a VC-funded start-up, founded in January 2001. It is headquartered in London. Title: Coordinating Loosely-coupled Actions using OASIS Business Transaction Protocol (BTP) Date & Place: April 16, 2002 (Tuesday), 10:00AM-12:00PM, Hawthorne GN-F15 Hosts: Thomas Mikalsen & Stefan Tai

Abstract:
Loosely-coupled distributed systems are based on business-driven contracts which specify the content, format, legitimate order of, and implied reactions to, inter-application messages. Such systems eschew shared knowledge of implementation and seek to minimize interlocked behaviours. The need for such systems is amplifed by the chaotic variety of systems and inter-connections deployed in user organizations. Different styles and qualities of message transmission, delivery and processing (oneway, request-response; attempted vs. guaranteed delivery; exactly-once delivery/processing) are all legitimately used to build systems of this type.

Loosely-coupled systems paradoxically also require consistent, assured outcomes to be achieved, where the reactions to multiple messages received by multiple parties can be gathered into synchronization points, irrespective of the style or quality of message delivery. The provision of assured common outcomes in this environment requires a process-oriented, interface-access stimulation of opaque state changes which are contract-observant. Operations must be defined in each of the parties to a coordinated outcome (or "business transaction") to allow provisional or tentative effects to be either finalized or countered in accordance with the will of a commanding application. At the same time the autonomy of independent departments or companies must be preserved, by allowing withdrawal at any time from such coordinations. Business transactions that supply these features are critical to simplifying and improving the reliablity and fault-tolerance of business processes.

OASIS Business Transaction Protocol is a standard initiative supported by Oracle, HP, Sun, BEA Systems, IONA and others. Choreology Ltd has played a leading role in proposing and defining the technical direction of the BTP specification. In particular Choreology introduced a key novel feature, the "cohesion", which allows flexible and dynamic views of the vitality and viability of the participation of parties in a coordinated business transaction. Other features of BTP, such as its use of a two-phase outcome protocol (reflecting the view that the state-undo rollbacks used in ACID transactions are a special case of compensation) and its ability to concurrently exploit multiple underlying carrier protocols, underpin implementation robustness and trustworthiness, and address key business needs in the task of integrating disparate systems.

BTP in its first incarnation provides a defined binding to SOAP-over-HTTP. Choreology has also commenced independent work on a WSDL binding for BTP, and a preliminary view of this work will be presented.

Roman Vitenberg, Technion, Haifa, Israel Title: Cascade: A Caching Service for Distributed CORBA Objects Date: 17 September, 2001 10:30 AM - 12:00 PM Location: Hawthorne 1S-F40 Host: Alexander Keller

Abstract:
Caching systems and architectures have proven exceptionally effective in the large scale World Wide Web environment, and Web caching techniques, tools and infrastructures have been becoming increasingly mature during the recent decade. At the same time, the more general problem of caching generic distributed objects remains mainly unresolved. In particular, very few caching systems have been developed for object oriented middlewares like CORBA, RMI, or COM+. This is believed to be one of the major reasons for hindering development of truly large scale distributed applications based on such middlewares. In this talk, I will present Cascade, a caching service for distributed CORBA objects, aimed at increasing the availability, locality, predictability and responsiveness of object accesses. CASCADE is unique in supporting dynamic migration and deployment of active objects (that include both data and code), and by employing a dynamically built hierarchy, which makes it suitable for scalable Internet services. Among the other main benefits of CASCADE are transparency for applications and powerful and flexible consistency maintenance.

Biography:
Roman Vitenberg is working toward his PhD in computer science at Technion, Haifa, Israel. He received his BS in computer science and in mathematics and his MS in computer science, both with distinction, from the Hebrew University, Jerusalem, Israel. His research interests include distributed systems, distributed objects and object oriented middleware, scalable computing, fault tolerance and high availability systems. He is a student member of the ACM.

Michael Macedonia, STRICOM, Title: "Real World VR" Date: Friday August 3rd, 2001 Time: 10:00 to 11:30 Location: GN-F15, Hawthorne Host: Rob Strom

Abstract:
Michael Macedonia will address the current state of virtual reality and how virtual environment applications fit in our society's future. The promises of cyberspace are aging rapidly, and there is no clear consensus on the state of VR. However, the ubiquity of high-bandwidth network connections and powerful desktop graphics has already lead to many novel uses of VR, particularly in support of training, entertainment and medicine.

Biography:
As the Chief Scientist and Technical Director for STRICOM, Dr. Macedonia is responsible for the planning and development of a comprehensive research program that provides the full spectrum of commercial and defense technology for a $ 1.1 billion per year enterprise. As Technical Director, he represents STRICOM to the Army and Department of Defense Science and Technology community. Macedonia also chairs the STRICOM Technical Council and provides outreach to the commercial simulation industry, academia and research institutions. He oversees the Technical Advisory Board composed of leading CEO's and scientists from industry and academia including Kodak, Department of Energy, and the Walt Disney Company. A graduate of West Point, Macedonia served as an infantry officer in a variety of command and staff positions in the United States and overseas assignments including Germany and the Middle East. He also served as a project manager for automated electronic warfare systems development and successfully managed research, development and acquisition for communication training systems, voice recognition, and communication analysis software and hardware. He is a veteran of Desert Storm. Following his military service, Macedonia became the Vice-president of the non-profit Fraunhofer Center for Research in Computer Graphics, Inc. (CRCG) in Providence, Rhode Island. Under his leadership at CRCG, the center grew to over 40 researchers and staff, developed unique programs with Brown University and the Rhode Island School of Design in the areas of computer graphics and design. CRCG became known worldwide for innovative research into virtual reality, networking, and tele-medicine. Macedonia then joined the Institute for Defense Analyses in Alexandria, Virginia as a Research Staff Member for Modeling and Simulation. Macedonia has a Ph.D. in Computer Science and a M.S. in Telecommunications. His research in networked virtual reality is widely cited in the technical literature. He is the author of numerous publications and worked extensively with the multicast networking community. He also contributed to the National Academy of Sciences report entitled "Virtual Reality: Scientific and Technological Challenges," detailing the further networking and communications research needed to continue the development of virtual reality systems. He is a member of the editorial board for IEEE Computer; the second most read technical publication in the world, and the department editor for "Entertainment Computing". He is the co-editor of "Projects in Virtual Reality, for IEEE Computer Graphics and Applications. He is the editor for the Modeling and Simulation Technology Trends column for Simulation Magazine. Macedonia is also on the editorial board for the MIT Journal Presence, the journal for virtual reality research. He chairs the IEEE Virtual Reality Conference, to be held in Orlando in March 2002. He has is a member of the IEEE Computer Society Technical Committee on Visualization and Graphics. He has been quoted in numerous publications including the New York Times, Los Angeles Times, National Defense, and the National Journal. He also been an invited speaker at numerous forums including the Aspen Institute and ACM SIGGRAPH and a lecturer at a wide variety of academic forums. Macedonia is also the Chair of the Industrial Advisory Board for the University of Central Florida School of Electrical Engineering and Computer Science and is on the Engineering Advisory Council for the University of Florida College of Engineering. He also has served as an associate member for the State University System of Florida Digital Media Education Coordination Group and as a member of e-Entertainment subcommittee for the Florida Information Technology Task Force. Dr. Macedonia is also on the NTSA Certification Board for M&S Professional Certification.

Vijay Karamcheti and his student Tao Zhao Title: Expression and Fine-grained Enforcement of Resource Sharing Agreements in Distributed Systemson Date: Thursday, June 28, 2001 Time: 10:00--11:00 Location: 2N-F28 Host: Mike Spreitzer

Abstract:
Advances in computing and networking technology and an explosion in network-accessible services has resulted in a growing number of distributed systems getting constructed out of resources contributed by multiple sources. Use of such resources is typically governed by sharing agreements between owning principals, which limit both who can access a resource and in what quantity.
Such systems raise several new resource management challenges particularly with respect to the expression and enforcement of sharing agreements. Recognizing the need for generic infrastructure-level support for agreements, we present an approach that builds on the concepts of tickets and currencies to express sharing agreements involving diverse resources in an abstract, dynamic, and uniform fashion. We also formulate the allocation problem of enforcing these agreements as a linear-programming model, automatically factoring the transitive availability of resources via chained agreements. Using this model, and combining it with efficient time-window based distributed queuing algorithms, we describe an architecture for the fine-grained distributed enforcement of resource sharing agreements.
A case study simulating resource sharing among ISP-level web proxies shows the benefits of enforcing transitive agreements: worst-case waiting times of clients accessing these proxies improves by up to two orders of magnitude. We have also implemented a prototype of the proposed architecture in which coordinating HTTP redirector nodes schedule requests from distributed clients to web-based servers. Measurements verify that our approach is low-overhead and effective: different client groups receive service commensurate with their agreements.

Gidon Gershinsky, IBM Haifa Lab Title: "Reliable Multicast for Bulk Data and Message Stream Delivery", Date: February 20, 2001, Time: 1PM Location: the Hawthorne Auditorium, GN-F15 Host: Rob Strom

Abstract:
The IP multicast, introduced over ten years ago, is a standard IP extension by now. Still, despite its promise of extremely scalable data delivery, the use of multicast was very limited both at corporate networks and in the general Internet, mainly due to the complexity of traffic administration and to the lack of deployment at wider area networks. This situation started to change recently when the ever increasing necessity to deliver large information amounts simultaneously to numerous recipients has made the multicast the only possible and unavoidable transmission mechanism. A number of companies specializing in content delivery and message queueing (like Tibco, Fiorano, Digital Fountain and others) are now putting multicast at the core of their business. We in IBM Haifa Research Lab have developed a reliable delivery system based on the standard IP multicast. In addition to providing reliability layer to the inherently unreliable IP multicast, our system incorporates cataloging (inventory announcement) mechanism that allows receivers to learn about data items available, choose one or more and request the transmission at the time convenient to them. The delivery is optimized for two data types: bulk information and message streams. In this talk, I will give a brief introduction to the field of reliable multicasting and describe the techniques we use for inventory announcement and optimal delivery of bulk and streaming data.