Go to Dana Petcu's homepage
GRAI: GRID COMPUTING AND ARTIFICIAL INTELLIGENCE
Five Postdoc positions are available at Western University of Timisoara for 18 months: January 2007-July 2008.
Four PostDoc positions still available !!!!
Young researchers (bellow 35 years) with PhD Degree who are interested in a full research position and in one of the following subjects are invited to submit a request.
Applications should be send asap to Dana Petcu (petcu@info.uvt.ro).
Agent based Grid clients are able to interact with the user and to drive him throughout the exploitation of Grid resources. They can provide facilities for dynamic and automatic upgrade of software of Grid nodes and of user devices. They can move to the client device and automatically analyze the available resource, in order to configure the device and to optimize the services exploitation. On server side the framework can use agent technology to share applications which can run on each node of the grid where the data reside or where the available resources fit the target requirements. Proactivity allows the adoption of new computing paradigms especially suited for narrow band network, or where the connection is unstable or too much expensive
Integrating Grid and software agents technologies bring new opportunities and helps in defining and developing new services. Once this integration is realized, software agent concepts and technologies will help to enable new and advanced operational and usage modalities of Grid. Software agents can act, as Grid users, on behalf of their owners, according to a delegation model, and accomplish a specific task by the exploitation of Grid services and resources. On the other hand agents can provide Grid services on their own by composing existing services or implementing new ones. Autonomy and mobility can be used to explore the Grid and to distribute the computation where data and resource are available.
A first task will be the development of a testbed based on agents that is similar with MAGDA (Mobile Agents based Grid Architecture [1]) in order to integrate the provided protocols, services and application programming interfaces at the different layers of the Grid infrastructure. Since MAGDA was originally implemented using Aglets, which are a dying-off agent platform, it will be reimplemented using JADE, the state of the art agent platform of today.
In order to achieve this goal, it is needed the intensive collaboration with the team that has created MAGDA. Collaboration relationships with the Napoli team have been established prior to this project. Moreover, the previous experience existig in the team that proposes the current project in the field on the multi-agent systems is benefic for the succesfull execution of the project.
A functional model and a component based architecture are necessary to be defined for the new platform. Those will beprovide to the project both by methodological and experimental contribution, aiming at the definition of new techniques and at the application of new technological solutions to realize Grid services based on the exploitation of stationary and mobile agents.
Possibility to build and export new innovative agent based GOTI services will be also investigated. For example software agents could provide added value services by composing the basic ones or by implementing new ones. Hence, the software agent paradigm will be exploited both for services management and for service provision. In the first case agents will exploit the system infrastructure to perform efficient service discovery and services composition to optimize the execution of their jobs. Furthermore, agents will be used to manage prototype economical model of resource sharing, involving price negotiations for service delivery. Finally, agents will be used to optimize load balancing for a given task as well as across the grid and to optimize overall system throughput. Here, existing techniques for scheduling and load balancing of grid applications will be utilized and enhanced using Agent technology. In the second case agents will implement services themselves and will be able to be invoked as any Grid service.
Simple prototypal agent based applications will be implemented to prove how the agents can exploit the Grid services and resource to accomplish a task on behalf of the user. Properly conceptualized stationary and mobile agents will use basic services of the Grid infrastructure to get system information, to collect the necessary resource and allocate them before they migrate and perform their computations. For instance, mobile agents can migrate from one host to another to optimize the system performance and utilization at server side, or to reconfigure the user device and interact with the user at the client side.
A collection of benchmarks will be developed to test the available resources and measure the achievable performance. At first, programming and execution environments will be provided as basic features of nodes (PVM, MPI, Mobile Agent server) together with services such as resource brokering, gatekeeper, Grid enabled libraries. An agent platform will be instantiated and combined with basic services to submit the execution from remote of simple agent based application.
[1] R.Aversa, B. Di Martino, N. Mazzocca, S.Venticinque, "MAGDA: a software environment for MobileAGent based Distributed Applications", Proc. of .11th Int. Workshop on Parallel and Distributed Processing, 2003, Genova, Italy.
Computational science researchers often gather multichannel and multimodal data from real-time collection instruments, access large distributed databases, and rely on sophisticated simulation and visualization systems for exploring large-scale, complex, multidimensional systems. Managing such large-scale computations requires new algorithmic approaches, powerful and distributed computing resources and efficient, scalable, and transparent software that frees the user to engage the complexity of the problem rather than of the tools themselves. Such computational application software does not currently exist.
The project will allow to test new tools and methods developed for datamining in very huge distributed databases, especially databases composed of unstructured data. Existing parallel algorithms have reached their limits: they are not scalable to massive databases and they are inappropriate for distributed context. High performance datamining requires to invent new approaches based on heuristics methods, suited to solve new kinds of problems as reduced support or multicritera search, and designed for efficient distributed execution on Grids.
Data partitioning will be studied based on distributed methods and will help the execution of new heuristics allowing both the decrease of the problem complexity and the exploitation of the distribution.
The project will allow to test on Grids new heuristics developed for datamining, to create new tools allowing better monitoring and testing on the Grid and to design new tools which assure increased transparencies for the programmer.
A starting point for the research will be the results of the FP6-European project DataminingGrid [1]. The Data Mining Tools and Services for Grid Computing Environments (DataMiningGrid, 2004-2006) is developing tools and services for deploying data mining applications on the Grid. Currently there exists no coherent framework for developing and deploying data mining applications on the grid. The DataMiningGrid project addreses this gap by developing generic and sector-independent data-mining tools and services for the grid: Grid-enabled data-mining data interfaces and services, Grid-enabled data-mining workflow management tools, and Grid-enabled text mining interfaces and components.
The cooperation with the developers of CosmoGIS [2] that aims at enabling the integration of distributed Earth remote sensing data and providing corresponding services to the users. Contacts with the Kiev team has been establish previously to this project.
Most of the data that will be studied will be taken from the field of satellite images and those of interes for meteorology, the activity from this research drection being related with the one of the group in the frame of the running project CEEX MedioGrid [3].
[1] DataminingGrid, http://www.dataminingrid.org
[2] CosmoGIS, http://www.cosmogis.org.ua
[3] MedioGrid, http://mediogrid.uttcluj.ro
A first aim is to develop Grid-based implementations of multi-objective evolutionary algorithms. The difficulty consists in adapting the parallelization strategies to the particularities of a Grid infrastructure (large computational resources, communication topology, dynamic character and heterogeneity of resources). To benefit from the computational power of the Grid the distributed version of an evolutionary algorithm should be adequately designed taking into account specific aspects like dynamic populations, asynchronous communication, tree topology communication.
The basic steps to be followed are: (i) selecting multi-objective optimization evolutionary algorithms (MOEAs) and develop strategies for their Grid implementation; (ii) selecting some challenging multi-objective optimization problems; (iii) testing the Grid implementation for the selected problems.
The support from the team that created ParadisEO-CMW [1] is requested. The cooperation relationships with the Lille team have been established prior to this project. The second aim is the development, implementation and testing of:
1. evolutionary algorithms with adaptivity to the dynamic environments
2. methods inspired by the behaviour of natural systems (ant colonies, bird swarm etc)in order to solve dynamical optimization problems to be apply in resource scheduling problems in Grid applications management, and in the data analysis coming from dynamical databases.
Note that the group that proposes the project has previous experience in:
- development of methods to adapt the parameters of the evolutionary algorithms
- theoretical analysis of the population diversity in the case of the evolutionary algorithms
- development of several variants of the clustering algorithms on non-supervized type based on the model of ant colonies.
The activity will be correlated with that planned in the frame of the project CEEX GridMOSI (2005-2008): Virtual Organisation for high performance modeling, simulation and optimization – the tasks that for the local group are concerning optimization methods on Grid architecture.
[1] S. Cahon, N. Melab, E-G Talbi. An Enabling Framework for Parallel Optimization on the Computational Grid. In the Proc. of the 5th IEEE/ACM Intl. Symp. on Cluster Computing and the Grid (CCGRID 2005), Cardiff, UK, 2005.
The implementation of the exploration model requires complex and significant case studies, to allow the identification and usage of relevant knowledge schemes and the mechanism that allows their application. The case study of systematic exploration of Groebner bases and related theory, will represent the object of research in the first phase of this project. It represents a continuation of the first steps carried out in this direction in the frame of the European SysteMaThEx project (2005-2007), currently executed by the team. In particular, the following will be pursued:
- synthesis of Groebner bases algorithms, that should be more efficient that the one synthesized at the moment, by adding efficiency specifications to the algorithm scheme used (CPC - Critical Pairs/Completion ).
- the generation of a general (axiomatic) description of domains in which the CPC scheme can be applied, by applying the exploration methodology based on schemes; this will allow the synthesis of an abstract class of algorithms (in this CPC domain), that can then be instantiated in any domains where the CPC axioms hold; in particular, algorithms like Knoth Bendix (term rewriting) or resolution (clause set solving) could become simple problems of domain classification.
The files containing the theory generated following the exploration process (expressed in the natural stil in the Theorema language), together with the system, can be used, in Grid and gent context as a mathematical service, that offers mathematical knowledge, or solution to problem like “find an algorithm to determine if an element is a member of the ideal generated by a polynomial set”.
Therefore, in a second phase of the project the activity will be cencentrate to realize the vision of the SCIEnce project: symbolic computation on Grid architecture by using the standards developed in the frame of MKM for mathematical object representation and the specific techniques of distributed applications. A number of major obstacles remain to be overcome if effective symbolic Grid applications are to be developed. Amongst the most important are mechanisms for adapting to dynamic changes in either computations or systems. This is especially important for symbolic computations, which may be highly irregular in terms of data and general computational demands. The starting point in what concern the resource allocation will be MathBroker [1] developed accordind the specifications of the MONET project. The key new and unsolved research issues that must be tackled by the project are: dynamic load management for irregular computations; cost estimations for complex Symbolic Computations; solving real world applications involving symbolic computations too complex to be solved on a single computers or a cluster; foresight of system and application needs; construction of contract resource negotiators and resource brokers for complex Symbolic Computations.
The cooperation with the team that has developed the MKM concept and MathBroker is crucial for the project. Collaboration relationships with the Linz team were established prior to the project.
[1] W. Schreiner, Brokering mathematical services in the global network. In Procs. IIWAS 2003, G. Kotsis, S. Bressan and I. Ibrahim, Austrian Computer Society, vol. 170 (2003)
Workflow design determines how workflow components can be defined and composed. Workflow scheduling focuses on mapping and managing the execution of workflow tasks on shared resources that are not directly under the control of workflow systems. In Grid environments, workflow execution failures can occur for various reasons such as network failure, overloaded resource conditions, or non-availability of required software components. Thus, Grid workflow management systems should be able to identify and handle failures and support reliable execution in the presence of concurrency and failures. For Grid workflow applications, the input files of tasks need to be staged to a remote site before processing tasks. Similarly, output files may be required by their children tasks which are processed on other resources. Therefore, intermediate data has to be staged out to corresponding Grid sites. Some systems require users to manage intermediate data transfer in the workflow specification, while some systems provide atomated mechanisms to transfer intermediate data.
This research direction aims to design and implement an automated workflow management system for Grids that allows the above required facilities.
A starting point for the research will be the results of the European project K-Wf-Grid [1]. The Knowledge-based Workflow System for Grid Applications (K-Wf-Grid, 2005-2007) introduces several new ideas and innovations in the area of Grid workflow application development, application of knowledge-based technology and performance monitoring services. The main innovation of the K-WfGrid workflow management technology includes a novel approach of supporting knowledge-based workflow orchestration and Grid Service selection by means of an expert system.
Note the previous experience of the local group in the frame of the European project VISP (Virtual Internet Service Provider [2], 2005-2008), in what concerns the workflow management systems. The activity from the current project will be correlated with the local team work in the frame of VISP project.
[1] K-Wf-Grid, http://www.kwfgrid.net
[2] VISP, http://www.visp-project.org
Last modification: 03.09.2006