Computer Science Department - University of Piemonte Orientale  
ShareGrid applications
  • Image processing
    • Pov-Ray creates three-dimensional, photo-realistic images using a rendering technique called ray-tracing.
      • The image to elaborare is split in many tiles.
      • The elaboration of each tile corresponds to a task.

  • Simulation of economic systems
    • Parei is an agent based simulation modelling the over 500k production units composing the economy of Piedmont. Parei is used for the understanding and the esteem of economic dynamics, particularly when they are difficult to be observed in vivo, and for the evaluation, in vitro, of economic effects of proposed public policies.
      Every set of parameters of the model corresponds to a task and to a different scenario that must be simulated. The parallel exploration of solution space allowed by several different tasks makes possible to neutralise deterministic effects in random number generation and to analyse the solution space (robustness analysis).

  • Simulation of molecular systems
    • The simulation of the dynamic behaviour of molecular systems is a computationally intensive task, especially when biological macromolecules are concerned, and solvent molecules are modeled explicitly. The availability of large computational resources is thus mandatory to accomplish reliable esteems of the free energy of binding between a ligand and its target. Techniques like thermodynamic integration (TI) require running several, independent, molecular dynamics (MD) simulations, and are particularly fit for a distributed computing scenario like the one provided by ShareGrid, since communication between tasks is not required. The availability of a cluster of independent CPUs also allows simultaneous virtual screening of a series of potential drug candidates by means of molecular docking and subsequent MD refinement, thus making in silico methods competitive with traditional in vitro wet-lab screening.

  • Distributed rendering
    • Distributed rendering: Scene rendering -- where a set of scenes of a given movie must be rendered via software in order to adds in static and dynamic features (like bitmap or procedural textures, lights, bump mapping, etc.), and to stitch them together for making the final animation ­ is a typical compute-intensive activity. The inherent nature of scene rendering, where different frames belonging to the same animation can be rendered independently from each other, makes it particularly suited to distributed processing. A distributed version of scene rendering can indeed be obtained by processing each frame independently from the other ones, and then merging together them to build the complete movie. Distributed rendering drastically reduces the rendering time (and the production time too) and the investment costs needed for buying and maintaining personal computing resources. For ShareGrid, we developed a Bag-of-Tasks version of blender (an open source rendering engine), in which bags correspond to scenes, and tasks of a bag to the different frames of the same scene. This application is being used by professionals working in the animation field to accomplish their activities.

  • Simulation of scheduling algorithms for distributed systems
    • Discrete-event simulation is often used in Computer Science to study the behavior of certain systems before actually implementing them. Job scheduling for distributed systems is one of the research areas in which discrete-event simulation is often the tool-of-choice. The study of the behavior of a given scheduling algorithm for different scenarios, or the comparison of different algorithms for the same set of scenarios, can be naturally performed by simultaneously executing many independent simulations in parallel. A Bag-of-Tasks discrete simulation engine has been developed by researchers of the Department of Computer Science at the University of Piemonte Orientale, and is used to perform parameter sweep studies of schedul-
      ing algorithms for desktop grids. The computing infrastructure provided by ShareGrid has enabled these researchers to enlarge the set of scenarios and scheduling algorithm that could be studied in a reasonable amount of time, thus significantly increasing their possibility of investigating interesting avenues of research that could not have been investigated by using only the limited computing resources owned by their research laboratory.


  • Evaluation of Classifier Systems
    • Classification is the task of recognizing an object or event as an instance of
      a given class and represents one of the problems most frequently found in computer applications. Medical and fault diagnosis, prognosis, image recognition, text categorization, adaptive user profiling are weel known instances of classification tasks. The task of automatically inferring a classification program (classifier)
      from a set of previously classified data has been investigated for more than two decades in pattern recognition, statistics, and in machine learning and produced
      a large number of powerful algorithms. When new application domains and/or new requirements are sought, existing algorithms as well as new ones need to be
      tested and evaluated extensively to assess their performances. This evaluation activity consists in running the algorithm to acquire a classification program and test the learned model on, so called, test data, and is very time-consuming, because, several runs are performed by varying configuration parameters of the algorithm and the datasets on which the algorithms are tested. Each run of an algorithm can then be seen as a separate task and, thus, a Bag-of-Tasks paradigm can be applied to organize a large number of experiments and get advantage of a grid computing environment. In particular, the researchers at the Department of Computer Science at the University of Turin used the computing infrastructure provided by ShareGrid to test the performances of an SVM classifier on a user identification task. The experimental setup consists in running the SVM classifier on 64 different datasets, by varying three parameters of the algorithm for a total of about 1200 runs. Each run takes about 1 hour cpu time on average to run. This means that it would take more than a month to terminate all the experiments using a single cpu system. By using the ShareGrid computing power, the full experiment took only a couple of days.

  • Evaluation and suppression of noise caused by flows over a cavity
    • At the Aeronautical Department of the Politecnico di Torino, a group of researchers and PhD students works on the evaluation and suppression of noise caused by flows over a cavity. In this framework a method has been developed, which allows the prediction of the emitted noise based on 2-dimensional velocity field data, experimentally obtained by Particle Image Velocimetry (PIV). This method is advantageous as it allows to retrieve information about the acoustic emission of a flow and about the noise sources (flow field) at the same time.

      Specifically, from the basic equations in Fluidmechanics, the continuity and momentum equations, a wave equation can be derived, which has a source term and a propagation term. Such a wave equation used to predict aerodynamically created noise is called an acoustic analogy, where several formulations exist (Lighthill, Fowks Willams and Hawkins, Curle, Powell and others). In the present research the Curle’s acoustic analogy is used, as it accounts for bodies in the flow. The source term in the Curle’s acoustic analogy consists mainly of the hydrodynamic pressure over which one has to integrate to obtain the emitted noise.

      The usage of the Sharegrid environment is especially useful during the calculation of the hydrodynamic pressure from the PIV velocity data, which is computationally expensive. The algorithm first computes the Lagrangean acceleration. Successively, the pressure gradient is calculated inserting the obtained acceleration in the Navier-Stokes equation. Finally, the gradient is integrated in space applying a multi-path integration method, which is the computationally expensive step in the algorithm. Moreover, this calculation has to be done for a time series of velocity fields, typically consisting of several thousand snapshots. Being able to submit these jobs to the Sharegrid network, reduces the computation time significantly.



Login Form

Lost Password?
  Design by &