POV-Ray m.in.
Transkrypt
POV-Ray m.in.
DISTRIBUTED EXECUTION OF DYNAMICALLY DEFINED TASKS ON MICROSOFT AZURE Piotr Wiewiura1, Maciej Malawski1, and Monika Piwowar2 1 AGH University of Science and Technology, Department of Computer Science, Krakow, Poland 2 Jagiellonian University, Department of Bioinformatics and Telemedicine, Krakow, Poland Background Requirements for DTL ExonVisualiser Goals Distributed execution Visualization of protein regions encoded by particular exons To evaluate suitability of Microsoft Azure as a platform for execution of computational applications Install-and-forget worker service To create a lightweight library to execute arbitrary tasks with arbitrary data on Azure. Simplicity Allows “tracking” gene expression by identification parts of the proteins being in relation with gene structure (codding regions - exons). To port existing ExonVisualizer application into Azure Dynamic deployment and execution Dynamic horizontal scaling Windows Azure features Virtual machines created on-demand Blob Storage Queue Services Dynamic Horizontal Scalability Good tool for analysis proteins that are synthesized in the alternative splicing and also in identification protein regions created in the post-translational modifications process. Portability and independence from execution platform Written in C# Support for Linux (Mono) Worker service auto-update → Result: Distributed Task Library (DTL) Architecture of DTL Implementation of DTL Microsoft Azure REST API Programming language and platform independence (Mono requirement) Dynamic task definition Binaries sent with tasks. Rapid development, testing and usage. Arbitrary data sources Byte streams, objects, files, storage services Experiments with DTL on Azure VM Instance creation times POV-Ray Banchmark ExonVisualizer Instantiating 1 to 20 Small Instance VMs Ray tracing 128x128 px benchmark scene. Large input size and fast execution First virtual machine is ready in about 235 s and each next takes further 45 s Combination of 1-16 segments and workers Conclusions References New Distributed Task Library (DTL) implemented and tested ExonVisualiser adapted to Azure using DTL Malawski, M., Meizner, J., Bubak, M., Gepner, P.: Component approach to computational applications on clouds. In: Proceedings of the International Conference on Computational Science, ICCS 2011 Azure is a worthy platform for computational applications Piwowar, M., Porembski, K., Piwowar, P.: ExonVisualiser - application for visualization exon unitsin 2D and 3D protein structures. Bioinformation 8(25), 1280–2 (2012) For additional information, please contact: Maciej Malawski Department of Computer Science AGH University of Science and Technology [email protected]