Demos Program

Demos Session I - Tuesday 10/22 4:30PM - 6:30PM

Chair: Jean-Francois Lalande

  • HyperMapper: a Practical Design Space Exploration Framework
    • Speaker: Luigi Nardi, Stanford
    • Abstract: This demo complete the paper presented in the Plenary Session. It will introduce HyperMapper 2.0 a new methodology and corresponding software framework. HyperMapper 2.0 handles multi-objective optimization, unknown feasibility constraints, and categorical/ordinal variables. This new methodology also supports injection of the user prior knowledge in the search when available.
  • Using MAQAO to analyse and optimise an application
    • Speaker: Cédric Valensi, Université de Versailles St Quentin en Yvelines
    • Abstract: MAQAO (Modular Assembly Quality Analyzer and Optimizer) is a performance analysis and optimisation framework operating at binary level, with a focus on core performance. Its main goal is to guide application developers along the optimization process through synthetic reports and hints. In this demo, we show how MAQAO can be used to analyse an application and provide advice on its code quality and on how to improve its performance. In particular, we will show how the MAQAO reports can assist users on optimising their application.
  • Addressing deviant I/O patterns on Parallel File systems
    • Speaker: Jean-Thomas Acquaviva, DDN Storage
    • Abstract: In this session, we will investigate the impact of the I/O access patterns on file systems. Starting with the emerging IO500 benchmark, we will demonstrate how specific patterns can be handled safely by some file systems, while at the opposite impact dramatically performance on others file systems. The purpose is not to showcase a specific file system but more to raise awareness of application developers on I/O design in their applications.
  • Leveraging cloud unused resources for Big data application while achieving SLA
    • Speaker: Jean-Emile DARTOIS, IRT B<>Com
    • Abstract: Although the use of virtualization has improved the utilization of computing resources in data centers, several studies have demonstrated that their average usage remains very low, between ~20% and 50% in the case of CPU. A promising alternative for optimizing deployment cost of applications on Cloud infrastructures is to opportunistically exploit their allocated but unused computing resources. The Institute of Research and Technology b<>com work on a project that aims to make unused and heterogeneous private IT resources available through a highly secured distributed Cloud to deploy applications at a cheaper price. The first use case of the project is to provide a framework that leverages unused Cloud resources to run Big jobs (i.e., Apache Spark). However, the operator (i.e., interface between the infrastructure owner and the customers) still have to meet the expectations of its customers in terms of Quality of Service while avoiding the interference between big data jobs and co-resident workloads (i.e., the resource providers). The demonstration will show how a consumer can use a community cloud to deploy Apache Spark jobs to process big data. It will also show how the Apache Spark can automatically scale in/scale out depending on farmers (i.e., infrastructure owner) workloads (used resources). To achieve that, the demo will show three mechanisms we have designed (i) a Forecasting builder to predict resource volatility, (ii) a QoS controller to ensure users SLA guarantee by avoiding interference, (iii) and a predictive reactive autoscaler to automatically scale in/scale out Apache Spark orchestrated by Kubernetes.

Demos Session II - Wednesday 10/23 Time TBD.

Chair: Jean-Thomas Acquaviva

  • Orange to create the web of things
    • Speaker: Pierre MEYE, Orange
    • Abstract: Thing in Scan & Scale demonstrates a set of tools designed to create objects into Thing in, a unique repository built by Orange to create the web of things. The demonstration shows not only how to add objects and relations into the Thing in graph through different technologies, but also how the platform’s architecture is designed to be able to scale according to the sharply growth of the graph.
  • PyMaO: Accelerating Analysis to tackle Android malware
    • Speaker: Jean-Francois Lalande, Supelec
    • Abstract: Experimenting with Android malware requires to manipulate a large amount of samples and to chain multiple analyses. Scripting such a sequence of analyses on a large malware dataset becomes a challenge: the analysis has to handle fails on the computer and crashes on the used smartphone, in case of dynamic analyses. We present a new tool, PyMaO, for handling such experiments on a regular desktop PC with the highest performance throughput. PyMaO helps to write sequences of analyses and handle partial experiments that should be restarted after a crash or continued with new unknown analyses. The tool also offers a post processing capability for generating number tables or bar graphs from the analyzed datasets.