1. A scalable and flexible platform to run various types of resource intensive applications on clouds ISWG2015 3-5 June 2015 Budapest, Hungary Tamas Kiss, Hannu Visti, Gabor Terstyanszky, Gregoire Gesmier t.kiss@westminster.ac.uk h.visti@westminster.ac.uk

2. Funded by the European Commission FP7 programme, FoF: Factories of the Future July 2013 – December 2015 EUR 4.5 million overall funding Coordinated by the University of Westminster 29 project partners from 8 European countries 24 companies (SMEs) and 5 academic/research institutions An example for intensive industry collaboration CloudSME - EU funded industry oriented research project

3. * The CloudSME project develops a cloud-based, one-stop-shop solution providing a scalable platform for small or larger scale simulations, and enable the wider take-up of simulation technologies in manufacturing and engineering SME’s. The CloudSME project Cloud-based Simulation platform for Manufacturing and Engineering * Defines generic and concrete business models for SMEs in the manufacturing/engineering sector to facilitate the take-up of cloud-based simulation solutions * CloudSME builds a simulation platform that allows seamless access to multiple heterogeneous cloud resources and provides a high level of abstraction to users when accessing these resources for simulations in a one-stop-shop solution. * Provides a Platform as a Service (PaaS) solution to build customised cloud applications * Enables simulation software providers to offer Software as a Service (SaaS) simulation solutions * Enables SMEs in the manufacturing and engineering domain to access simulation services * Provides seamless access to HPC resources in order to speed up the simulations on-demand

4. The CloudSME Simulation Platform

5. Targeted application areas * High Performance Computing (HPC) * relatively small number of end-users with high computation demand * many used simulation before * enable software vendors to extend their product with cloud support * High Throughput Computing – parameter sweep (HTC) * use of multiplier entities (e.g. consultant companies) and technologies (e.g. templates) * although smaller in scale, still computationally intensive, typically parameter sweep * Scalable Web Applications (SWA) * Use CloudSME Simulation Platform (CSSP) to deploy and manage web services on public Internet.

6. High Performance Computing * Example Case: Fluid dynamics simulation * Benefits * Provides access to various clouds, grids and clusters – allowing execution of code where it is optimal either from performance or cost perspective * Challenges * Scales well within a single node by adding more cores. Communication between instances is a problem if more than one node is needed. CloudSigma are working on a solution in CloudSME project, and there is also supercomputer access from the platform.

7. High Throughput Computing * Example Case: Repast * Repast: Recursive Porous Agent Simulation Toolkit * Requires a model (Java code) and a parameter file * Benefits * CloudSME infrastructure provides parameter sweep workflow functionality and possibility to orchestrate jobs to different clouds. * Challenges: * Each model is different. Some are computationally heavy and a single run takes a long time to complete, while others complete in fractions of a second. * Currently each run runs exactly one simulation with a certain model and parameter file. If thousands or tens of thousands of runs are required, this needs to be optimised.

8. Scalable Web Applications * Example Case: Outlandish and Tidybooks * Three-tier architecture (proxy server with public IP address, application server and database server) * Can run LAMP or MEAN stack – not limited to any particular architecture * Can be managed from Cloudbroker GUI or Cloudbroker API * Benefits * CloudSME infrastructure allows access to different clouds. A Cloudbroker job is a representation of application functionality and data, which fits to Web application architecture neatly. * Setting up the architecture is basically a workflow

9. HTTP proxy1 HTTP proxy 2 HTTP proxy N App server 1 App server 2 App server 3 App server N DB server 1 DB server 2 Replication Internet Public IP Private IP Admin server Admin server Scalable Web Applications

10. * Challenges: * Needs public IP address. Academic clouds generally do not provide this. * Cost and performance optimisation + need to add mechanisms of measuring. * If the web service traffic volume varies a lot, it would need to be scaled up or down on the fly. Managing this, especially between clouds, would require a portable representation of the application, its configuration and also data. Means of transport exist already in Cloudbroker input-output parameters, but there are problems (potentially massive volumes of data – financial optimisation would need a way of querying this) * Currently not tested with WS-PGRADE portal – the setup workflow is implemented as a Python script. There is not currently an easy way to run Cloudbroker API commands from workflows. This could be solved by a front-end application deployment but doing this would not be trivial.

11. Any questions? http://cloudsme.eu