Project Description
Distributed Service Oriented Engineering Design
YourGrid.Net: ‘our technology your grid’
Prof Simon Cox and Dr Kenji Takeda
Microsoft HPC Institute - Southampton University
School of Engineering Sciences, University of Southampton, UK
The application of numerical optimisation in engineering design using sophisticated modelling and analysis tools can provide a vital competitive edge to companies. Engineering design is an iterative, multidisciplinary process that is often data intensive and computationally expensive due to the application of high fidelity analysis models for the simulations of physical phenomena. In the past few decades, engineering design has become increasingly dependent on computing and IT to underpin the process of design from initial modelling and analysis through optimisation to fabrication and testing of prototypes. We demonstrate why, where, and how we are exploiting current and future Microsoft tools and technologies to make the engineering design process faster, cheaper and better.
In particular, we currently use:
· Microsoft Cluster Compute Edition to provide ‘high performance’ capability computing,
· Microsoft tooling to build Open Standards based Web Services for access to distributed capacity computing resources and to integrate heterogeneous resources,
· Windows Communication Framework to provide a secure, reliable and transacted messaging framework,
· WinFS and SQL server for data handling and integration, and
· Tablet PC technology for users to interact with all aspects of the design process.
Demo Cluster Version 1 (SC'05)
OUR DEMO+TEST CLUSTER VERSION 1 (as used for demo at SuperComputing'05):
· Dual Opteron 246 HE (single core, low power version, revision E, 2.0GHz)
· CC nodes virtualised with VMWare, with a third machine used as domain controller, firewall and web service host.
· Gigabit Ethernet
· We use 1 NIC per node. As our cluster is virtualised we decided to keep the configuration simple as additional virtual NICs in a VMware node would not give much performance benefit (as the virtual NICs would still share one physical NIC as only one NIC was available). We will look into installing more physical NICs for later test clusters, although this will not be required on the final production cluster.
Spitfire Production Cluster (June 2006)
The cluster consists of 64bit dual-core AMD Opteron based machines that support a total of 304 independent processes, 800GB memory and 30TB storage space. Software installed on the cluster includes Windows Server 2003 Compute Cluster Edition, MS SQL Server 2005, Windows Workflow Foundation, Fluent and Matlab.