Windows Azure and HPC Scenarios and workload Overview

Windows Azure and HPC Scenarios and workload Overview

Problem Statement To meet its statutory obligations for daily processing of space and terrestrial radio service notices and to solve complex problems, ITU has a need to perform CPU intensive calculations in a limited timeframe. Thus, as a component of IT system, ITU have developed a local distributed infrastructure which is currently used to manage the workload of calculations-on-demand submitted by our Membership. Such a workload is greatly timedependent and is based on the pull model, with a database acting as the master and a number of Windows services as the slaves.

The Windows Azure Platform Developer Experience Use existing skills and tools. Windows Azure Platform Compute Storage Management “Operating system in the cloud” CDN “Middleware in the cloud” “Relational database in the cloud”

Microsoft HPC Scenarios overview

Microsoft components for HPC On-premise HPC Head Node Desktop Compute Cloud via Idle Win 7 Workstation Cores HPC Edition Desktop User HPC Cluster Broker Node(s) Azure Compute Proxies Azure Compute Instances

SOA on an HPC Cluster Compute Nodes 1. Create Session 3. Requests Head Node 2. HN provides WCF Broker node to Client and Session Manager assigns WCF Broker node for Client job 4. Requests Workstation 5. Responses 6. Responses WCF Broker Nodes Azure. CN 1……. Azure. CNx

On Premise Solution Windows HPC Server 2008 R 2 +SOA scheduler Dedicated nodes – HPC Servers 2008 R 2 Cycle scavenging – Windows 7 desktops

Burst in Azure Solution Windows HPC Server 2008 R 2 Azures Nodes Extend the on-premise cluster to the cloud, absorb peak demand, keep windows HPC applications performing without change. With your Azure subscription additional resources when you need them Pay only when you use them

Applications in Azure Solution Application in hosted entirely in Azure • No on-premise Head Node Application is accessed from • • • Rich client on premise Web Application or portal Web Service Multiple business models • • Application usage is licensed to customers Application and compute is provided as a service in the cloud Scheduler and Runtimes Supported in Azure • • Embedded Azure Scheduler, not a Head Node All runtimes: Parametric Sweep, MPI, Cluster SOA, Excel

Timescale compression 1 1 server 100 weeks 100 Buy a server $8 k 100 servers 1 week* £ 8 k 1

Examples of Azure HPC Workloads ü Finance (CM+ Insurance) § § Risk/Analytics ESG/ALM ü Media § § Encode/Transcode Analytics ü Life science ü Genomics ü Cell profiling ü Engineering/Universities ü Some optimisation simulations

Excel Workbooks on the Cluster st ue on si es S q Re sig s r. A Spreadsheet e ok r B Job nt. Head Node e nm nm k Sub mis s Ta sion sig s A lts u s Re Res ults Brokers t en

Thank you. .