Managing and Deploying High Performance Computing Clusters Using Windows HPC Server 2008 R2Saptak Sen [[email protected]]Senior Product ManagerTechnical ComputingMicrosoft Corporation
SESSION CODE: WSV320
Session Objectives and TakeawaysSession Objective(s): First glance at System Management features for Windows Server 2008 R2, HPC pack.Key Takeaways
WHPC2K8R2 will enable diskless compute nodes, via iSCSI boot New Heatmap, allows the cluster admin to view cluster state at a glance.Diagnostic extensibility allows the ISVs and administrators to author custom tests for windows HPC.Reporting extensibility will allow cluster administrators to create custom reports.WHPC2K8R2 will enable customers to expand the capacity of HPC clusters by utilizing desktop cycles
Enable personal supercomputing
Scale to divisional sized clusters, enable new
categories of interactive HPC applications, and interoperate with existing IT infrastructure.
Integrate well with existing HPC and IT infrastructure.
Highly performant and scalable parallel applications.
Deploy, manage, and troubleshoot clusters of up to
1000s of nodes.
HPC Server Mission and Vision
Make Windows Server the mainstream way to solve scaled-out business, engineering and scientific computational problems.
System Management Vision
MONITORING: Ability to see at a glance current/recent changes to the system across the whole stack( that affect my app)
Deployment Monitoring
Diagnostics Reporting
Provide reliable and consistent system management and “excellent
manageability” at departmental and divisional scale (1000s of
nodes).
WS2K8WS2K8 R2WS2K8WS2K8 R2
WS2K8WS2K8 R2
Windows HPC Server 2008 R2 Platform
Head Node Windows Server 2008 R2SQL Server 2008 Powershell 2.0
Compute NodesWindows Server 2008 and/or Windows Server 2008 R2
Client support back to XPFlexible setup
Database configuration: Files and location
V2
V2 V2
WS2K8
V2…
SQL 2008
V3
V3 V3 V3
Backward Compatibility
System ManagementSide-by-side support for Admin Console on same workstationV2 node.xml file works on V3 cluster without changes.Upgraded V2 Node Templates work on V3.
User Console and Job Scheduler API Compatibility more details in the Job Scheduler Presentation
V2 Head node
V3 Head node
+
- +. . . .
: Search:
s
Admin Workstation
•Node.xml•Node Templates•Images
Deployment in Windows HPC 2008
We deploy 250 nodes, from bare metal in under an hour.
Built on Windows Deployment Services TransportUses PXE and TFTP servers.Image based DeploymentCustom Window Deployment Services provider to orchestrate the deployment process.OS image is multicast by default.
Support five basic network topologiesOffline Driver injection via Admin console
One step driver ( INF ) injection/removal Including IB drivers!
Patching in Windows HPC 2008
Ability to patch compute nodes via node deployment templatesIntegrated with Microsoft Update.Provide a patching level - Critical, All or None.Provide specific patch IDs to be applied to the compute nodes.
Cluster wide Patching reports via DiagnosticsDetermine the current patch level across the cluster.Determine which nodes have patches pending.
Deployment Goals for V3
Scale support for 1000 compute nodes.Support for Wk28 and Wk28 R2 compute nodes.Upgrade HN from HPC v2 to HPC v3Support for Remote DB on HNSupport for Diskless Compute Nodes
Deployment Scale
1000s of node support.
Better troubleshootingDeployment Progress columns to spot outliers.Leverage Extensible Diagnostic framework
Investments in higher resilience and better failure recovery.
Remote Database.Move to asynchronous communication model.
WS2K8WS2K8 R2WS2K8WS2K8 R2
WS2K8WS2K8 R2
Upgrade
Upgrade Head Node to Windows Server 2008 R2Install HPC Server R2 on Head Node
Upgrade HPC Databases to SQL Server 2008 and HPC V3 Schema
Reimage Compute NodesWindows Server 2008 and/or Windows Server 2008 R2
SQL Express
2005V2
V2 V2
WS2K8
V2…
SQL Express
2008V3
V3 V3 V3
WS2K8WS2K8 R2WS2K8WS2K8 R2
WS2K8WS2K8 R2
Remote Database
Management and Scheduler Database can be ‘remote’.Cannot upgrade from local DB to remote DB.Any SQL SKUCluster Admin needs permissions on Remote DB
V2
V2 V2
WS2K8
V2…
SQL 2008
V3
V3 V3 V3
Data Center
Diskless HPC Cluster
Support diskless compute nodes at scale, by allowing compute nodes to boot from Storage Array on the network via iSCSI.
Diskless Compute NodesEnable diskless compute nodes via iSCSI boot.Compute nodes use volumes(LUNs) on Remote Storage as system disk. Use DHCP reservations to provide Storage Server information to the diskless compute nodes.Leverage Storage vendor provided differencing capacity to store Differencing Disks per compute node.
New in HPC Server 2008 R2
DHCP reservationsNew Patching WizardDeployment ProgressCapture Golden ImageUse Excel to edit node list XML
Monitoring Goals
• Manage and Monitor the cluster at scale.• Manage all resources from a single console.
Diagnostics in WHPC 2008Integrated diagnostics experience
Starting
• Select nodes and tests• Start test run
Running
• View Progress• Cancel Test
Completed
• View test result history• View detailed results• Pivot to failed nodes• View and clear alerts• Export result to Excel
New features in Diagnostics
Extensible diagnostics platformCustomizable parametersAdditional in-the-box test
Diagnostics Extensibility - Workflow• Each step is defined by a command line.
– This allows test to be developed in any coding/scripting language.
Reporting in V2
Four categories of built-in reports
Node Availability Report Job Resource Usage Report Job Throughput Report Job Turnaround Report
Reporting Data Extensibility in V3
Windows HPC Server Customer Data Analysis
Data Warehouse
Other Authoring Tools
DATADATADATA
Where are data generated and used?
Data Sources Management Job Scheduler Performance Metrics Event Logs
Data Collection Data Collector
Data StoreData Store
Data Presentation Historical ReportsMonitoring Charts
Scal
abili
ty, R
elia
bilit
y
Heat Map
Reporting Data Extensibility
Data Behind the Reports
• Node Attributes• Node Event History
Node Data
• Job Configurations• Job Task Summary• Job State and
Statistics• Job Allocation History
Job Data
• Performance Metrics History
Performance Metrics
Cluster of Workstations (CoW)
Feature SummaryThe ability to add Windows 7 workstations as compute nodesView and monitor workstations the same as dedicate compute nodes Time of day scheduling for Workstation availabilityDraining interval for graceful preemption
Expand the capacity of HPC clusters while increasing the return on your existing technology investments by utilizing desktop cycles as part of your over HPC infrastructure.
Requirements– Windows 7 Professional or Enterprise, 32 or 64-bit (requires
Active Directory support)– Desktops and cluster in same AD Domain– Desktops, head nodes, compute nodes, and broker nodes need
to be on the same physical, public, network (topology 5)
Security policy prevents unintentional deployment of an OS to desktop nodes
CoW DeploymentDesktop Nodes
Windows 7HPC Pack 2008 R2 Desktop
Compute Nodes
Windows HPC Server
Database Server(Optional)
Windows Server 2008 R2SQL Server
Head Node
Head node
Broker Nodes
Windows HPC Server
Windows HPC Server
Public Network
Additional Resources• Windows HPC Server 2008 R2 Technical Library - http
://go.microsoft.com/fwlink/?LinkId=168271• Windows HPC Server 2008 R2 Beta site on Microsoft
Connect - http://go.microsoft.com/fwlink/?LinkId=169523• Diagnostics Extensibility in Windows HPC Server 2008 R2
CTP2 Step-by-Step Guide - http://go.microsoft.com/fwlink/?LinkId=177604
• Test Focus Guide for Windows HPC Server 2008 R2 Beta 2 - http://technet.microsoft.com/en-us/library/ee783563(WS.10).aspx
• Twitter - @windowshpc
Resources
www.microsoft.com/teched
Sessions On-Demand & Community Microsoft Certification & Training Resources
Resources for IT Professionals Resources for Developers
www.microsoft.com/learning
http://microsoft.com/technet http://microsoft.com/msdn
Learning
Complete an evaluation on CommNet and enter to win!
Sign up for Tech·Ed 2011 and save $500 starting June 8 – June 31st
http://northamerica.msteched.com/registration
You can also register at the
North America 2011 kiosk located at registrationJoin us in Atlanta next year
© 2010 Microsoft Corporation. All rights reserved. Microsoft, Windows, Windows Vista and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries.The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market conditions, it should not be interpreted to
be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.
JUNE 7-10, 2010 | NEW ORLEANS, LA