Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

HPC DAY 2017 | Altair's PBS Pro: Your Gateway to HPC Computing

346 views

Published on

HPC DAY 2017 - http://www.hpcday.eu/

Altair's PBS Pro: Your Gateway to HPC Computing

Dr. Jochen Krebs | Director Enterprise Sales Central & Eastern Europe at Altaire

Published in: Technology
  • Be the first to comment

  • Be the first to like this

HPC DAY 2017 | Altair's PBS Pro: Your Gateway to HPC Computing

  1. 1. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. Altair's PBS Pro: Your Gateway to HPC Computing Jochen Krebs • Director Enterprise Sales Central & Eastern Europe • Oct 2017
  2. 2. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. 2500+Engineers, scientists and creative thinkers Founded 1985 Headquartered in Troy, MI US 60,000+Users $313M 2016 Billings 5000+Customers globally 50+ ISV partners under our unique, patented licensing model 67offices in 23 countries ALTAIR AT A GLANCE
  3. 3. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. Altair is the Only Company That… …makes HPC middleware: …develops HPC applications: …and uses these to do HPC every day:
  4. 4. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. Software & Solutions For HPC Clusters & Clouds Optimize HPC clusters and clouds: ensure policies, maximize efficiency, deliver resilience and security via job scheduling and data management Admins & Managers Control HPC resources and provide 360o visibility and agility: configure, deploy, monitor, troubleshoot, report, simulate, optimize Engineers & Researchers Access HPC resources naturally (no IT expertise): run solvers, view progress, manage data, and use 3D remote visualization
  5. 5. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. Clouds Clusters PBS Works Control AccessOptimize PBS Pro Engineers & Researchers HPC Admins
  6. 6. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. PBS Professional: Optimize HPC • Ubiquity: Open Source and commercial licensing • Performant: faster time-to-results, better throughput and utilization • Secure: EAL3+ security certification and MLS support • Intelligent: policy-driven and topology-aware scheduling • Scalable: proven to run millions of jobs per day • Green Provisioning™: power management and control • Robust: known in the industry for stability and support • Open Architecture: implement virtually any new policy Workload Management & Job Scheduling PBS Pro schedules 250k+ cores for NASA PBS Works chosen for Quriosity EFFICIENT AND EFFECTIVE UTILIZATION 
 OF EXPENSIVE HPC RESOURCES
  7. 7. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. PBS Pro: Speed, Scale, Resilience & The “Whole” HPC World • 5x more scalable • Tested to 50,000 nodes • Tested to 500 concurrent portal users • 10x faster • Both scheduling & analytics • 10x smaller footprint • Analytics memory & disk • 100% health check framework • Dual licensing: commercial & Open Source PBSPro.org
  8. 8. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. Topology-aware Scheduling Inter-node & intra-node placement Switches, clusters, and NUMA All networks Infiniband, Ethernet, custom Dynamic (runtime changeable) All topologies Before After Average runtimes ~ 45% Faster ** actual US Customer Reported Results Faster, Predictable Performance and Better Utilization
  9. 9. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. PBS Plugins (“Hooks”) • Meet unique enterprise requirements • Deliver site-specific and 3rd party integrations • Support platform-specific features 
 on day 1 • Build novel extensions • Crowdsource solutions • Share (even sell) via PBS ecosystem Say “Yes!” to unique requests PBS Plugin Plugin Framework for Agility and Innovation
  10. 10. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. PBS Pro® Dual-licensing Commercial The HPC World Public Sector Risk takers Early adopters Natural collaborators Open Source Risk avoiders Later adopters Natural competitors Private Sector
  11. 11. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. HPC Cloud & Exascale Ecosystems Exascale Cloud Computing Power Management Big Data GPUs & Xeon Phi Open Source
  12. 12. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. Ecosystem: Containers • Cgroups • Limit, account, isolate resource use • CPU, memory, disk I/O, network, etc. • Enforced by PBS Pro based on allocated resources • Docker & nvidia-docker • Specify and run inside containers • PBS hook-based implementation • Big issue: security model grants root access (inside containers) • Singularity • Newer container system (so not as mature) • No security issue, as container runs as user • Compatibility with Docker containers (via import) • No PBS config needed – “just works”. nvidia-docker cgroups
  13. 13. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. Power Management - Real Life Test Results Customer has tested power capping per Computer node level Testing was done with Harmonie 1.5km resolution job. Harmonie is a very CPU power and memory throughput demanding application. During a job run all Node CPUs cores (2*14c) nearly all the time were performing at 100% load. Job was run on 36*Compute nodes (1008c, 72*E5-2680v4 2.4GHz 14c, 128GB RAM per node). Job was repeated few times with each specific power cap and an average result was fixed. Green Provisioning™ Results: 1/ Without power cap -> walltime ~885s, power consumption per job run ~ 2,05kW 2/ Power cap 220W -> walltime increase ~2%, power consumption decrease per job run by 10% 3/ Power cap 200W -> walltime increase ~3%, power consumption decrease per job run by 17%. => Seems to be an optimal option. 4/ Power cap 180W -> walltime increase ~8%, power consumption decrease per job run by 20% 5/ Power cap 160W -> walltime increase ~28%, power consumption decrease per job run by 17%
  14. 14. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. PBS Works: Access HPC • Novice to Expert: simple and powerful • Same UX: desktop, web, and mobile • Secure: protected access to HPC resources • End-to-end: submit, monitor progress, steer, fix, and rerun jobs • Collaborate: shared 3D analysis • 3D Remote Visualization PORTAL FOR ENGINEERS & RESEARCHERS Gateway to collaborative innovation
  15. 15. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved.Copyright © 2014 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. 15 Simplify HPC services – no IT degree needed Compute Manager Run | Monitor | Manage • Single “pane of glass” for HPC work 
 (zero client install – browser only) • Drag-and-drop input decks to submit HPC work (custom options auto- populate) • Watch progress & view intermediate results 
 while jobs are running • Post-process and visualize remote results 
 (e.g., graph energies) without moving files • Even fix inputs “in place” and re-run 
 (without re-uploading big input files) • Secure, unified access, anywhere… anytime “Compute Manager is a key enhancement to our PBS Works environment.” – Tower International
  16. 16. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved.Copyright © 2014 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. 16 Key Application Integrations (by industry) Life & Chemical Sciences Energy / Oil & Gas Manufacturing And many more Amber LAMMPS NAMD CFX Permas NONMEM OpenFOAM • PBS Works products are integrated with the most popular and widely used software across all vertical industries • Hundreds of commercial, open source and proprietary code integrations
  17. 17. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved.Copyright © 2014 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. 17 Visualize your datasets for rapid collaboration and innovation Display Manager Visualize | Access | Collaborate “Display Manager is our most significant functionality update since the move to HPC itself. In most cases we are saving a full day of time or more.” – PING Golf • Avoid commute and keep data near your compute! (consolidate hardware and software resources) • A secure and traceable way to access IP • Easy and accessible collaboration tools 
 for globalized infrastructure • Reduces software maintenance and deployment costs • Instantaneous application invocation • Flexible application resource allocation • Allows for instant collaboration on big data through a simple interface, anywhere & anytime
  18. 18. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. PBS Works: Control HPC • Single pane of glass: single command center • Real-time monitoring: simplify troubleshooting and maintenance • Reporting: PBS Analytics powered by Carriots Analytics™ • Workload simulator: simulate and optimize infrastructure sizing • Multi-cloud bursting: burst to any cloud for peak loads • One-click appliance deployment: effortless for public, hybrid, and private clouds • Modern UX: drag-and-drop simplicity Manage HPC Clusters & Cloud Appliances Altair’s own CAE Appliances HyperWorks Unlimited HPC ADMINISTRATORS’ CONTROL CENTER FOR MANAGING, OPTIMIZING AND FORECASTING HPC RESOURCES
  19. 19. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. PBScloud.io (SaaS offering) • Graphical IDE: easily create HPC appliances • Easy deployment: create once and deploy in minutes • Multi-cloud: deploy complete appliance everywhere • Security: customize security polices and role-based access control • Intelligent management: appliance lifecycle management (CRUD) • Control expenditure: quotas to optimize consumption Democratizing HPC by providing fully configured HPC Appliances in multi-cloud environments Democratizing HPC by providing fully configured HPC Appliance on multi clouds A SOFTWARE-AS-SERVICE TO MODEL, CREATE, DEPLOY, 
 MANAGE & MONITOR HPC APPLIANCES ON MULTIPLE CLOUDS Learn more at: pbscloud.io
  20. 20. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. PBS Works: Cloud Bursting Microsoft Azure Amazon Web Services … PBS Works On-demand use of cloud resources to maximize efficiency
  21. 21. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. • Video Cloud
  22. 22. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved.Copyright © 2014 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. HP Insight CMU + PBS Professional: The Connector enables: So Users/Admins can: Automation of common admin tasks Do their jobs faster and easier Dynamic OS provisioning Trust that all existing OS images are being comprehended by PBS for scheduling Dynamic node group creation Visualize utilization and performance of jobs on nodes Integrated PBS menus Easily register nodes, online/offline (or “drain”) nodes, perform OS provisioning, delete jobs and much more Automatic network topology configuration Optimize job placement Maintenance mode for nodes Move 'bad' nodes into maintenance mode for troubleshooting without competing with users’ jobs One-click access to online resources Quickly and easily access PBS Professional documentation, user forums and support
  23. 23. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. Reimagined User Experience
  24. 24. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved.
  25. 25. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved.Copyright © 2014 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. 25 PBS Analytics Accounting & Analytics Portal for HPC Visualize historical usage for optimized returns on HPC investments • Allocate costs & plan for future capacity needs • Visualize workload & historical usage • Easily drill down to underlying data • Filter by project, app, user, group, queue, host, … • Canned reports out-of-the-box, and customize your own • Aggregate data from multiple PBS Professional servers • Analyze historical data as far back as PBS Pro 5.3 • Slideshow mode for continuous display (in lobby) • Export to customer billing system “Maximizing our license utilization means we don't have to buy a 
 new license, set up another workstation, and hire another engineer to keep up with demand.” – Trelleborg View | Report | Forecast
  26. 26. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. Copyright © 2015 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. 2619 Example: Users Request Too Much Memory… Users are asking for more memory than needed. 
 This could have caused some jobs to wait unnecessarily…
  27. 27. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. pbs_snapshot PBS Works: Configuration PBS Works: Analytics Simulator Solver PBS Pro Scheduler Core PBS Snapshot • • Policies • Infrastructure • Workload PBS Snapshot Simulator Architecture
  28. 28. © 2017 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. Why PBS Works? Save Money Save Time Gain Peace of Mind Ensure User & Admin Satisfaction Scale Up “Tens of millions of dollars in upfront savings.” –Lockheed Martin “We are saving a full day of time or more.” –PING Golf “Admin overhead is reduced significantly.” –IT4I “Altair… drew on their engineering expertise to make creative suggestions. They are problem solvers.” –GE Oil & Gas “We’ll be running orders of magnitude more cores than in the past….PBS will play a central role in enabling us to do that well.” –Chrysler

×