Cycle Computing Record-breaking Petascale HPC Run


Published on

In this slidecast, Jason Stowe from Cycle Computing describes the company's recent record-breaking Petascale CycleCloud HPC production run.

"For this big workload, a 156,314-core CycleCloud behemoth spanning 8 AWS regions, totaling 1.21 petaFLOPS (RPeak, not RMax) of aggregate compute power, to simulate 205,000 materials, crunched 264 compute years in only 18 hours. Thanks to Cycle's software and Amazon's Spot Instances, a supercomputing environment worth $68M if you had bought it, ran 2.3 Million hours of material science, approximately 264 compute-years, of simulation in only 18 hours, cost only $33,000, or $0.16 per molecule."

Learn more:
Watch the video presentation:

Published in: Technology, Business
  • Be the first to comment

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Cycle Computing Record-breaking Petascale HPC Run

  1. 1. Record-breaking Petascale CycleCloud HPC Production Run 156,000-core Cluster (1.21PetaFLOPS) Accelerates Schrödinger Materials Science and Green Energy November 2013 Cycle Computing
  2. 2. Cycle Computing believes utility high performance computing accelerates invention
  3. 3. Records broken, Science done On November 3rd, ran a “MegaRun” cluster that had: • 156,314 cores and 1.21 PetaFLOPS of theoretical peak compute power • Ran 2.3 Million hours, totaling 264 years of computing, in 18 hours • Executed world-wide, across all 8 public AWS Regions (5 continents) • Compared to $68Million to purchase – done on CycleCloud with Spot Instances for just $33K THE SCIENCE • Finding Organic Photovoltaic Compounds that are more efficient, easier to manufacture to help remove the US’s reliance on fossil fuels. • Designing, synthesizing, and experimenting with a new material can take 1 year of a scientists time requiring hundreds of thousands of dollars in equipment, chemicals, etc. or With Schrödinger Materials Science’s tools, on Cycle and AWS Spot Instances, it cost $0.16 per molecule • The run analyzed 205,000 compounds in total • This is the exact kind of science being outlined in the Materials Genome initiative from the White House
  4. 4. Challenge of Materials Science Traditional Materials Design •  Design, Synthesis, Analysis are challenging for an arbitrary material •  Low hit rate for viable materials •  Total Molecule Cost: •  Time: A year for a grad student •  $100,000s in equipment, chemicals, etc. With Schrödinger Computational Chemistry & Cycle •  Schrödinger Materials Science tools simulate accurate properties in hours •  Simulation guides the researcher’s intuition •  Focus physical analysis on promising materials •  Total cost: •  Time to enumerate molecules: Minutes/ hours •  $0.16 per molecule in infrastructure using AWS Spot Instances
  5. 5. Designing Solar Materials The Challenge is efficiency •  Need to efficiently turn photons from the sun to Electricity The number of possible materials is limitless •  Need to separate the right compounds from the useless ones •  If the 20th century was the century of silicon, the 21st will be all organic How do we find the right material, without spending the entire 21st century looking for it?
  6. 6. The Challenge for the Scientist Dr. Mark Thompson Professor of Chemistry, USC “Solar energy has the potential to replace some of our dependence on fossil fuels, but only if the solar panels can be made very inexpensively and have reasonable to high efficiencies. Organic solar cells have this potential.” Challenge: run a virtual screen of 205,000 molecules in continuing analysis of possible materials for organic solar cells
  7. 7. The right needle in the right hay stack Before: Trade-off between compute time vs. sampling Coarse screen, Small samples Now: Better analysis, more materials è Better results Higher Quality Analysis, More materials More Materials More Materials
  8. 8. Solution: Utility HPC On-demand compute power is transformative for users, but hard to make production —  Big Opportunity to help Manufacturing, Life Science, Energy, Financial companies: —  Rise of BigData, compute, Monte Carlo problems that power modern business and science —  Applications, like Schrödinger Materials Science tools, offer a compelling alternative to physically testing products —  Amazon Web Services makes infrastructure easily accessible —  AWS Spot instances decrease the cost of compute —  Science & engineering face faster time-to-market, increased agility requirements —  Capital efficiency (OpEx replacing CapEx) are organizational goals
  9. 9. Why isn’t everyone doing this? Because it is really complicated, and really hard to orchestrate technical applications, securely, at scale We’re the first and only ones doing this including the wellpublicized: 2000, 4000, 10000, 30000, and 50000 core clusters in 2010-2013 Clients including: Johnson & Johnson, Schrödinger, Pfizer, Novartis, Genentech, HGST, Pacific Life Insurance, Hartford Insurance Group …
  10. 10. Cycle Computing Makes Utility HPC a Reality Easily orchestrates complex workloads and data access to local and Cloud HPC —  Scales from 100-1,000,000 cores —  Handles errors, reliability —  Schedules data movement —  Secures, encrypts and audits —  Provides reporting and chargeback —  Automates spot bidding —  Supports Enterprise operations
  11. 11. Challenge: 205,000 compounds totaling 2,312,959 core-hours, or 264 core-years
  12. 12. Solution: “MegaRun” Cluster New record: MegaRun is the largest dedicated Cloud HPC Cluster to date on Public Cloud Tool   Description   Schrödinger  Materials  Science   tools   Set  of  automated  workflows  that  enable  organic  semiconductor   materials  to  be  simulated  accurately   CycleCloud   HPC  clusters  at  small  to  massive  scale:  application  deployment,   job/data  aware  routing,  error-­‐handling   Jupiter   Cycle’s  massively  scalable,  resilient  cloud  scheduler     Chef   Automated  configuration  at  scale   Multi-­‐Region  AWS  Spot  Instances   Massive  server  resource  capacity  across  all  public  regions  of  AWS  
  13. 13. 205,000 molecules 264 years of computing 16,788 Spot Instances, 156,314 cores!
  14. 14. 205,000 molecules 264 years of computing 156,314 cores = 1.21 PetaFLOPS (Rpeak) Equivalent to Top500 Jun2013 #29
  15. 15. 205,000 molecules 264 years of computing Done in 18 hours Access to $68M system for $33k
  16. 16. 8-Region Deployment US-West-2 US-East EU Tokyo US-West-1 Brazil Singapore Australia
  17. 17. Jupiter Scheduler —  Make large cloud regions work together —  Spans many regions/datacenters to resiliently route work with minimal scheduling overhead —  Batch/MPI Schedulers get 10k cores doing 100k jobs —  Jupiter seeks to get Millions of cores doing 10Ms tasks —  Currently 100k’s cores doing 1M tasks on large runs —  Can survive machine, availability zone, and region failure while still executing the full workload
  18. 18. Resilient Workload Scheduling
  19. 19. MegaRun – Facts and Figures Metric Count Compute Hours of Work 2,312,959 hours Compute Days of Work 96,373 days Compute Years of Work 264 years Molecule Count 205,000 materials Run Time < 18 hours Max Scale (cores) 156,314 cores across 8 regions Max Scale (instances) 16,788 instances
  20. 20. Accelerated Time to Result Cluster Scale Cost Run-time 156,000 core CycleCloud $33,000 ~ 18 hours 300-core Internal cluster (stopping all other work) $132,000 ~ 10.5 months
  21. 21. CycleCloud–156,000 cores
  22. 22. CycleCloud – 16,788 instances
  23. 23. 8 Public Regions across AWS
  24. 24. Ramping up to full capacity
  25. 25. Solution: 205,000 compounds, 264 core years, 156k core Utility HPC cluster in 18 hours for $0.16/molecule using Schrödinger Materials Science tools, Cycle & AWS Spot Instances