HPC Cluster Computing from 64 to 156,000 Cores 
5 likes1,997 views
In this video from ChefConf 2014 in San Francisco, Cycle Computing CEO Jason Stowe outlines the biggest challenge facing us today, Climate Change, and suggests how Cloud HPC can help find a solution, including ideas around Climate Engineering, and Renewable Energy. "As proof points, Jason uses three use cases from Cycle Computing customers, including from companies like HGST (a Western Digital Company), Aerospace Corporation, Novartis, and the University of Southern California. It’s clear that with these new tools that leverage both Cloud Computing, and HPC – the power of Cloud HPC enables researchers, and designers to ask the right questions, to help them find better answers, faster. This all delivers a more powerful future, and means to solving these really difficult problems." Watch the video presentation: http://insidehpc.com/2014/09/video-hpc-cluster-computing-64-156000-cores/
Ad
More Related Content
What's hot (20)
Viewers also liked (14)
Ad
Similar to HPC Cluster Computing from 64 to 156,000 Cores  (20)
Ad
More from inside-BigData.com (20)
Recently uploaded (20)
HPC Cluster Computing from 64 to 156,000 Cores 
- 1. HPC Cluster Computing from 64 to 156,000 cores Jason Stowe, Founder
- 2. You’ll learn about: How continuous/instant delivery helps invention/discovery
- 3. You’ll learn about: Climate, Organic Semiconductors, Lean manufacturing Technical Computing
- 4. About Cycle Computing We believe utility access to technical computing accelerates discovery & invention
- 5. Our mission: Easy HPC applications for anyone, on any resources, at any scale
- 6. About real use cases: Life Sciences Life Sciences • 50,000-core utility supercomputer in the Cloud • Analyzed 21 million compounds in 3 hours • 12.5 processor years for < $15,000 Nuclear engineering Utilities / Energy • Approximately 600-core MPI workloads run in Cloud • Ran workloads in months rather than years • Introduction to production in 3 weeks Rocket Design & Simulation Asset & Liability Modeling (Insurance / Risk Mgmt) • Completes monthly/quarterly runs 5-10x faster • Use 1600 cores in the cloud to shorten elapsed run time from ~ 10 days to ~ 1-2 days Manufacturing & Design • Enable end user on-demand access to 1000s of cores • Avoid cost of buying new servers • Accelerated science and CAD/CAM process • 39.5 years of drug compound computations in 9 hours, at a total cost of $4,372 • 10,000 server cluster seamlessly spanned regions • Advanced 3 new otherwise unknown compounds in wetlab • Moving HPC workloads to cloud for burst capacity • Amazon GovCloud, Security, and key management • Up to 1000s of cores for burst / agility
- 7. About me: I’m a geek from the 90s
- 8. 90s geek?? Whilst I like hacker news, reddit, etc. Slashdot is my homepage
- 12. It will affect everyone
- 15. What do you think?
- 16. Buy land in Canada?
- 17. Maybe, but hopefully you’re getting: this is a big deal
- 18. Just in case… Chairman Rajendra Pachauri Intergovernmental Panel on Climate Change "Nobody on this planet is going to be untouched by the impacts of climate change”
- 19. US Secretary of State John Kerry: "Unless we act dramatically and quickly, science tells us our climate and our way of life are literally in jeopardy. Denial of the science is malpractice."
- 23. Possibilities? Wind power Nuclear Fusion GeoThermal Climate Engineering Nuclear Fission energy Solar Energy BioFuels
- 24. How do we simulate/engineer a solution?
- 25. How do they create these?
- 26. They use a supercomputer…
- 27. Supercomputers = great • Supercomputers have special features, like fast interconnects • So good they should never run R, genomics, monte carlo, etc. that don’t use *Fast Interconnect* • During peak usage, our users have wanted: – Latency sensitive – Large scale/grid MPI Running on bare metal, put other jobs elsewhere Flickr: LouisRix
- 28. Limitations of fixed infrastructure Too small when needed most, Too large every other time… • Upfront CapEx anchors you to aging servers • Costly administration • Miss opportunities to do better risk management, product design, science, engineering
- 29. Lets talk about materials science
- 30. Specifically…
- 31. Designing Better Solar Materials The challenge is efficiency - turning photons to electricity The number of possible materials is limitless: • Need to separate the right compounds from the useless ones • Researcher Mark Thompson, PhD: “If the 20th century was the century of silicon, the 21st will be all organic. Problem is, how do we find the right material without spending the entire 21st century looking for it?”
- 32. Needle in a Haystack Challenge: 205,000 compound families totaling 2,312,959 core-hours or 264 core-years
- 33. Supercomputers = Fast & costly Flickr: LouisRix
- 34. Another tool in the box: Cloud • At peak load, use up to massive scale to run throughput workloads on cloud: – Needle in a Haystack – Whole Sample set – BigData/Throughput i/o – Integration, monte-carlo paths – Interactive, Service oriented workloads (master worker) • Workload run-time rather than individual job run-time Flickr: MaCDavis
- 35. Most of Cycle’s users use 40 - 4000 cores, Mark Thompson’s problem wasn’t that problem
- 36. Needle in a Haystack Challenge: 205,000 compounds totaling 2,312,959 core-hours or 264 core-years
- 37. 205,000 molecules 264 years of computing 16,788 Spot Instances, 156,314 cores
- 38. 8-Region Deployment US-West-1 US-East EU US-West-2 Brazil Tokyo Singapore Australia
- 39. 205,000 molecules 264 years of computing 156,314 cores = 1.21 PetaFLOPS (~Rpeak)
- 40. 1.21 PetaFLOPS (Rmax+Rpeak)/2, 156,314 cores
- 42. 205,000 molecules 264 years of computing Done in 18 hours Access to $68M system for $33k
- 44. How did we do this? Automated in 8 Cloud Regions, 5 continents, double resiliency Auto-scaling Execute Nodes JUPITER Distributed Queue Data … 14 nodes controlling 16,788
- 46. Now Dr. Mark Thompson Is 264 compute years closer to making efficient solar a reality using organic semiconductors…
- 47. Important?
- 50. Lean Manufacturing (Japanese Manufacturing to turn raw materials to inventory)
- 51. Continuous Delivery (Modern software process to turn builds to production applications)
- 52. Users want to decrease Cycle time = (prep time + queue time + run time)
- 53. The Scientific Method Ask a Question Hypothesiz e Predict Experiment / Test Analyze Final Results Test and Analyze stages require the most time, compute, and data
- 54. What’s holding us (Engineering, Science, Risk) back?
- 55. Utility HPC Access = Continuous HPC capacity delivery
- 56. The Innovation Bottleneck: Scientists/Engineers forced to size expectations to the infrastructure they have access to
- 57. How does cloud help?
- 58. Put the important, fast interconnect jobs on fast interconnect machines Put everything else on cloud capacity, when needed
- 59. Workloads are actually helping Serial or MPI job Serial or MPI job (may need 1 core or many servers) Serial or MPI job Serial or MPI job (may need 1 core or many servers) Serial or MPI job (may need 1 core or many servers) Serial or MPI job (may need 1 core or many servers) Fixed Cluster Workload can benefit from dynamic cluster For these workloads: $$$/science and throughput **are as important as** individual task performance RUN TIME
- 60. Utility Access is Revolutionizing clients Life Sciences Life Sciences • 156,000-core utility supercomputer in the Cloud • Used $68M in servers for 18 hours for $33,000 • Simulated 205,000 materials (264 years) in 18 hours Nuclear engineering Utilities / Energy • Approximately 600-core MPI workloads run in Cloud • Ran workloads in months rather than years • Introduction to production in 3 weeks Rocket Design & Simulation Asset & Liability Modeling (Insurance / Risk Mgmt) • Completes monthly/quarterly runs 5-10x faster • Use 1600 cores in the cloud to shorten elapsed run time from ~10 days to ~ 1-2 days Manufacturing & Design • Enable end user on-demand access to 1000s of cores • Avoid cost of buying new servers • Accelerated science and CAD/CAM process • 39.5 years of drug compound computations in 9 hours, at a total cost of $4,372 • 10,000 server cluster seamlessly spanned US/EU regions • Advanced 3 new otherwise unknown compounds in wet lab • Moving HPC workloads to cloud for burst capacity • Amazon GovCloud, Security, and key management • Up to 1000s of cores for burst / agility
- 61. Cycle HPC Deployment Architecture Internal & External Jobs & data Faster interconnect capability machine File System (PBs) Cloud Filer Cloud Filer Cloud Filer Blob data (S3) Glacier Auto-scaling external environment HPC Cluster Internal HPC Blob data (S3) Glacier Auto-scaling external environment HPC Blob data Cluster Cold Storage Auto-scaling external environment HPC Cluster Config Data
- 62. Example Use Cases Cycle software tools securely orchestrate scientific, engineering, and finance workloads across AWS 10,600 instance Top 10 Pharma Molecular Modeling for Cancer Targets 160 core MPI/Engineering Nuclear Reactor, Hard Disk, & Rocket 1 Million hours RNA Analysis Genomics/Next- Gen Sequencing
- 63. Survey of Use Cases þ Drug Design þ CAE þ Genomics
- 65. Every day is ! crucial and costly
- 66. Challenge: Novartis run 341,700 hours of docking! against a cancer target (impossible to do in-house)
- 68. Utility technical computing server count:
- 70. Cycle Computing’s cluster view:
- 71. Metric Count Compute Hours of Science 341,700 hours Compute Days of Science 14,238 days Compute Years of Science 39 years AWS Instance Count 10,600 instances CycleCloud & Cloud! finished impossible run, ! $44 Million in servers, in…
- 72. 39 years of drug design in 11 hours on 10,600 servers for < $4,372
- 73. Does this lead to new drugs?
- 74. Novartis announced 3 new compounds going into screening based upon this 1 run
- 75. Survey of Use Cases þ Drug Design þ CAE þ Genomics …
- 77. Aerospace Corp @ AWS re:Invent
- 78. Reached out to Cycle Computing in Summer 2013
- 79. Aerospace Corp @ AWS re:Invent
- 80. Rocket Design on AWS/Cycle Remove caps on scale, speed time to result No QueueWait for users TBs FS Engineer File System Corporate Firewall Of 3 Quarters TC Cluster Of 3 Quarters TC Cluster Auto-scale Secure HPC Cluster Internal Cluster Application AWS GovCloud
- 81. HGST, A Western Digital Company
- 82. HGST, A Western Digital Company
- 83. Burst for TC workloads
- 84. Results • Gave 10 minutes to get 64-core to 512-core environments! • Parallel scaling greatly improved time-to-results • Simplified submission interface removed file transfer and remote computing complexity • Auto-scaling environments decreased costs while increasing science completed.
- 85. Survey of Use Cases þ Drug Design þ CAD/CAM þ Genomics …
- 90. Results • Continuously deliver up to 8,000 core of compute power to support online genomic analysis • Complicated application w/Web, NoSQL, TC clusters • Analysis as a Service for all IonTorrent users
- 91. Survey of Use Cases þ Drug Design þ CAD/CAM þ Genomics …
- 92. Now hopefully…
- 93. You believe utility access to technical computing accelerates invention
- 94. Supercomputers = great Flickr: LouisRix
- 95. And so is Cloud… Flickr: MaCDavis
- 96. Computing can revolutionize materials science
- 97. We need to engineer solutions for climate, and lower our footprint
- 98. Buy land in Canada?
- 99. We can solve this problem
- 100. Across many science, engineering, and risk Life Sciences Life Sciences • 156,000-core utility supercomputer in the Cloud • Used $68M in servers for 18 hours for $33,000 • Simulated 205,000 materials (264 years) in 18 hours Nuclear engineering Utilities / Energy • Approximately 600-core MPI workloads run in Cloud • Ran workloads in months rather than years • Introduction to production in 3 weeks Rocket Design & Simulation Asset & Liability Modeling (Insurance / Risk Mgmt) • Completes monthly/quarterly runs 5-10x faster • Use 1600 cores in the cloud to shorten elapsed run time from ~10 days to ~ 1-2 days Manufacturing & Design • Enable end user on-demand access to 1000s of cores • Avoid cost of buying new servers • Accelerated science and CAD/CAM process • 39.5 years of drug compound computations in 9 hours, at a total cost of $4,372 • 10,000 server cluster seamlessly spanned US/EU regions • Advanced 3 new otherwise unknown compounds in wet lab • Moving HPC workloads to cloud for burst capacity • Amazon GovCloud, Security, and key management • Up to 1000s of cores for burst / agility
- 101. To continuously deliver HPC to accelerate the human mind…
- 102. Does this resonate with you? We’re hiring like crazy: Software developers, HPC engineers, devops, sales, etc. jobs@ cyclecomputing.com