Axceleon Presentation at Siggraph 2009
1 like469 views
High performance computing, which was traditionally used by scientific and engineering organizations, is now being used widely in the visual effects and animation industry. Large render farms with hundreds or thousands of nodes are now common at major studios. The amount of computing power and storage needed for projects keeps growing dramatically. Resource management software like EnFuzion helps optimize these large render farms and enable on-demand, cloud-based rendering.
Axceleon Presentation at Siggraph 2009
- 1. High Performance Computing Comes to Hollywood Mike Duffy President and CEO Axceleon Inc.
- 2. Presentation Overview • Expected Users of High Performance Computing • UN-Expected Users of High Performance Computing • Elements of a High Performance Computing Compute Farm • The Evolving Image Rendering Ecosystem • Looks like HPC to us • What Happened? • Where Is The Industry Going? • Where Does EnFuzion Help In The Evolving 3D HPC Ecosystem • On-Demand Rendering Next Step • Axceleon
- 3. High Performance Computing (HPC) Expected Users of High Performance Computing • US Department of Energy National Laboratories – Lawrence Livermore, Sandia, Los Alamos • Airplane Manufacturers – Boeing, Airbus • Automotive Companies – GM, Ford, Toyota, Mazda, Honda, Isuzu • Engineering Companies – Lockheed, Northrup Grumman, Caterpillar, Peterbilt Trucks etc. • All need major computing horsepower to run their business
- 4. High Performance Computing (HPC) Un – Expected High Performance Computing Users • Motion Picture Companies – Disney, DreamWorks • Visual Effects houses building content for movies, games • Post production houses – get to market faster • Broadcasters world-wide – where time to the air is critical • Advertising Companies – those products with catchy ads • Gamers – fastest growing segment of the DCC marketplace • Medical Imaging – Better diagnosis, faster cures • These companies too need major computing horsepower to run and effectively manage their businesses
- 5. High Performance Computing (HPC)Farms Elements of a HPC Compute Farm • Usually HPC systems are located in Datacenters • Have “Big Iron”, hundreds of Multi-Core machines – Fast Intel processors – Large memory sizes, 32G to 64+ Giga bytes of memory per machine • Very fast internal networks – Min 1 Gigabit speed, moving to 10 Gigabits – Larger compute farms have Quadrics, Myrinet, Infiniband network fabrics • Large Storage requirements, minimum is Terabytes most at Petabyte levels
- 6. Image Rendering Ecosystem The Evolving Rendering Ecosystem • HD driving disruptive growth • Render Farms getting bigger • Extreme demands on network and storage • Render nodes switching from desktops/white boxes to servers, with Blade Servers being deployed more and more as render farms • Scalability, Reliability and Performance are real concerns as farms and processor densities grows • OS switching from Windows to Linux, with OSX becoming more popular • Demand for Quality Render Farm Management Software on the rise • High Availability, Auto-Fail Over, Remote Node Management Mandatory
- 7. HPC Rendering Farms Looks Like HPC To Us • Bigger and Bigger Machines, all now seem to be 8 core and now moving to 16 usable cores with the Intel Nehalem Chip set • More and more of these (Commodity off the shelf) COTS and Blade Machines • Fast Networks, 1 Gigabit or 10 GigE showing up in Hollywood studios more and more, not quite Quadrics network fabrics yet • Terabyte moving to Petabyte storage arrays 7
- 8. HPC Rendering Farms • Shrek (2001) – 525 processor server farm plus 300 company desktops – 5 million computer hours of render time needed • Shrek 2 (2004) – 1:45 in length – 20 Terabytes of data – 10 million computer hours of render time needed – Some frames took 30-40 hours to render – 3000-processor render farm of HP servers and 300+ HP workstations • Shrek 3 (2007) – 20 million computer hours of render time needed • DreamWorks = 4500 nodes, Disney in excess of this, smaller no-name studios in LA are running 2000 – 3000 nodes • Even with the advancements and speed in hardware for these Render / HPC compute farms the need for more render time still doubled with each Shrek movie reinforcing the need for resource management software (EnFuzion® ) to optimize the hardware/storage resources Source: The San Francisco Chronicle Source: Exodus Film Group HPC* = High Performance Computing
- 9. HPC Rendering Farms Transformers Stats • Bumblebee = approx 8,000 pieces/elements • Optimus Prime = Approx 10,000 pieces/elements • 80,000 compute/render hours to render 1 scene Source:9 Renee Dunlop, CG Society.org. 18 June 2009, Clare Kaufmann, http://education.yahoo.net/degrees/articles/digital_transformers.html Source: Matt Sullivan, www.popularmechanics.com/technology/industry/42/8826.html . Transformers: Bonus Material, Special Features. Dir. Michael Bay
- 10. Industry Direction Where Is The Industry Going? • More disruptive growth driving demands for HPC like systems and functionality inside studios large and small • Must be 24x7, becoming mission critical and driving demand for full redundancy of render farm Head Nodes, Compute Nodes and Cluster/Render Management software • Bigger and faster machines for rendering will then enable even more complex and grandiose visual effects within required project timelines • Single computer “Real-Time Rendering” solutions coming on the scene, not clear at all if this GPU based computing will displace the traditional CPU based render farms in use world-wide today
- 11. EnFuzion3D Resource Management • EnFuzion® – 11 Years of continuous deployment in Mission Critical environments • Supported 3D Animation since 1998, Plug-ins for 90% of 3D Applications – Heterogeneous robust Render Farm resource manager – Performance – fastest scheduler in the Industry – Scalability – Built to scale with minimum network load, (10,000s’ of CPU/Cores) – Drives all cores in multi-core machines on 32 and 64 bit processors (Windows and Linux) – Full OSX support, PowerPC (G4/G5) and Intel architectures for Mac OS (Tiger, Leopard) – Reliable 24x7 fault tolerance features – work around HW, SW, memory, network, renderer problems – OPEN ARCHITECTURE, Full API, scripting interface , extensive admin/ reporting tools • High Availability and Auto-Fail over options available, Remote Node Management • Support’s “green grid” initiative with WoL (Wake on Lan) and can shut down unused machines and rendering nodes when work done. • Web based Control Center , Web based Front End for “remote job submission” • Solid Customer Satisfaction
- 12. EnFuzion Resource Management How Does EnFuzion® Work
- 13. EnFuzion EC2 Plug-In for Amazon EC2 THE EVOLVED RENDER FARM = EnFuzion EC2 + Amazon EC2 Cloud EnFuzion EC2 EnFuzion®
- 14. On-Demand Rendering THE EVOLVED RENDER FARM “On Demand” Rendering EnFuzion EC2 EnFuzion EC2 EnFuzion® EnFuzion® RENDER FARM AMAZON EC2 CLOUD
- 15. EnFuzion High Availability – Auto Fail Over EnFuzion EnFuzion Submit host Submit host On Network/Hardware Failure EnFuzion EnFuzion Root host Root host IP: 66.124.53.150 IP: 66.124.53.152 On Network/Hardware Failure EnFuzion EnFuzion Compute Nodes Compute Nodes
- 17. Axceleon Snapshot • NEW YORK • Axceleon is a software company founded in 2002 with a focus on high performance computing, clustering, grid computing and resource management for render farms. HQ in Sunnyvale CA., presence in Los Angeles, China, Singapore, Tampa, New York, R&D center planned for Ireland in late 2009. • Customers in over 18 countries. • Hardware Channel Partner: IBM, HP, Dell, Sun and SGI • Key ISV Partners: Siemens, Novell, Red Hat, Microsoft, Intel, Autodesk, Apple, The Foundry, Maxon, and AMD • Engaged with select resellers and system integrators in the US, Japan, China, Taiwan and Europe that focus on clustering and the 3D animation market.
- 18. Thank you
- 19. EnFuzion Demos @ at Axceleon Booth 2224 • EnFuzion3D 2009 • EnFuzion Remote Desktop • EnFuzion EC2 with the elastic EC2 render farm • Thank you • Any questions? Footnotes are very small and need not be capitalized.