Fat Nodes & GPGPUs - Red-shifting your infrastructure without breaking the budget

Fat Nodes & GPGPUs - Red-shifting your infrastructure without breaking the budget

• 1.
  Red-shifting yourinfrastructure without breaking the budget – Fat Nodes & GPGPUs Irish Centre for High End Computing Dr. Eoin Brazil, Technology Transfer
• 2.
  Outline of PresentationRed-shift versus Blue-shift Scaling – Up or Out – Why not both ? The secret recipe and how to bake it Capacity planning Rock stars wanted Irish Centre for High End Computing (ICHEC) - Optimisation with GPGPUs
• 3.
  Red-Shift / Blue-shift Irish Centre for High End Computing (ICHEC) - Optimisation with GPGPUs Rapidly growing compute and / or data demand beyond Moore’s Law
• 4.
  Characteristics of a Red-shift Company Rapidly growing compute and / or data demand beyond Moore’s Law Not just Google or YouTube Irish examples include Paddy Power, Tullow Oil, MET Éireann, NewBay and many more! Irish Centre for High End Computing (ICHEC) - Optimisation with GPGPUs
• 5.
  Google – A Red-shift Company "Between 1998 and 2009 it is reported that Google's infrastructure scaled performance (queries processed per day) by 1,000 times while scaling its infrastructure by 1,000 times ." J. Dean, “Challenges in Building Large-Scale Information Retrieval Systems,” keynote talk, Proc. 2nd Ann. ACM Conf. Web Search and Data Mining (WSDM 09), ACM Press, 2009; http://wsdm2009.org/proceedings.php Irish Centre for High End Computing (ICHEC) - Optimisation with GPGPUs
• 6.
  YouTube – A Red-shift Problem Estimates that 24 hours of video are being uploaded per minute HD rate is 2 – 5 Mbps that equals about 45-75 terabytes of data per day Transcoding is a typical compute dependant task Irish Centre for High End Computing (ICHEC) - Optimisation with GPGPUs
• 7.
  Scaling – WhichWay ? Up – Bigger server or capability per server Out – More servers Hybrid – Do both NextIO, SuperMicro, Appro, etc. Irish Centre for High End Computing (ICHEC) - Optimisation with GPGPUs
• 8.
  Off the ShelfNextIO vCORE Extreme, 3u, 16 double-wide GPU’s and 8 server connections over x16 PCIe – 8 Tflops DP / 16 Tflops SP SuperMicro SuperServer 7046GT-TRF-FC407, 4u tower, 4 double-wide GPU’s Irish Centre for High End Computing (ICHEC) - Optimisation with GPGPUs
• 9.
  Big Data -Hadoop MapReduce Black Scholes option pricing uses Monte Carlo methods – MITHRA, ~2008/9 Cluster: 62 CPU Quad Core Cluster = 248 cores Machine: 4 CPU Quad Core Cluster (16 cores) + 2x 9800 GX2 GPUs (256 cores) = 272 cores ~2008/9, 1/10 the size and 508x speed up ~2011, x2 CPU versus x4 GPU improvement Irish Centre for High End Computing (ICHEC) - Optimisation with GPGPUs
• 10.
  Recipe for FATNodes Keep processing local Use dual 12 Core CPUs with 64GB Ram on each CPU (124 GB per node) Connect 10+ Fermi GPUs to each node Use SSDs (1 SSD @ 200K = 50 ~ 100 HDD in parallel on single PCIe card) Connect with 40Gb/s InfiniBand Irish Centre for High End Computing (ICHEC) - Optimisation with GPGPUs
• 11.
  Scaling & Optimising<~> Capacity Planning Irish Centre for High End Computing (ICHEC) - Optimisation with GPGPUs Benchmark real runs against Amdahl's Law and Neil J. Gunther’s Universal Scalability Law (USL) to give insight into future planning for system and/or infrastructure.
• 12.
  Irish Centre forHigh End Computing (ICHEC) - Optimisation with GPGPUs
• 13.
  Training on CUDAFor Academics: FREE, HPC, MPI, OpenMP and CUDA For Businesses: COMMERICAL RATES, all the same courses plus custom tailoring Consultancy NVIDIA Certified Courses Irish Centre for High End Computing (ICHEC) - Optimisation with GPGPUs
• 14.
  5 forthcoming newpositions at ICHEC http://www.ichec.ie/about_us/employment Email us: [email_address] Irish Centre for High End Computing (ICHEC) - Optimisation with GPGPUs
• 15.
  Acknowledgements Supported byScience Foundation Ireland under grant 08/HEC/I1450 and by HEA’s PRTLI-C4. GPGPU - Gilles Civario, ICHEC