DeIC Watson Agreement - hvad betyder den for DeIC medlemmerne Preben Jacobsen Solution Architect Nordic Lead, Software Defined Infrastructure Group IBM Danmark 2014 IBM Corporation
Link: https://www.youtube.com/watch?v=_xcmh1lqb9i 2
DeIC IBM Watson Aftalen Hovedelementer i aftalen Membership of the Global IBM Academic Initiative program IBM Analytics Scheduling for BigData Watson Foundation and Compute Environments. IBM Application and Process Control for Compute Environments IBM Watson Foundation solutions for BigData IBM BigInsight 3
IBM Academic Program 4
Link: https://www.youtube.com/watch?v=-wvcnlqafnw 5
Data is the new basis of competitive value Oil & Gas Healthcare Banking Retail 6
IBM LSF Family IBM Analytics IBM Application Center IBM RTM IBM Process Manager IBM MPI LSF IBM Data Manager for LSF Hadoop Connector MapReduce Accelerator Docker Connector IBM Dynamic Cluster Scheduling Extensions IBM Session Scheduler IBM License Scheduler 7 Planned Products Elastic Storage (GPFS)
IBM LSF accelerates time to results Benchmarks designed to simulate workloads typically found in high throughput EDA, Life Sciences. Comparison with free schedulers at small scale (128 cores) LSF Delivers Superior throughout with consistent performance 150x TORQUE TORQUE is used for execution by Moab, and shares a similar architecture with PBSPro 12xOpen Grid Scheduler Open source fork of the Grid Engine codebase 6x SLURM Higher throughput = Quicker results: Time to Market More simulations in a given time: Better products 8
IBM Analytics Data Centre IBM Analytics LSF Cluster 1 LSF Cluster N Resources Users Workload Provides detailed insight into current and historic view: Compute Slots Groups Applications Capacity and Utilization Memory Departments Projects Workload Throughput Hardware Resources Business Units Multiple Lines of Business Usage or Resources Cost of Compute Application Licenses Parallel Workload Workload Efficiency Complex Workflows Pending Workload Wasted Resources 9
Example: Requested Memory 10
IBM Software Defined Infrastructure High Performance Analytics (Low Latency Parallel) Hadoop / Big Data High Performance Computing (Batch, Serial, MPI, Workflow) Application Frameworks (Long Running Services) Example Applications Homegrown Homegrown Workload Engines Symphony Symphony (MapReduce) LSF Application Service Controller Resource Management IBM Computing Scheduling & Acceleration with Infrastructure Sharing 11
IBM Software Defined Infrastructure High Performance Analytics (Low Latency Parallel) Hadoop / Big Data High Performance Computing (Batch, Serial, MPI, Workflow) Application Frameworks (Long Running Services) Other Commercial Applications Workload Engines Resource Management Symphony Symphony (MapReduc e) LSF IBM Computing Application Service Controller Other Management Software Data & Storage Management IBM Spectrum Storage Infrastructure & Cloud Management IBM Cluster Manager IBM Cloud Manager with OpenStack Flash Tape Disk Power x86 Linux on z docker VM 12 On-premise, On-cloud, Hybrid Infrastructure (heterogeneous distributed computing and storage environment) Support and Services
Scalability from regional to national or global Cluster A Cluster B Grid Cluster E Cluster C Cluster D Manage your assets as a single, virtual computer from anywhere in the world 13
Link: https://www.youtube.com/watch?v=jlcj4jqi3q8 14
Data is growing exponentially and demands new approaches (technology and strategy) 44 zettabytes unstructured data You are here structured data 2010 2020 15
16
ESS Scaling 17
ESS GNR Disk Failures = Rebuild = Low Impact 1 disk fails 3 disks fail Critical rebuild finished, continue normal background rebuild Normal rebuild <5 minutes critical rebuild Normal rebuild Data-oriented rebuild vs. Disk-oriented rebuild OK for multiple concurrent failures 1 or 2 drive failures per drawer in all drawers = near-zero impact Priority for critical rebuilds Then back to background rebuild at low priority = low impact Rapid recovery, even with rolling failures TIME 18
Link: https://www.youtube.com/watch?v=pe9kevtxgxm 19
Announcing the IBM Power Systems S812LC optimized for entry and small Hadoop workloads Improve the agility and reduce the cost of running Spark and Hadoop workloads 16X the memory capacity of Xeon E3 servers, 2X 1P Xeon E5 servers Complete the same Spark workloads for <½ the cost of Intel Xeon E5-2690 v3 systems 2.3X BETTER performance per dollar spent 94% more Spark workloads in the same rack space as Intel Xeon E5-2690 v3 systems 1.94X BETTER performance per system (10 core S812LC vs 24 core DL380) 1-socket, 2U Up to 10 cores (2.9-3.3Ghz) 1 TB Memory (32 DIMMs) 115GB/sec memory bandwidth 14 LFF (HDD/SSD) 84TB storage 4 PCIe slots, 2 CAPI enabled Default 3 year 9x5 warranty, 100% CRU 20
THE END 21