X 86 Vision Product Line Bricks Blades Virtualisation
X 86 Vision Product Line, Bricks, Blades, Virtualisation, Energy Efficiency Tikiri Wanduragala Snr. Consultant Server Systems
Overview • Event Driven World • Centralised computing – SW Model – Server Consolidation – Desktop Consolidation • Scale up and Scale out • Market driven systems – Volumes – Rapid Evolution of Technology • Virtualisation Revolution in x 86 market • Energy efficiency
Innovation and Technical Leadership - The Server Portfolio System i 5™ System x™ System p 5™ Blade. Center™ Enterprise Storage Server™ System z 9 ™ Customer value Autonomic Computing Work. Load Mgr, Virtualization, Partitioning, Security, Systems Mgmt Series Unique Technologies z/OS, AIX 5 L™, OS/400, Windows Operating System, FICON Express Shared Components Blade. Center, Linux, Processors, I/O Power, Hardware Console, Adapters, Switches, Power/Mechanical frames § IBM's best technology § Shared innovation § Faster servers § Improved availability § Faster to market § Investment leverage Source: x. Series Linux Marketing Management & Strategy
Virtualisation – The Killer Application • “Build your own server” • Increase workloads by combining applications • Reduce physical population of servers • Reduce management by automating tasks • SW server – break link with HW • Efficient Energy Utilisation
Virtualisation – Path • Virtualise Server Apps • Virtualise I/O / Storage – IBM Open Fabric Manager SW • Virtualise Desktops – Security – Service / Support – Energy Efficiency 150 W -> 15 W
“a server or desktop is a file”
Combining Bricks & Blades A Smart, Flexible, Energy Efficient - Computing Platform Management & Control Storage Backup IBM Advantage • Blades • Stable Compatible Proven Platform since 2002 • Choice • Processors • Switches • Chassis • Energy Efficiency IBM Advantage Network • Bricks • Industry Leading Scalability • Industry Leading Performance • Optimised Memory System • Virtualisation ready • Proven Platform since 2001 • Energy Efficiency Brick Blade Combination No Single Point of Failure Energy & Systems Management Scale-up Scale-out Investment/Technology Protection
Portfolio that Covers the Spectrum of Business Needs Enterprise e. X 4 i. Data. Plex Server Consolidation, Virtualization Web 2. 0, HPC, Grid Scale Up Blade. Center Infrastructure Simplification, Application Serving Scale Out IBM give you the RIGHT choice for your business needs • Scale Up Enterprise e. X 4 Rack Servers for Consolidation , Virtualization and Database Serving • Scale Out Blade. Center for Infrastructure Simplification and Application Serving • Scale Out i. Data. Plex for Internet Scale Applications.
Scale up / SMP computing Large symmetrical multiprocessing (SMP) ‘Bricks’ Platforms for Virtualisation Blade. Center T Blade. Center H ‘Blades’ ‘Classic Servers’ x 3755 HS 21 Scale out / distributed computing Source: IDC LS 21
IBM System x 3950 M 2 The Next Generation of X-Architecture ØIntel® Quad Core Xeon processor for greater peak protection ØOutstanding performance leveraging both the advancements of the X 4 Architecture and the new Intel® Core. TM Microarchitecture ØEasy to expand scale with Scalability Wizard ØGreater Memory support in junction with VMware 3. 1 128 GB memory support ØSecure design with the onboard Trusted Platform Module (TPM) that ØDelivers embedded virtualization with the new VMware ESX 3 i on board ENTERPRISE
IBM System x 3950 M 2 – Server Consolidation Industry Leading Virtualisation Platform "This is the game changer here: an IBM data-class machine with four quad -core processors. . . running an embedded hypervisor. Mark Schlack vice president of Editorial, Storage and Windows Media Groups, Tech. Target Inc. ENTERPRISE
Public Internet/ Intranet Clients Step 1 Consolidate Servers Firewalls Routers (Layer 3 Switches) Layer 4 -7 Switches SSL Appliances Clients Layer 2 Switches ERP Servers App Web F/P SCM Gate WS ERP Sec Caching Appliances App Sec SCM Web Application Servers Security Servers SCM Servers Web Servers Storage Fibre Switches SAN Gate Security Gateway F/P WS File/Print Servers Web. Sphere Application Servers
Public Internet/ Intranet Clients Step 2 Integrate First Layer of the Network (L 2) Firewalls Layer 4 -7 Switches SSL Appliances Layer 2 Switches Caching Appliances Web App Gate F/P ERP WS Storage Fibre Switches SAN Routers (Layer 3 Switches) Caching Appliances SCM Sec
Public Internet/ Intranet Clients Step 3 Integrate Storage Fabric Firewalls Layer 4 -7 Switches SSL Appliances Routers (Layer 3 Switches) SSL Appliances Caching Appliances Web App Gate F/P ERP WS Storage Fibre Switches SAN SCM Sec
Public Internet/ Intranet Clients Step 4 Integrate Second Layer of the Network (L 4 -7) Firewalls Layer 4 -7 Switches SSL Appliances Routers (Layer 3 Switches) SSL Appliances Caching Appliances Web App Gate F/P ERP WS SAN SCM Sec
Step 5 Consolidate Applications Public Internet/ Intranet Clients Firewalls Routers (Layer 3 Switches) SSL Appliances Web App Gate F/P ERP WS SCM Sec Caching Appliances SAN
Result Step 6 Blade. Center Consolidate Clients Collapses Complexity: Public Internet/ Intranet Clients Firewalls ERP WS Sec ERP Gate F/P WS Gate Web App SCM F/P Web App Reduce costs 1/4 Reduce power costs 1/3 Reduce floor space 1/2 Integrate into 1 SAN SCM Sec Routers (Layer 3 Switches)
IBM Blade. Center – 5 Reasons 1. No 2. Single 3. Point 4. Of 5. Failure
IBM Blade. Center® Extend blade benefits to your entire business A common set of blades A common set of industry-standard switches and I/O fabrics A common management infrastructure IBM Blade. Center E Best energy efficiency, best density IBM Blade. Center S IBM Blade. Center T Ruggedized IBM Blade. Center H Distributed, small office, easy to configure IBM Blade. Center HT High performance Ruggedized, high performance HS 21 XM LS 21 LS 41 HC 10 QS 21 JS 12 Proc: Xeon Quad Core Mem: 4 DIMM / 16 GB HDD: 2 Proc: Xeon Quad Core Mem: 8 DIMM / 32 GB HDD: 1 Proc: Opteron Dual Core Mem: 16 DIMM / 64 GB HDD: 2 Proc: Conroe dual core Mem: 4 DIMM/8 GB HDD: 1 Proc: Cell BE Mem: 2 GB (XDR) HDD: 1 Proc: Power. PC Mem: 4 DIMM / 16 GB HDD: 2 RIGHT © 2006 IBM Corporation
IBM Blade. Center S All-in-one gets you up and running fast • Integrated business-in-a box foundation with configurable shared storage • Big IT results even from the smallest IT staffs to deliver big IT results • Easy with “select-and-click” configurability • Office-friendly 240 v power • Grows with your business • Optimised for small office environments RIGHT
Blade Deployment Made Easy Open Fabric Manager § Software utility in Advanced Management Module q MAC, WWN address assignment and management for initial blade deployment and re -deployment § Optional standalone utility for advanced functions q MAC, WWN Address assignment and management for initial blade deployment and re -deployment q Creates blade failover pools – Event Action Plans q Provides I/O parameter and VLAN migration in case of failover § Embedded switch enhancements q Easy to use GUI for switch configuration q Switch stacking for network simplification IBM Blade. Center Open Fabric
Open Fabric Manager in AMM simplifies server deployment and failover Blade. Center MAC 1 MAC 4 3 MAC 2 Management Module VLAN A Blade 1 VLAN B Blade 2 VLAN C VLAN D Blade 14 Any switch or pass through Ethernet MAC addresses are assigned to blade slot by AMM
Open Fabric Manager in AMM simplifies server deployment and failover Management Module Blade 1 VLAN A MAC 2 MAC 1 MAC 34 Blade. Center MAC 4 Spare Blade 2 Blade MAC 3 VLAN B VLAN C VLAN D Blade 14 Any switch or pass through New blade inherits I/O addresses move to new blade assigned to slot
Manage failover across your data center 2) Failover pool event action plan executes 1) Blade X fails in chassis SIX 3) Blade Y in chassis FORTY receives blade X I/O parameters
The real picture from data center input to usage: Where are the W consumed ? Servers Hardware Data Center Server Loads Typical Utilization Mainframe 80 – 90% Unix 10 – 20% Wintel 5 – 12% AC/DC losses; 25% HVAC, UPS 40% Fans; 9% Used Resource Idle; 20% = 3, 6% of total 80% IT power 60% Processor DC/DC losses; 10% 30% Standby; 2% Drives; 6% PCI; 3% Planar; 4% Memory; 11% Need one more W of compute energy ? 27 W data center X 1, 7 16 W equiv IT power SPL Sites & Facilities GTS STG Lab Services GTS DC Design – Southbank Ground Floor Lab x 3 5 W equiv processor x 5 +1 W equiv. used ressource STG Systems Design
• Power components: – Active power – Passive power • Sub-threshold leakage (source -drain leakage) – Gate leakage Power Density (W/cm 2) Passive Power Continues to Explode Gate Leakage Gate Length (microns)
Multi-Threaded, Multi-Core Chips Are Enabling Power-Efficient, High-Performance Processors • • Use multiple cores at lower frequency to speed-up parallel part of a program, and single core at higher frequency for sequential parts, to minimize power consumption Multi-core exploitation Single core Variety of techniques are becoming available to exploit multi-threaded and multi-core processors – – • MIPS/Watt Enhancements to compilers and operating systems Extensions to and new programming languages Application optimization at chip level will be achieved with different size cores, cores with dedicated SW functionality and accelerators MIPS Scale-out Scale-up C C C C C L 2 C CLarg e Core C C A A C C C A A A Medium Small Core Accelerator
What’s Using the Power? • The processor power growth is the largest single contributor but there are many other areas - the more you pack into a server the more power it needs! Blade. Center helps in this area OTHER? • AC to DC Transitions • DC to DC Deliveries • Fans and air movement Page 28 March 11, 2005 IBM Confidential Low Voltage processors help in this area
Simple, powerful and easy power management Compare actual vs. name plate power at system level View inlet and exhaust temperature Track heat emitted Compare rack actual power vs. label power Trend power use over time Trend temperature over time GREEN
Power, Can Virtualization Help? • Typical Intel type server utilization is quite low (15 - 40%) • Tools like VMWare can increase utilization and unlock new processing capability for scale up and scale out without adding to power at the rack level 2500 AC watts 2000 1500 1000 500 0 Idle 25% 50% 75% Max
Power and Cooling Financial Services – High Density Computing Deployment Client requirements § High density blade deployment for compute intensive financial applications § Limited floor space and limited power availability § High availability and resiliency § Cost and energy efficient Solution § 2, 100 blades in 1100 m 2 § IBM Integrated Rack Solution for High Density § Integrated networking, power management and environmental monitoring Benefits § Centre enclosed cold aisle layout enables 6 k. W per m 2 or 25 k. W/rack § Significant reduction in data center size and number of air conditioning units – cooling efficiency increased by 40 -60% § Reduction of 15 -25% in space needed for cabling; improves air flow distribution
HYBRID supercomputing x 86 Linux master cluster Cell/B. E. accelerator cluster Targeting 1. 4 petaflop peak, 1. 0 petaflop sustained performance
Combining Bricks & Blades A Smart, Flexible, Energy Efficient - Computing Platform Management & Control Storage Backup IBM Advantage • Blades • Stable Compatible Proven Platform since 2002 • Choice • Processors • Switches • Chassis • Energy Efficiency IBM Advantage Network • Bricks • Industry Leading Scalability • Industry Leading Performance • Optimised Memory System • Virtualisation ready • Proven Platform since 2001 • Energy Efficiency Brick Blade Combination No Single Point of Failure Energy & Systems Management Scale-up Scale-out Investment/Technology Protection
- Slides: 33