Bob Thome Senior Manager Grid Computing Enterprise Grid

Bob Thome Senior Manager, Grid Computing

Enterprise Grid Computing

The best thing about the Grid is that it is unstoppable. The Economist, June 21, 2001 2

Top 10 Grid Computing Lies 10. The grid will be unreliable because power grid failed last year 9. The grid is 5 years away 8. The grid is just for research and academic users 7. The grid requires multiple administrative domains 6. Al Gore invented the grid

Top 10 Grid Computing Lies 5. You need to rewrite your apps to take advantage of the grid 4. There is only one Grid 3. You need to move everything to the grid at once 2. Oracle 10 g is a grid in a box 1. The grid only runs on Power. Point

Problem with Today’s IT Infrastructure Ÿ Statically Assigned Islands of Computing Resource – – Some are melting down Some are almost idle Ÿ High Costs – – – Hardware Labor Software EMAIL ERP Ÿ Hard to Align with Business Priorities DW

Example

Example: In December Ÿ Order Entry maxes out processing orders Ÿ Financials is idling below capacity Order Entry Financials

Example: In January Ÿ Order Entry drops off from season high Ÿ Financials maxes out on year end close Order Entry Financials

What is Grid Computing? “In basic terms, grids are clusters of interconnected servers, enabling shared computing resources utilization” “Defining Grid Computing”, Giga Research, Agosto 2002

Grid Computing Vision Ÿ Computing as a utility – A network of clients and service providers Ÿ Client-side: Simplicity – Request computation or information and receive it Ÿ Server-side: Sophistication – – Availability, load balancing, utilization Information sharing, data management Ÿ Virtualization – – Clients see a large virtual server Underlying infrastructure hidden

Benefits of Grid Computing Ÿ Better information faster – – – Perform more work with fewer resources Spread work across resources Access to resources on demand Ÿ Faster response to changing business priorities – Instantly and dynamically realign IT resources as business needs change Ÿ Reduced IT costs – – Improve utilization of existing resources Utilize less expensive commodity platforms 8 Oracle Confidential

Technology Trends Ÿ Blades: Every vendor offering them – – – Huge cost advantages Software vendors have to enable usage Dell Power. Edge, HP Proliant BL, Sun Fire Blades, Fujitsu Primergy BL Ÿ Linux: Fastest growing OS – – – Commodity OS Ready for blades today Linux and blades naturally complement each other Ÿ NAS, SAN, and IB provide storage access from any blade 6

Grid Computing Evolution Outsourcing Enterprise Grids Desktop Computing Grids • Collaborate • Example: SETI@home Shared Server Grids • Share • Example: CERN • Dedicated Servers In a Data Center • Example: • Electronic Arts • Oracle Corp.

The Final Phase: Outsourcing Ÿ Problem: – – – Many apps are already standardized Replicating admin knowledge to administer standard components is not cost effective SMB does not have scale to realize grid benefits Ÿ Solution: – Buy the application as a service Ÿ Implementation: – – Available today from many vendors, especially for SMB Potentially explosive in under-automated economies and industries…remember cell phones? 14

Enterprise Grid Computing Ÿ Standardization – – Standard blade servers, Linux Fast interconnects for storage and network Ÿ Virtualization and provisioning – – Resources dynamically assigned Realign IT resources as business needs change Ÿ Scale out – Add additional resources to grow capability of system

Grid Computing Components Ÿ Storage Ÿ Database Servers Ÿ Application Servers Ÿ Provisioning and Management Tools

Align Storage with Business Ÿ Islands of storage – “My storage is underutilized and growing 50% a year”

Align Storage with Business Ÿ Islands of data – “My storage is 30% utilized and growing 50% a year” Ÿ Disk farms of industry standard disks – – Consolidate into SAN or NAS Provision as needed

Storage Grid Ÿ Oracle Automatic Storage Manager – – – Provisions storage capacity automatically to Oracle 10 g as needed Stripes and Balances I/O Mirrors: Immune to disk failure

Grid Computing Components Ÿ Storage Ÿ Database Servers Ÿ Application Servers Ÿ Provisioning and Management Tools

Align Processing with the Business Ÿ Islands of computation – “ 15% utilization of CPU is exceptional”

Align Processing with the Business Ÿ Islands of computation – “ 15% utilization of CPU is exceptional” Ÿ Standardize resources – – Blades provide lowest cost, highest performance Not Self-healing, Disposable Ÿ Share virtual resources Ÿ Provision resources as required Ÿ Scale out

Issues Ÿ Blades typically 1 -4 CPUs Ÿ Many databases require greater than 4 CPUs Ÿ Platform must scale to meet future/peak demand Ÿ Databases may require more memory or I/O than many blades provide

Solution Ÿ Run database workload across clusters of multiple blades – – Federated database Shared database

Federated Database Ÿ Partition large database into many small subsets Ÿ Provide a federated (union) view of all data Ÿ Strengths: scalable, extensible Ÿ Challenges: inflexible, limited application support, availability Federation Layer Data Subsets

Shared Database Ÿ Multiple blades access a single database Ÿ Any instance access any data Ÿ Strengths: High availability, broad application support, dynamic scalability Ÿ Challenges: Requires shared disk, fast interconnect Listener/ Balancer All Data Oracle Real Application Clusters

Databases on the Grid Ÿ Database clustering with shared disk – – – Low cost highest quality of service Scalability AND availability Ÿ Add/drop servers as needs change Ÿ Automatically balance load across servers Ÿ Proven – Hundreds of customers running enterprise applications

CPU Provisioning on Demand Shared database supports dynamic CPU provisioning Shared Database Federated Database Add blade while running Drop blade while running Dynamically allocate CPU Add blade, reload/repartition Repartition/reload, drop blade CPU allocation static All Data Subsets 21

CPU Provisioning on Demand Ÿ Quarter end sale on the website – Web site load serviced by blades Quarter ends, GL closes the books – – GL higher priority, add nodes Capacity on Demand Increase the allocated portion of the blade farm – Add blades or increase the sandbox Scale out automatically according to your priorities General Ledger Web Site

Example: In December Ÿ Order Entry maxes out processing orders Ÿ Financials is idling below capacity Order Entry Financials

Example: In January Ÿ Order Entry drops off from season high Ÿ Financials maxes out on year end close Order Entry Financials

Example: With Grid Computing Ÿ Load balance based on a policy to optimize around both of these peak load conditions Order Entry & Financials

Policy based CPU Provisioning Ÿ Specify service levels – – Response time CPU utilization Ÿ Monitor service levels Ÿ Automatically add/drop resources to meet service level objectives Ÿ Frees administrator from provisioning activities

Resonance Ÿ Automatically provision CPU between databases as loads change – – Completely automatic and policy driven Automatically add/drop instances servicing a RAC database Ÿ Load-based session management and migration – – – Automatically migrate sessions to rebalance workload across RAC instances Intelligently direct sessions to instances Service-based Ÿ Transparent to applications – No application code changes required

Demo

Provision Data Ÿ Move data to available cpu – – – Access on demand Replicate Move Ÿ Provision data in bulk or incrementally with Streams Ÿ Build a CPU rich analytic farm – – Provision data in for processing Maintain it or throw it away

Grid Computing Components Ÿ Storage Ÿ Database Servers Ÿ Application Servers Ÿ Provisioning and Management Tools

Application Server Grid Ÿ Complete, integrated application server clusters Ÿ End-to-end transparent application fail-over – Fast fault recovery in seconds Ÿ Application-specific load balancing policies – – Schedules Runtime metrics

Grid Computing Components Ÿ Storage Ÿ Database Servers Ÿ Application Servers Ÿ Provisioning and Management Tools

Management Tools Ÿ Enterprise Manager Grid Control – – – Manage sets of systems as one Application service level management Policy-based standardization Automated provisioning of Oracle components Automated administration

Provisioning Tools Ÿ Many third-parties (systems vendors) provide provisioning tools Ÿ Designed to manage an entire heterogeneous grid Ÿ Create virtual lans, clusters, and application sandboxes on demand Ÿ Must interoperate with applications and application specific provisioning infrastructure

Transition to Grid Computing Ÿ Start small – – – Move an application Get experience Establish standard components Create standard procedures and patterns Create “known good” configurations Continue moving things

Scale Out Ÿ When you run out of capacity, buy more – – Clone components Gain economies of scale Never make a big capital investment Never take a risk Ÿ Savings and flexibility increase as Grid grows

Enterprise Grid Computing Ÿ Enterprises can realize the benefits of grid computing now Ÿ New technologies make it easy – – Standardize on modular low-cost hardware components Pool resources across applications Provision resources as required Scale out to add resources

More Information Ÿ Grid on OTN – http: //otn. oracle. com/grid/