Sunteți pe pagina 1din 85

<Insert Picture Here>

Oracle Real Application Clusters


Barb Lundhild, Product Manager, Oracle Real Application Clusters

The following is intended to outline our general product direction. It is intended for information purposes only, and may not be incorporated into any contract. It is not a commitment to deliver any material, code, or functionality, and should not be relied upon in making purchasing decisions. The development, release, and timing of any features or functionality described for Oracles products remains at the sole discretion of Oracle.

Agenda
What is RAC? RAC for Scalability RAC for High Availability Consolidation with Oracle Clusterware and RAC RAC as the Grid Infrastructure
<Insert Picture Here>

Real Application Clusters


Benefits Highest Availability On-demand flexible scalability Lower computing costs World record performance

Database

Storage

RAC Adoption Rate Growing


13,000 12,000 11,000 10,000 9,000 8,000 7,000 6,000 5,000 4,000 3,000 2,000 1,000 0

Source: Oracle Internal Reporting, FY07

F Y 02 F Y 03 F Y 04 F Y 05 F Y F Y 07 F Y 08 F Y 09 06 (Q 2)

RAC: The Cluster Database


Network

Centralized Management Console Interconnect High Speed Switch or Interconnect

Users
No Single Point Of Failure

Clustered Database Servers

e hared Cach S
Storage Area Network Drive and Exploit Industry Advances in Clustering

Hub or Switch Fabric Mirrored Disk Subsystem

Oracle RAC Architecture


public network

Node1

VIP1 Service Listener instance 1 ASM


cluster

VIPn Service Listener instance n ASM

Node n

Oracle Clusterware interconnect Oracle Clusterware

Operating System

Operating System

shared storage

Managed by ASM RAW Devices

Redo / Archive logs all instances Database / Control files OCR and Voting Disks

Automatic Storage Management


Simplify and Automate Database Storage management Fraction of the time is needed to manage database files Increase Storage Utilization Eliminate over provisioning and maximize storage resource utilization Predictably Delivers on Service Level Agreements Never get out of tune delivering higher performance than RAW & File System over time Uncompromized availability empowering low cost storage deployment reliably

Automatic Storage Management


Eliminates need for conventional file system and volume manager Capacity on demand
Add/drop disks online

Automatic I/O load balancing


Stripes data across disks to balance load Best I/O throughput

Automatic mirroring Easy

Automatic Storage Management


Whats New in 11g
Fast Disk Re-sync changes tracked when disk goes offline, only relevant changes syncd when back online Preferred Mirror Read - Perform reads from the closest disk, even if it is not the primary mirror side Fast VLDB Start - For large files, automatically use larger extent sizes (memory consumption reduced by factor of 16) ASM Rolling Migration - Allow instances in an ASM cluster to be upgraded independently. (Some operations e.g. rebalance are restricted during rolling upgrade)

<Insert Picture Here>

RAC for Scalability

Real Application Clusters


Flexible Scalability for Applications
Application scalability by load balancing work across cluster As workload increases, easily add new node(s) to the cluster Cache Fusion performance improvements with 11g Batch
Order Entry Supply Chain

End users are reporting that Oracle's Real Application Clusters and Automatic Storage Management efficiently leverage commodity-class servers and storage respectively.
- MarkA.Beyer Gartner December 15th 2006

http://mediaproducts.gartner.com/reprints/oracle/144742.html

Oracle Real Application Clusters has been moved to the mainstream and is providing significant advantages to customers.

- Donna Scott, Donald Feinberg Gartner 6 February 2009

http://mediaproducts.gartner.com/reprints/oracle/article61/article61.html

RAC Provisioning
Simple, Accurate, Reliable Capacity on Demand

Automate RAC provisioning while providing service


Convert DB to RAC One-click-extend RAC to meet increased demand De-provision RAC nodes

Simplified way to scale up or down RAC


Integrated provisioning of RAC, ASM, and Clusterware Bare metal provisioning integrated with RAC provisioning

Best on Oracle

Gold image based initial RAC and Clusterware provisioning Flexible and Adaptable to your IT environment

RAC Deployment Cycle


Test users

Testing and Staging

1 Create reference RAC systems


Production

2 Stage gold images

1 5 Scale down 4 Scale up RAC cluster 3


Create Production clusters

ADDM for RAC

Database-Level ADDM 11g


Self-Diagnostic Engine

New in 11g

Performance expert in a box Identify the most Globally Significant performance problems for the entire RAC cluster database Database-wide analysis of:
Global cache interconnect issues Global resource contention, e.g. IO bandwidth, hot blocks Globally high-load SQL Skew in instance response times

Instance-Level ADDM

Inst 1 AWR 1

Inst 2 AWR 2

Inst 3 AWR 3

Runs proactively every hour when taking AWR snapshots (default)

SAP SD and SAP SD Parallel Standard Benchmark


Near Perfect Scaling across SMP and Cluster
5 Node RAC
37,040

40,000 35,000 30,000 3 Node RAC SD Users 25,000 20,000 15,000 10,000 Single Node SMP 5,000 0 4
4,010 2,035 22,416

4 Node RAC
30,016

2 Node RAC
15,520

8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80 # of CPU Cores

These results, as of March 25, 2008, have been certified by SAP AG, www.sap.com/benchmark. Please see notes page for benchmark certification details for the above results.

Cost Comparison: SMP vs. Grid


Telecom Workload 550,000 transactions/hour

Millions $8 $7 $6 $5 $4 $3 $2 $1 $0
Unix 72 CPU SMP

Server Network Storage Infrastructure Software

Ten 2-cpu Dell PE1750s (Xeon)

Four 4-cpu Dell PE7250s (Itanium)

Call Data Records Architecture


Using CDRLive from LGR Telecommunications

Call Data Records Architecture


Using CDRLive from LGR Telecommunications
System Usage

Call Data Records Architecture


Using CDRLive from LGR Telecommunications
LGR has done extensive benchmark testing of nearly all the major DW solutions available. Chose to standardize on Oracle databases nearly a decade ago because:
Proven ability to handle high data volumes Highly scalable (50 500 TB) Rapid query response times on hundreds of TB Familiar, easy-to-use for customers High level of support globally Low TCO allowing for a rapid ROI

Low Cost Servers - HP TPC-C Benchmark


1,200,000 1,000,000 800,000 600,000
TPM-C $7.9 M Total Cost $2.6 M Total Cost 1,008,144 1,184,893

400,000 200,000 0

1x64

16x4

Same 1.5 GHz Itanium2 CPUs


Source http://www.tpc.org - Audited Results

TPC-H Three Terabyte Best Performance Results by Database


as of June 8, 2006

100,000 80,000

110,577

QphH

60,000 40,000 20,000 0

54,466 30,957

Source: www.tpc.org/tpch

Oracle RAC based World-Record SpecJ2004


SpecJ2004 at IR=5700 (9,459 JOPs) 11 mid-tier blades (2P/4C Montvale itanium) and 2 RAC DB nodes, each with 40 cores Oracle Application Server 10g Release 10.1.3.3.2 and Oracle RAC 10g Release 10.2.0.3 using Automatic Storage Management
Application Server

Database

Storage

http://www.spec.org/jAppServer2004/results/res2007q4/jAppServer2004-20071106-00093.html

Automatic Workload Management


Services and Connection Load Balancing

LISTENER
Service RAC? RAC1 on N1 RAC2 on N2
Listeners
Ne ork tw

RAC3 on N3
RAC Database

Clients

Automatic Workload Management


Load Balancing Advisory
Load Balancing Advisory is an advisory for balancing work across RAC instances. Load balancing advice Is available to ALL applications that send work. Directs work to where services are executing well and resources are available. Adjusts distribution for different power nodes, different priority and shape workloads, changing demand. Stops sending work to slow, hung, failed nodes early.

Automatic Workload Management


Runtime Connection Load Balancing Supported by JDBC, ODP.NET, OCI Client connection pool is integrated with RAC load balancing advisory When application does getConnection, the connection given is the one that will provide the best service. Policy defined by setting GOAL on Service Need to have Connection Load Balancing
New in 11g

PQ integration with Services

New in 11g

Parallel Query slaves will only execute on nodes where the service of the query owner is active. No longer have to code instance_groups
Node-1 Node-2 Node-3 Node-4 Node-5 Node-6

OLTP1 OLTP2 OLTP3 OLTP4


Report ing Batch

DW

Load Balancing Advisory


Load Balancing Advisory is an advisory for balancing work across RAC instances. Load balancing advice
Is available to ALL applications that send work. Directs work to where services are executing well and resources are available. Adjusts distribution for different power nodes, different priority and shape workloads, changing demand. Stops sending work to slow, hung, failed nodes early.

Client connection pool is integrated with RAC load balancing advisory When application does getConnection, the connection given is the one that will provide the best service.

Improved XA Support with RAC

New in 11g

Tightly coupled Global txns can span RAC instances CLUSTERWIDE_DISTRIBUTED_TRANSACTIONS, which is TRUE by default Allows the units of work performed across these Oracle RAC instances to share resources and act as a single transaction I.E. tightly coupled 2PC requests can be sent to any node in the cluster

Mercado Libre
eBay in Latin America Runs marketplace on RAC Scaled incrementally as marketplace grew
Business Volume 2 0 ,0 0 ,0 0 0 1 0 ,0 0 ,5 0 0 1 0 ,0 0 ,0 0 0 50 0 0 ,0 0 0

2004

2005

2006

2007

Nodes

Mercado Libre
Load Distribution Marketplace Searches, Listings, Bidding, Reporting Customer Service

Newer nodes are faster and have more memory Marketplace workload is not partitioned

The size, complexity and varying needs of our customers require highly scalable, reliable and available platforms on which to run core applications. Therefore, when we looked to expand PolicyWriters platform compatibility, Oracle DB and RAC was the logical choice based on its capabilities and reputation in the market.
John Pettit, President and CEO Adaptik Corp

http://www.oracle.com/corporate/press/2008_apr/migrations.html

Eugene Park Senior Director of Application Services PG&E

With Oracle RAC, the utility can use smaller, more-efficient servers that offer excellent performance and scalability while handling 720 times the amount of data. Much of what we've accomplished we attribute specifically to Oracle RAC. Instead of scaling up a massive mainframe, we've distributed across smaller, more energy-efficient systems.

Leading Internet Brand, Most Trafficked Internet Destination


Largest online network of integrated services Provides Internet services to users and businesses worldwide

Needs:
Easier scaling with standard configuration Reduce costs - Migrate From Big Iron SMP systems to Commodity Hardware w/Linux Minimize down time and provide fail back option

Migrated to standard RAC configuration


2002 Started with RAC 9.2, 32-bit Linux, 4 nodes Today RAC 10.2.0.2 64-bit RHEL 4, up to 10 nodes/cluster

Products/projects are supported by over 100! dedicated RAC clusters

Amazon Data Warehouse Superstack 3rd Cluster


Having a 3rd copy allows us to buy 1.5x the amount of processing power instead of 2x. We replace the oldest cluster every year to keep up with growing requirements 1. Extract from source systems 2. Integrate, transform, denormalize, query and analyze 4. Data access and publishing

Extract Servers
17nodes, 8CPUs, 16-32 GB RAM, 100+ TB, 1Gb IC

ETL Manager Amazon S3


Amazon DCS 16 nodes, 16 CPUs, 64GB RAM, 100+ TB, IB IC

"At 22 000 GB the Trace Archive is in the Top Ten UNIX databases in the world. That's not bad for a research organisation of 850 employees in the countryside just outside Cambridge." "It is possibly the biggest single (acknowledged) scientific RDBMS database in Europe, if not the world."
Martin Widlake, Database Services Manager Wellcome Trust Sanger Institute
The Winter Corporation database survey 2005 suggests the Trace Archive would rank fifth behind such giants as AT&T, Yahoo and other large international corporations.

<Insert Picture Here>

RAC for High Availability

High Availability
Removes Server as Single Point of Failure
Integral part of Oracles Maximum Availability Architecture Enables applications to mask component failures in infrastructure
Order Entry Spare Supply Chain

Server Failure

Automatic Workload Management


Service fails over to surviving Node

Order Entry

Supply Chain

Re-allocate spare to Order Entry

Oracle: High Availability Without the High Cost


Traditionally High Quality = High Cost
High quality systems were built by combining high quality, high cost parts Mainframe model

Oracle enables a new model: High Quality + Low Cost


Oracles vision is to attain the highest possible availability using software that can also easily scale out to meet additional load requirements Fully integrated and simple to manage
Enable widespread deployment of HA configurations

Causes of Downtime
System Failures

Unplanned Downtime

Storage Failures Human Errors Corruptions Site Failures

Data Failures System Changes Data Changes

Planned Downtime

Oracles Integrated HA Solution Set


System Failures
Real Application Clusters

Oracle MAA Best Practices

Unplanned Downtime

Data Failures

ASM Flashback RMAN & Oracle Secure Backup H.A.R.D Data Guard Streams Online Reconfiguration Rolling Upgrades Online Redefinition

Planned Downtime

System Changes Data Changes

Fast Connection Failover


Supports multiple connection caches Datasource for each cache mapped to a service Keeps track of service and instance for each connection Cleans up connections when failures occur Distributes new work requests across available instances
Mid-Tier
CACHES SERVICE 1 SERVICE 2 SERVICE 3 INST Z INST X

Database Tier

INST Y

JDBC, ODP.NET, OCI

OrderEntryBus

A P IC T NT D Y P L A IO O A
TradingApplication

PoolA
FAN

PoolC

PoolB
ServiceBServiceC

Connection Cache

ServiceAServiceC

2nodeRAC

ASM

Notification Callouts
Users can write callout programs that are invoked when notifications occur
Notifications include node up/down, instance up/down, or service up/down

Example usages:
Send email, page Log status info Start / stop programs

RAC Patching
Simple, Accurate, Reliable Maintenance of DB and RAC

Automate patching while continuing to provide service Comprehensive support for all types of patches Adaptable to your IT environment Reduce possibility of errors and improve reliability

Best on Oracle

EM driven RAC Rolling Patching

1 Both nodes running 2


Node 1 shut down/patched/started

EM

3 Nodes in mixed mode

Node 2 shut down & patched

Amtrust

formerly

One of the nations largest mortgage lenders Needed a continuously available scalable infrastructure to replace stressed legacy systems that supported a .NET development environment Full payback of investment in bank call center in 17 months, 165% ROI in five years Achieved internal rate of return (IRR) of 38% Availability, scalability, and performance improved
Server/OS maintenance performed online in rolling manner Automatic Workload Management enables individual management and control of workloads

What, The Saraswat Co-op Bank is Saying

We believe very strongly that Oracle offered the most robust and reliable system we evaluated. We decided that whatever system we acquired needed to last three to five years as a minimum. Oracle 10g and the chance to adopt grid computing meant Oracle could clearly address that scalability requirement.
Samir K Banerji, Managing Director, The Saraswat Co-op Bank

What SK Telecom is Saying


We successfully completed a Big Bang migration from a mainframe system to a clustered server environment. The Oracle environment is highly stable and offers powerful performance and easy scalability, which has enhanced user satisfaction and customer service.
NoCheol Park, CIO & Senior VP, NGM Program Manager, SK Telecom

<Insert Picture Here>

Consolidation with Oracle Clusterware and RAC

Cluster Consolidation
listener database cluster database 2 ORACLE_HOME ORA_CRS_HOME

Protected App A listener cluster database ORACLE_HOME ORA_CRS_HOME

Protected App B listener cluster database ORACLE_HOME ORA_CRS_HOME

Protected App C listener cluster database 2 ORACLE_HOME ORA_CRS_HOME

Cn o a dP o o S ra ew o s lid te o l f to g ith A to a S ra eMn g mn u mtic to g a a e e t

Be careful to size your servers to fit the required capacity.

Oracle Clusterware Licensing


See Oracle Database Licensing Information B14199-01

At a minimum, one of the machines involved in the cluster must be licensed using the appropriate metric for any Oracle Server Technology product in either its Standard Edition or Enterprise Edition, or all of the machines in the cluster must be supported via Oracle Unbreakable Linux support. A cluster is defined to include all the machines that share the same Oracle Cluster Registry (OCR) and Voting Disk

Consolidation to RAC
Node-1 Node-2 Node-3 Node-4 Node-5 Node-6

OLTP1 OLTP2 OLTP3 OLTP4


Report ing Batch

DW

formerly

Consolidated Clusters
St a b nd y

St

b nd

8 SKU databases consolidated into two 18 node Linux clusters. ASM provides pooled storage Database consolidation next

Consolidated Clusters
St a b nd y

St

b nd

Instances are brought up on nodes as workload or maintenance requires.

Consolidated Clusters

In case of outage, Data Guard recovers, and work is adjusted over remaining nodes

Migrated from decentralized Sybase system to a central Oracle 9i database Upgraded to Oracle RAC 10g with ASM, Grid Control, and Data Guard Improved reliability and removed single point of failure Provides rapid growth and scalability with reduced cost (hardware, support, and staff) Ease of Management with integrated Oracle solution Able to provide faster response to internal support requests with Enterprise Manager 10g

<Insert Picture Here>

RAC as the Grid Infrastructure

Plamen Zyumbyulev Head of Database Administration M-Tel

The grid capabilities of Oracle Database 11g have enabled us to consolidate several databases and run mix workloads in a performant and scalable environment.

Fernando Gaviria Lareo Team Application Leader Pago Facil

Oracle Database 11g enabled us to achieve a steady and fast growing platform, lower our costs and optimize the responsiveness to our customers by 50 percent.

Data Center 1

Oracle Real Application Clusters Standard Building Block Benefits from Oracle RAC:
# DB n

# DB

# DB

Easier Scaling - testing shows linear scalability from 4 to 6 to 8 nodes Better performance, higher throughput Reduced costs through using commodity hardware with Linux Standard RAC Cluster building blocks enables faster response to customer opportunities

G EM

Supplier to Business & Home retail space MegaGrid Concept


Many Servers in One Grid Many Cluster Databases Drives Consolidation, Scalability & Flexibility at Dell IT Add Capacity on Demand to individual Cluster Databases

Dell on Dell

European Order Management System


Oracle 11i EBS 10,000 TPS, 3 TB Already scaled up to meet additional demand

Midtier

11i Web/Forms/PCP

11i PCP

12 node Dell Power Edge 6650 (2 way IA32)

DB

Batch Concurrent Processing


10 node Dell Power Edge 6650 (4 way IA32)

DW

Grid Architecture
Corporate DW (9i + raw) SAP SEM/BI (9i + ocfsv1) Electricity Dispat. (10g + asm) SAP ISU (10g + asm) Siebel Brazil (10g + asm)
Provisioning Between Systems

Combined OLTP & DW

Production Servers

Development/Test

Siebel Brazil planned production Feb 07, SAP ISU Mid 07 Provisioning Nodes is Future Plans

Siebel Scalability with Oracle RAC 10g


Oracle/Siebel benchmark
80% scalability 80% scalability

8000 7000 6000 5000 4000 3000 2000 1000 0 1-node 2-node 4-node # of users
80% scalability 80% scalability

http://download.oracle.com/oowsf2004/1635_wp.pdf

<Insert Picture Here>

IPD/OS

Oracle Problem Detection Tool (IPD/OS)


Real-time Monitor

IPD/OS Node View

Node Eviction Diagnosis


Root Cause Analysis
Identify Problems
Experience Reboot Review Logs Call Support Escalate When it Reoccurs

Identify and Fix Problems


Monitor in Real-Time Replay State of Node Determine Root Cause Fix the Problem

Root Cause Analysis Node Eviction


Look in Oracle Clusterware CSSD log file (ocssd.log) for the time the node eviction occurred Run IPD in Historical Mode
$ crfgui -d "<HH>:<MM> Where "<HH>:<MM> is the time (hours and minutes since the eviction occurred)

Example: It is now 10:00 AM and node 1 was evicted at 09:45 AM, then you would enter:
$ crfgui -d 00:16

This will show you what the cluster looked like 1 minute before the eviction and you can replay the last minute

Download the Tool

otn.oracle.com/rac or
http://www.oracle.com/technology/products/database/clustering/ipd_download_homepage.html

For 32-bit and 64-bit Linux and Windows

<Insert Picture Here>

Adoption of Oracle RAC

Industry Dominance with ISVs deploying RAC


Fin Services Healthcare
Manufacturing

Retail

Telco

Government

Utilities

Life Sciences

Transportation

Education

Over 10,000 RAC Customers


in every industry
Financial Services Retail & Consumer Products

Manufacturing & Transportation

Technology & Communications

Government & Public Sector

Healthcare

Proven with ISV Partners


Vertical (and Horizontal) Solutions
Financial Services Retail & Consumer Products

Life Sciences

Technology & Communications

Government & Public Sector

Healthcare

Q & A
http://otn.oracle.com/rac

QUESTIONS ANSWERS

For More Information

http://search.oracle.com
Real Application Clusters

or oracle.com/clusters

Real Application Clusters (RAC)w SAP

Active

Active

Oracle Instance Clusterware , OS

Oracle Instance Clusterware, OS

Oracle Instance Clusterware, OS

Oracle Instance Clusterware, OS

World Record Oracle 10g Parallel SD Benchmark


40,000 35,000 30,000 25,000 20,000

sr es U DS

15,000 10,000 5,000 0 1 node 2 nodes 3 nodes


Scalability: 1.9

4 nodes

5 nodes

Scalability: 1.9

These results, as of March 25, 2008, have been certified by SAP AG, www.sap.com/benchmark. Please see notes page for benchmark certification details for the above results.

http://www.oracle.com/corporate/press/2008_mar/sap-sd-ibmp570.html

S-ar putea să vă placă și