Oracle Real Application Clusters

10 - the loundation or
Lnterprise Grid Computing

.v Oracte !bite Paer
etevber 200:





OracIe ReaI AppIication CIusters 10 - the Foundation for Grid Computing Page 2
Oracle Real Application Clusters 10 - the
loundation or Lnterprise Grid Computing

ORACLE REAL APPLICATION CLUSTERS - FUNDAMENTAL TO
ACHIEVING ENTERPRISE GRID COMPUTING BENEFITS

In the past ew years, there hae been a number o signiicant improements in
computing hardware, networking, chip architectures, data storage systems and in
standards-based open-source operating systems. Oracle is the irst endor that
oers a complete enterprise computing sotware inrastructure that leerages all o
these improements to delier signiicant economic return on inestment through
better aailability, scalability, lexibility, and manageability. Oracle Real Application
Clusters ,RAC, is a key element in this unique, popular and successul solution.
\hether you run an enterprise data center sering 200,000 users, a small to
medium-sized operation sering 200 users or a web-based business handling 25,000
transactions per hour, you are concerned with economically maximizing the
aailability, scalability, lexibility and the manageability o your computing resource.
Oracle`s Real Application Clusters, introduced in 2001, oers a proen alternatie
to traditional expensie hardware and proprietary sotware solutions or achieing
these goals. 1his paper presents seeral scenarios representing the common
questions or decisions I1 executies ace and how enterprise customers hae used
Oracle Real Application Clusters to address these business needs.
OracIe ReaI AppIication CIusters - the foundation for Enterprise Grid
Computing
In June 2001, Oracle released Oracle9i with Real Application Clusters. Oracle9i
RAC made it possible or a collection o database serers to cooperate
transparently in the management o a single Oracle database. 1his allowed Oracle
to delier greater scalability and aailability than had preiously been possible while
simultaneously reducing costs and improing lexibility. In January 2004 Oracle
released the second generation o its clustered database technology, Oracle Real
Application Clusters 10g release 1 and in June o 2005, 10g Release 2.
OracIe ReaI AppIication
CIusters (RAC) is a key enabIing
technoIogy for OracIe
Enterprise Grid computing.
OracIe RAC 10g ReIease 2
offers a compIete grid
infrastructure for each Iayer of
the enterprise grid -
appIication, database and
storage grids.
OracIe ReaI AppIication CIusters 10 - the Foundation for Grid Computing Page 3


Figure 1: OracIe RAC -cIustering database servers - foundation for Enterprise Grid
Computing deIivering high avaiIabiIity, scaIabiIity and fIexibiIity.

QUESTIONS AND DECISIONS
\e speak with many companies throughout the course o the year, many CIOs,
C1Os and executies responsible or the computing inrastructure or a multitude
o enterprises, large to small. Increasingly, they look to endors such as Oracle or
adice and guidance on the decisions and questions they ace like:
- My SMP infrastructure reaches its financial and technology
refresh point in fourteen months. What do we do? Is this an
opportunity to consider an enterprise grid architecture based upon
Oracle 10g Real Application Clusters?" A major on||ne reta||er made the
sh|ft and we've documented |t here.

- Our business is successful and growing that's the good news
and the bad news. We seem to reach capacity on our critical
systems and it requires a forklift upgrade to get more scalability.
Lxpensie and disruptie. \ill Oracle Real Application Clusters 10g
and the Oracle Grid ision proide a less disruptie, more lexible and
RAC: The CIuster Database
CIustered
Database Servers
Mirrored Disk
Subsyst em
High Speed
Switch or
Interconnect
Hub or
Switch
Fabric
Network
CentraIized
Management
ConsoIe
Users
No SingIe
Point Of FaiIure
Private Interconnect
Storage Area Net work
ReaI AppIication CIusters

OracIe ReaI AppIication CIusters 10 - the Foundation for Grid Computing Page 4
scalable solution` 1he 6h|cago 8tock Exchange tack|ed th|s concern
w|th 0rac|e RA6 as to|d be|ow.

- Our average server utilization is well below S0 -- sure we hae
spikes to oer 90 but that happens at dierent times o the month or
each o the arious applications we`e dedicated to these systems.` \ill
Oracle`s 10g grid solution improe our utilization AND enhance the
aailability, scalability and lexibility o my enironment Ta|k Amer|ca
|s a b|g proponent of 0rac|e 10 Rea| App||cat|on 6|usters because |t
enab|es a 8erv|ce 0r|ented Arch|tecture wh|ch v|rtua||zes the under|y|ng
resources prov|d|ng better hardware ut|||zat|on, f|ex|b|||ty, |mproved 8LAs
and |ower cost.



E-RETAILER - SMP REFRESH? NO - OPTING INSTEAD FOR LINUX-
BASED REAL APPLICATION CLUSTERS FOR THEIR MISSION
CRITICAL APPLICATIONS.
\ell-run companies manage their capital budgets and inrastructure spending ery
careully. 1wo years ago, the core one o the largest e-retailers computing
inrastructure was a large SMP system handling OL1P and Data \arehousing
actiities. 1hat asset was due or reresh,replacement. 1his system ran a number o
applications and Oracle databases rom their Catalog, Customer database and their
order,shipment,payment database.
\hat they thought was going to be a simple contract renewal wasn`t. 1his customer
suered seere sticker shock when presented with a nearly >10M million quote to
reresh the SMP UNIX enironment that had been their mainstay. At this point,
the customer had to look or other ways to achiee their high aailability and
scalability goals while maintaining their perormance leels.
Instead, they architected a cluster o Intel-based commodity serers running Linux.
Oracle Real Application Clusters with its embedded clusterware and cluster ile
system. Deploying on twenty-eight nodes in three clusters ,one containing 16
nodes,, this e-retailer was able to migrate to an aailable and scalable enironment
or a total capital cost o 25 o what was initially proposed by the hardware
endor saing them nearly >6M in capital outlays. 1he annual support and
maintenance ees were correspondingly reduced by 0. Already amiliar with the
Oracle Database operating in a single instance mode on their SMP system, this e-
retailer elt comortable considering a Real Applications Clusters solution.
High AvaiIabiIity: One o the key driers to moe to Oracle RAC or this e-Retailer
was its compelling actie,actie` architecture and the high aailability that oered
Sticker Shock of nearIy $10M for a
UNIX-SMP pIatform refresh made the
decision to move to a of RAC Linux-
based cIuster easy for this weII-
known e-retaiIer. RAC running on Iow
cost servers enabIed them to avoid $6
million in capital outlays.
OracIe ReaI AppIication CIusters 10 - the Foundation for Grid Computing Page 5
while proiding better resource utilization.. \ith Oracle RAC, i a node in the
cluster ails, the database keeps on running. All other nodes in the cluster remain up
and users continue processing. \ith a single SMP serer, i anything ails the whole
machine goes down - user processing is interrupted and can take 20 to 30 minutes
to resume on a cold standby ,actie,passie, system. In a cold,standby
enironment, when the primary system ails, administrators hae to manually
dismount and remount storage olumes, user connections hae to be established to
the new serer and the buer cache in the standby`s memory needs to be populated
with the most requently used data. A signiicant part o the cost saings this e-
Retailer experienced was due to not haing to purchase and warehouse a cold
ailoer SMP system.
Oracle Real Application Cluster`s actie,actie system oers irtually uninterrupted
processing or users and signiicantly better, more complete utilization o
computing resources as compared to the traditional cold standby solutions. Oracle`s
RAC eliminates the need or these recoery steps - all cluster nodes are always
connected to all the storage so no olume dismounting and remounting is required.
Users on the suriing nodes remain connected only the users on the ailed nodes
need to reconnect and RAC does this:
- rapidly and transparently such that users oten don`t een know a
ailure has occurred
- automatically - no need or administrator interention and pre-set
ailoer preerences to speciied nodes can be established through
Oracle Lnterprise Manager.
Oracle RAC systems can be conigured to hae no single point o ailure, een
when running on low-cost, commodity hardware and storage. In an Oracle RAC
enironment i database serers ail their applications simply keep running. lail-
oer is requently transparent to applications and occurs in seconds. Oracle also
protects rom major site ailures by combining Oracle RAC with Oracle Data
Guard - part o Oracle`s Maximum Aailability Architecture ,MAA,.

THE CHICAGO STOCK EXCHANGE - MORE CAPACITY WITHOUT THE
HEAVY LIFTING
1he Chicago Stock Lxchange ,ClX, is the third most actie stock exchange in the
United States by olume. It is a technologically adanced exchange, proiding ast,
quality executions. Due to the nature o the business, continuous uptime, scalability
on demand, and rock-solid perormance are absolutely critical.
lacing shrinking reenues in the late 1990s, ClX's I1 department needed to ind a
way to do more with less money. At the same time, it needed to eliminate
bottlenecks created by massie data loads on heay trading days. Seeking peak
reliability and perormance, ClX chose Oracle Database with Real Application
Clusters on lP hardware or its optimal solution. ClX`s needs were growing, most
Our onIy question was: ReaI
AppIication CIusters or not?" said
David MiIne, director of database
technoIogies for the Chicago Stock
Exchange. "Oracle's clustering
technology was irresistible. Most
important was its reIiabiIity and
avaiIabiIity. Its seIIing points were
ease and fIexibiIity in adding Iarge or
smaII hardware as needed, where
needed, and no changes required to
appIications."

OracIe ReaI AppIication CIusters 10 - the Foundation for Grid Computing Page 6
likely requiring it to procure the next class o SMP machine. But, ClX was
concerned about signiicantly oer-buying or under-buying serer capacity - trading
olumes spiked by 2 times in 2000 but dropped signiicantly in 2002. 1hey also
were concerned that as newer technology emerged, they would not be able to take
adantage o it without replacing the entire serer inrastructure.
In a two node cluster, a hardware ailure on one node could negatiely impact all
users, and result in lost trading reenues Migrating to a our node RAC,Grid
enironment signiicantly reduced the exposure to a single hardware ailure. Oracle
RAC allowed the Chicago Stock Lxchange to easily and quickly add more low-cost
serers to the cluster. 1his has signiicantly reduced their exposure and extended
the useul lie o existing hardware. \ith Oracle RAC a hardware ailure on one
node is not catastrophic, the workload is automatically shited to other hardware in
the Grid. Most o the time the users are not aware o the ailure. In addition, the
results o a serer ailure are less urgent in the Grid enironment, while still a
critical eent, it is not necessarily the all hands on deck` scenario.
Improved ScaIabiIity, FIexibiIity and ManageabiIity can be achieved
Lnterprises hae also had to oer-proision to accommodate changes in business
transaction olume and in the spikiness o web-based transaction actiity. Scaling
in a traditional SMP enironment required long lead time adance planning due to
the lead times in obtaining the hardware, preparing the enironment and shiting
enironments.
ScaIe Out: Oracle RAC allows multiple serers in a cluster to transparently manage a
single database. Oracle RAC allows database systems to scale-out rather than
haing to scale-up to meet additional workload increases. In the scale-up model o
computing, once a serer has been ully conigured with CPUs and memory, the
next step is an expensie "ork-lit upgrade", but with Oracle RAC users just plug
in another serer. i additional processing power is needed.
RAC also proides the opportunity to rapidly retire excess capacity and re-deploy
that or a dierent purpose either as a node in an application serer mid-tier layer
or as an additional database cluster node in a dierent cluster.
In addition to eliminating scalability constraints, RAC has proen to scale better
than SMP when ealuated in a head-to-head comparison. Gien the same number
and type o CPUs, the same amount o total memory, RAC has proen to oer
greater scalability and processing throughput than a big SMP enironment.
Project MegaGrid - Proving "scaIe out" as an economicaIIy attractive aIternative
As a ounding partner o Project MegaGrid, in phase one o this multi-phase
collaboration with Dell Computers, LMC and Intel, compelling eidence o the
power o a low-cost RAC enironment was gathered. Starting in 2004, Oracle,
Dell, LMC and Intel came together to alidate Oracle`s grid computing ision and
establish grid computing best practices as a way to encourage adoption and reduce
OracIe ReaI AppIication CIusters 10 - the Foundation for Grid Computing Page 7
market conusion and hesitation to embrace this new model. In phase 1, a standard
telecom serice proisioning application rom Cramer Systems, used by telecom
companies such as British 1elecom and Bell South, was benchmarked on both a
large 2CPU SMP system and an Oracle RAC cluster o low-cost commodity Dell
serers. 1he goal was to process as many transactions o a mixed workload per
hour while maintaining a maximum our-second response time. 1he >2.8 million
SMP serer enironment was able to handle 550,000 transactions per hour. 1he
same olume o transactions while maintaining the SLAs was achieed on a 10-
node, 20 CPU Dell PowerLdge 150 serer cluster costing about >60,000. Another
test but maintaining the same SLAs was achieed on a our-node our CPU 64-bit
Dell 250 serer cluster costing >160,000.

Figure 2: CIustered, Iow-cost servers running ReaI AppIication CIusters can offer an
economic, avaiIabiIity and scaIabiIity advantages over traditionaI architectures


TALK AMERICA
In business or 16 years, 1alk America proides local and long distance telephone
serices to residential and small business customers in the United States. Oer the
last ie years, their emphasis has shited more to local telephone serice where the
marketplace or those serices is ery actie and where customer serice and where
1alk America s customer serice and responsieness is a competitie adantage -
something they take so seriously they were awarded the J.D. Powers Serice
"The beauty of OracIe 10g
and a RAC-enabIed grid is
that the choice is mine. I can
manage SLAs based upon
my business needs."
Laurence Grant - Vice
President of Enterprise
Computing Systems - TaIk
America
Cost Comparison: SMP vs. Grid
$0
$1
$2
$3
$4
$5
$6
$7
$8
MiIIions
Server
Network
Infrastructure
Software
Storage
Unix
72 CPU
SMP
Ten 2-cpu
DeII PE1750s
(Xeon)
Four 4-cpu
DeII PE7250s
(Itanium)
TeIecom WorkIoad - 550,000
transactions/hour
OracIe ReaI AppIication CIusters 10 - the Foundation for Grid Computing Page 8
Lxcellence award or top customer satisaction rating in the telecommunications
segment. Core to 1alk America`s competitieness is their I1 operation and the
proprietary integrated order processing, proisioning, billing, payment, collection,
and customer serice inormation systems which enables them to oer
competitiely priced serice plans, high-quality serice and simplicity through
consolidated billing and responsie customer care.
lundamental to high-quality customer serice is a highly aailable systems
enironment. 1alk America`s legacy enironment, based upon 20 dierent
Inormix databases, was preenting them rom meeting their business goals -
database reliability was a problem. Additionally, 1alk America was acing a number
o other challenges that was stressing their I1 enironment:
- Rapid growth o their business and the exponential growth o call
records and transaction olumes
- Need to proide a 24x operation
- Increasing systems dependencies - less time or maintenance
- Shrinking nightly batch windows
- Growing multitude o systems to satisy growing end-user, internal
users ,sales,serice, and increased reporting requirements
,Sarbanes,Oxley etc.,
- Much larger data olumes growing to 60 terabytes rom only 1
terabyte eight years earlier
1alk America had to make some changes - changes that mandated an upgrade o
their inrastructure to a high perormance, highly aailable, lexible enironment.
Laurence Grant, Vice President o Lnterprise Computing Systems at 1alk America,
and Vishal Anand, Director o Database Administration, were concerned that
simply upgrading their silo`ed enironment would be expensie and would not
really sole the problems they aced. Inariably, some decisions about serer and
storage capacity would result in an oer allocation or some applications and under
allocation or others.
1alk America decided to pursue Oracle Database 10g and Real Application Clusters
as their core database and high aailability inrastructure. Len though the solution
was ery new at the time, Larry and Vishal belieed that the adantages 10g brought
to their problems ar outweighed the risks o going with a relatiely new product
release.



OracIe ReaI AppIication CIusters 10 - the Foundation for Grid Computing Page 9
OracIe Database 10 and ReaI AppIication CIusters - the underpinning
of a Service Oriented Architecture:
1alk America was interested in the economies that could be gained i they
consolidated databases and migrated to a serice-oriented architecture. It meant
greater hardware utilization, more lexibility to moe resources to accommodate
spikes in workload demands, and cost saings by not haing to inest in an
expensie, dedicated solution or each major system. 1alk America had experience
with big iron solutions and they didn`t want to buy hardware eery six, nine or
twele months to keep pace with the business - they wanted to aoid the business
disruption and the additional costs o shuling equipment in and out. Oracle`s 10g
RAC was the key that enabled the high aailability and workload sharing and
balancing allowing 1alk America to gain better resource utilization and aoid
changing the hardware ootprint eery ew months.
1alk America wanted an enironment that would enable them to manage resources
to satisy the arious serice leel agreements ,SLAs, established or their users.
1hey took adantage o the 10g Grid capabilities and consolidated all their 20
separate databases into one 601B Oracle database. 1he arious serices such as
Order Processing, Proisioning, Leads Processing, Data Marts, CRM etc. run on
the RAC cluster and share access to the single database - administration is easier,
aailability is higher by irtue o Oracle RAC, and lexibility is greater in terms o
adding or remoing processing power to match the needs o the application at any
point.
Leveraging AII the Hardware AII the Time
1he Oracle Database 10g serice-oriented architecture enabled by Oracle Real
Application Clusters, binds serices to the underlying hardware - but in a lexible,
irtual way. It is ery easy within a 10g architecture to run OL1P applications and
data warehousing within the same RAC cluster. And it`s easy to change the
allocation o resources as business processing needs change - hence leeraging all
the hardware all the time.
Larry Grant explains the one o the key adantages an Oracle Real Application
Clusters 10g - enabled grid proides.
. .ivte eavte, .vo.e bare O1P .errice attocatea to fovr voae.. f ove voae goe. aorv
`v .titt bigbt, araitabte becav.e tbree voae. are v ava rvvvivg attbovgb ,ov vigbt be rvvvivg
ritb aegraaea erforvavce. 1bat va, be accetabte, borerer, a. tovg a. ,ov`re rvvvivg. vt rbat
Oracte R.C 10g .a,., `re got fovr voae. rvvvivg O1P ava tro voae. aoivg reortivg iv v,
.i voae ctv.ter. Reortivg i. vot vi..iovcriticat, borerer. Ove of v, fovr O1P voae. goe.
aorv vor iv.teaa of rvvvivg ;v.t ov tbree voae., `v goivg to reairect .ove of v, traffic to ove of
tbo.e otber reortivg voae.. `v goivg to be rvvvivg O1P ritb fovr voae. ;tbree .vrririvg voae.
ava ove revro.ea reortivg voae). av cov.ciov.t, bac/ivg off v, reortivg .o iv.teaa of v,
O1P rvvvivg aegraaea, it. goivg to rvv at fvtt .teav ava `v goivg to tet v, reortivg .,.tev
aegraae. 1be beavt, of Oracte 10g ava a R.Cevabtea gria i. tbat tbe cboice i. vive. cav
vavage .. ba.ea vov v, bv.ive.. veea..
OracIe ReaI AppIication CIusters 10 - the Foundation for Grid Computing Page 10


Figure 3. TaIk America's typicaI service aIIocation among the six OracIe RAC nodes
impIemented on a 96 CPU two HP Superdome environment. Services can be reIocated
and resource aIIocation expanded or contracted as daiIy, monthIy, year-end processing
requires.

1alk America has achieed greater hardware utilization, better SLA achieement,
and signiicant cost saings by irtue o the ull technology stack supplied by Oracle
Database 10g and Real Application Clusters ,no need to acquire >500,000 o third
party clustering sotware,. Again - leeraging all the hardware all the time.

CONCLUSION
Oracle 10g Real Application Clusters oers enterprises:
- Radical economic beneits when migrating rom an aging SMP
enironment to a modern enterprise grid o low cost serers enabled with
Oracle RAC.
- An agile response to changes in business requirements through the
addition or subtraction or re-purposing o serer resources as transaction
olumes, processing schedules, number o user sessions or application mix
changes.
- Increased hardware utilization which enables your I1 unction to und
more strategic initiaties by re-directing budget away rom under-utilized
legacy hardware,sotware inrastructure to strategic deelopment
initiaties. 1his results in enhanced competitieness and customer
responsieness.


OracIe ReaI AppIication CIusters 10- the Foundation for Enterprise Grid Computing
September 2005
Author: R. Hietter


OracIe Corporation
WorId Headquarters
500 OracIe Parkway
Redwood Shores, CA 94065
U.S.A.

WorIdwide Inquiries:
Phone: +1.650.506.7000
Fax: +1.650.506.7200
oracIe.com

Copyright 2005, OracIe. AII rights reserved.
This document is provided for information purposes onIy and the
contents hereof are subject to change without notice.
This document is not warranted to be error-free, nor subject to any
other warranties or conditions, whether expressed oraIIy or impIied
in Iaw, incIuding impIied warranties and conditions of merchantabiIity
or fitness for a particuIar purpose. We specificaIIy discIaim any
IiabiIity with respect to this document and no contractuaI obIigations
are formed either directIy or indirectIy by this document. This document
may not be reproduced or transmitted in any form or by any means,
eIectronic or mechanicaI, for any purpose, without our prior written permission.
OracIe, JD Edwards, and PeopIeSoft are registered trademarks of
OracIe Corporation and/or its affiIiates. Other names may be trademarks
of their respective owners.