Cluster :

 It is a Set of Inter connected instances

 or Systems to perform a common task.

Clustering are of four types:

1) Network
2) Operating System
3) Hardware
4) Application

Node 1 Node 2
Operating System Operating System

Application Application
Software Software

Common Storage

DAS/NAS/SAN

DAS : Direct Attached Storage.

NAS : Network Attached Storage.
SAN : Storage Area Network.

Node represents complete system which is having its own independent resources.

As a DBA we are more concerned about Application level clustering.

Application cluster :

1) Failover cluster 2) Parallel cluster 3) Hybrid/High Availability cluster.

Failover Cluster :

N1 N2
U1 (ACTIVE) (PASSIVE)
U2
U3

C/R/D
DAS/NAS/SAN

 Max Nodes are two only

 3 Users are connecting to one node(N1)
 If node N1 goes down users sessions will be shifted to N2.

Disadvantages : At a time only one node is active another one passive. So no load
balancing feature.

Maximum we can have two nodes only.

Each node consists one instance.

Parallel Cluster : - Ex : RAC

If we use 10g RAC Max Nodes are 100 Nodes.

If we use 9i RAC Max Nodes are 64 Nodes.

In this environment some load balancing will be maintained.

Here we have all nodes in active state.
 N1 N2
(ACTIVE) (ACTIVE)

C/R/D
DAS/NAS/SAN

Hybrid Cluster : Combination of failover & parallel.

Minimum Nodes : 4 (2 Bunches)

N1
N1 N2 N2
(PASSIVE)
(ACTIVE) (PASSIVE) (ACTIVE)
F1 F2

C/R/D
DAS/NAS/SAN

Parallel Cluster between F1 & F2.

Failover Cluster between N1 & N2

ADVANTAGE is Failover and Load Balancing feature.

Components for Cluster Environment :-

1) Minimum 2 Nodes.
2) Private Network between two nodes.
3) Public Network for end users.
4) Shared Storage(DAS/NAS/SAN)
5) Cluster Software(If 9i Cluster Manager(CM) and for 10g Cluster Ready
Service)

6) Cluster file system :

1) Raw Partitions

10g 2) OCFS2 (Oracle Cluster

File Systems)
3) VXFS
4) ASM
 1) Raw Partitions

9i 2) OCFS (Oracle Cluster

File Systems)
3) VXFS

If we use 10g RAC on IBM,HP we need 3rd party cluster software.

If we use 9i RAC for

 MC SERVICE GUARD –HP UX

 HACMP (IBM)
 Veritas Cluster Software
 Sun Cluster
 Red Hat High Availability Cluster
 Novell Software.

Version :::::::::::::::::::::::::::10.2.0.1.0

Master Release Operating Specific

No. Release Number.

Maintainance Component
Release Number. Specific Release
Number.
Application Server
No.

Availability :

1) Low Availability  Data Loss + Down Time

2) Medium Availability  No Data Loss + Down Time

3) High Availability  No Data Loss + No Down Time

Failures ::
1) Hardware Level
2) Network Level
3) O/s (media) Level
4) Instance Level

Oracle Parallel Server Architecture (8i Onwards):

OBSP

IDL IDL
D M D M
B B
C SP C SP

RLB
RLB

C/R/D

IDLM : Integrated Distributed Locking Mechanism.

IDLM will be synchronized between two instances of nodes by OBSP(Oracle Block

Server Process).

So Sql statement which should not go through hard parsing, since segment data is
available in DBC(Database Buffer Cache).
Buffer available Cache Coherency(To
SNo Instance X Instance Y in requesting Maintain read consistency
instance whether that block
available in requesting
instance or not)
1 Select *from Select *from YES Cache Fusion
emp; emp;
2 Update emp set Select *from No Soft Ping (read from disk) In
sal=sal+100; emp; doubt Process
3 Commit; (I/O) Select *from No False Pinging
emp; (I/O)
4 Update emp set Update emp set No Block Pinging
sal=500 where sal=1000 where
eno=1; eno=2

From 9i Shared lock at table level,exclusive lock at row level; where as in 8i OPS the
problem is we can’t perform updation process from two instances. Because of segment
level locking
 9i RAC ARCHITECTURE

Inter Process Communication

CM + GSD CM + GSD
Global Service Daemon Global Service Daemon

DIAG DIAG

INSTANCE  X INSTANCE  Y

Variable Redo Log Data Variable Redo Log Data

Part of Buffer Base Part of Buffer Base
SGA Buffer SGA Buffer
Cache Cache

LMON LMD LCK LGWR LMSn DBWR

LMON LMD LCK LGWR LMSn DBWR

Redo Log Files Voting Disk Redo Log Files

OCR CONFIG

DATABA SE C+R+D
To overcome OPS disadvantages we use RAC.

Mandatory Background process from 9i

1) SMON 2) PMON 3) DBWR 4)LGWR 5)CKPT 6) RECO

RAC Mandatory processes

1) DIAG 2) LMSn 3)LMD 4)LMON 5)LCK

1) LMD : Global Enqueue Service Daemon.

To Serialize the transaction it is allocating enqueues inside the buffer memory(DBBC)

Detects dead locks and submitting to GRD (Global Resource Directory)
Maintaining the enqueue request status in the clustering environment.

2) LMSn : Global Enqueue Cache Service.

Responsible for shifting of enqueues that is cache fusion and synchronization of GRD along with library cache and data
dictionary cache from one node to another.

Performing cache fusion(DBBC) and non cache fusion(GRD and variable part)

3) LMON : Global Enqueue Service Monitor:

Monitor the enqueue status.

Repairs enqueues based on parity report of enqueue which is available in another node taking help of cluster group service(CGS O/s
Service). Any corroupted blocks will be check here.
.
4) DIAG: Diagnostic Daemon:

 Gives errors related to node health inside alert log file.

It also helps LMD to detect deadlocks.
Publishes deadlock information inside alert log.

5) LCK: LOCK PROCESS:

 Shifting of buffer memory

 Responsible for generating RPC calls, which is used for shifting of data from
one node to another or related to heart bit check.

6) CM : Cluster Manager : (Foreground process,OS Level ORACM)

 It will vote the node health with in specific time period.

(Hang Check timer module)
o Hang check tick
o Hang check margin

7) GSD : Global Service Daemon: Helps to perform admin tasks.

8) SRVCTL : Server control utility we can manage all remote nodes.

9) GRD (Global Resource Directory): It resides in SGA.

 In GRD it maintains status of Buffer Information. Status is nothing but current
location of Buffers.
 Which role that Buffer consist whether that is Global Role or Local Role.
 Which mode that Buffer consists of
o Exclusive (X)
o Shared (S)
o Null (N)
 Which Image
o Post Image (I)
o Current Image (O)

Instead of cluster manager we have 10g specific daemons:

CRSD-Cluster Ready Service Daemon

EVMD-Event Manager Daemon.
ORS-Oracle Notification Service.
VIP-Virtual IP
CSSD – Cluster Synchronization service daemon

9i QUORUM DISK,OCR CONFIG

10gVOTING DISK,OCR
 SWITCH

SWITCH

Etho Eth1 Etho Eth1

PIV S S PIV
C C
S S
I I

250G 1 GB
250G 1 GB

SCSI DISK

OS : LINUX AS 4.0
Software : 10.2.0.1.0

Partitions :

/ 2000MB
/boot 500MB
/opt 10000MB
/oraeng 7000MB
/crs 4000MB Not Required in 9i RAC
/tmp 2000MB
/home 4000MB
/usr 8000MB
swap Double of the RAM
/var 2000MB

O/s Installation :
#df –h
#ifconfig –a|more
#cd /opt
#cd e1000 or Intel
cd e1000-7.3.20/src
#pwd
/opt/e1000/e1000-7.3.20/src
#make install
#exit

Application  System tool  N/w device contrl

Configure  New  Ethernet Connection  Forward
Other Ethernet card  forward  forward

Statistically set IP address 192.168.100.117 Subnet Mask:255.255.255.0

Forward  Apply.
Click  eth1  Activate  Save  Quit

# init 6

# ifconfig –a
eth0-private (200.200.100.117)
eth1-public (192.168.100.117)

Note : ::: For character mode set Id=3 in /etc/inittab

CSSD : (Cluster Synchronization Service daemon)

1) It will vote the node health and node name, node status inside the voting disk.
2) If any node your adding to clustering Environment or any node your deleting
it will update that information inside voting disk.

CRSD : (Cluster Ready Service Daemon)

1) Its going to register all the services i.e CRSD, CSSD, EVMP, DNS, VIP,
Listener, Instance inside OCR (Oracle Cluster Registry).
2) If any service is not up it will try for 5 times to bring the service up but still
service is not up it generates event.
EVMD : (Event Manager Daemon)

1) The event published by CSSD is captured by EVMD and that event

information will be published in alert log file.

Log File Location : /crs/oracle/log/node_name/CRSD

/CSSD
/RVMD

ONS : (Oracle Notification Service)

1) It subscribes and publishes event meant for FAN(Fast Application

Notification)

Note : RAC is not supporting any kind of DML Failure. FAN we need to configure at the
front end application we need to configure call out script at FAN Level.

FAN Supported Connectivity’s ::: JDBC,OCI,ODP.Net

VIP : (Virtual IP) Mandatory Process

1) Its not under control of O/s, Its under control of Cluster ware Software.
2) It’s a Backup for public IP

Voting Disk : Consists Node Name, Node Status

1) Contains Software version

2) We should create in RAW disk.
3) Min Space for Voting Disk and OCR  250 MB
4) Voting disk and OCR should be in a separate disk, other than DB disk.

Hangcheck timer module:

 Hang check-tick=60
 Hang check-margin=180