You are on page 1of 12

Cluster: Set of Inter connected instances or Systems to perform a common task.

Clustering are of four types: 1) Network 2) Operating System 3) ardware !) "pplication

Node 1 Operating System "pplication Software

Node 2 Operating System "pplication Software

Common Storage #"S*N"S*S"N

#"S: #irect "ttac$ed Storage. N"S: Network "ttac$ed Storage. S"N: Storage "rea Network. Node represents complete system w$ic$ is $a%ing its own independent resources. "s a #&" we are more concerned a'out "pplication le%el clustering. "pplication cluster: 1) (ailo%er cluster 2) )arallel cluster 3) y'rid* ig$ "%aila'ility cluster.

(ailo%er: N1 +1 ."C6I7/) +2 +3 C*1*# #"S*N"S*S"N N2 .)"SSI7/)

,a- Nodes are two only 3 +sers are connecting to one node.N1) If node N1 goes down users sessions will 'e s$ifted to N2.

#isad%antages: "t a time only one node is acti%e anot$er one passi%e. So no load 'alancing feature. ,a-imum we can $a%e two nodes only. /ac$ node consists one instance. )arallel Cluster: 0 /-: 1"C If we use 12g 1"C ,a- Nodes are 122 Nodes. If we use 3i 1"C ,a- Nodes are 4! Nodes. In t$is en%ironment some load 'alancing will 'e maintained. ere we $a%e all nodes in acti%e state. N1 ."C6I7/) N2 ."C6I7/)

C*1*# #"S*N"S*S"N y'rid Cluster: Com'ination of failo%er 5 parallel.

,inimum Nodes: ! .2 &unc$es) N1 .)"SSI7/) (2

N1 ."C6I7/)

(1

N2 .)"SSI7/)

N2 ."C6I7/)

C*1*# #"S*N"S*S"N

)arallel Cluster 'etween (1 5 (2. (ailo%er Cluster 'etween N1 5 N2 "#7"N6"8/ is (ailo%er and 9oad &alancing feature. Components for Cluster /n%ironment:0 1) 2) 3) !) :) 4) ,inimum 2 Nodes. )ri%ate Network 'etween two nodes. )u'lic Network for end users. S$ared Storage.#"S*N"S*S"N) Cluster Software.If 3i Cluster ,anager.C,) and for 12g Cluster 1eady Ser%ice) Cluster file system : 1) 1aw )artitions 3i 2) OC(S .Oracle Cluster (ile Systems) 3) 7<(S 12g 1) 1aw )artitions 2) OC(S2 .Oracle Cluster (ile Systems) 3) 7<(S !) "S,

If we use 12g 1"C on I&,; ) we need 3rd party cluster software. If we use 3i 1"C for

,C S/17IC/ 8+"1# = ) +< "C,) .I&,) 7eritas Cluster Software Sun Cluster 1ed at ig$ "%aila'ility Cluster No%ell Software.

7ersion::::::::::::::::::::::::::: 12.2.2.1.2
,aster 1elease No. ,aintenance 1elease Num'er. "pplication Ser%er No. Operating Specific 1elease Num'er. Component Specific 1elease Num'er.

"%aila'ility: 1) 9ow "%aila'ility #ata 9oss > #own 6ime 2) ,edium "%aila'ility No #ata 9oss > #own 6ime 3) ig$ "%aila'ility No #ata 9oss > No #own 6ime (ailures: 1) ardware 9e%el 2) Network 9e%el 3) O*s .media) 9e%el !) Instance 9e%el Oracle )arallel Ser%er "rc$itecture .?i Onwards): O&S)

# & C 19&

I#9 , S)

# & C

I#9 , S)

19&

C*1*#

I#9,: Integrated #istri'uted 9ocking ,ec$anism. I#9, will 'e sync$roni@ed 'etween two instances of nodes 'y O&S) .Oracle &lock Ser%er )rocess). So SAl statement w$ic$ s$ould not go t$roug$ $ard parsing; since segment data is a%aila'le in #&C .#ata'ase &uffer Cac$e). &uffer a%aila'le Cac$e Co$erency.6o SNo Instance < Instance B in reAuesting ,aintain read consistency instance w$et$er t$at 'lock a%aila'le in reAuesting instance or not) 1 Select Cfrom Select Cfrom B/S Cac$e (usion empD empD 2 3 ! +pdate emp set salEsal>122D CommitD .I*O) +pdate emp set salE:22 w$ere enoE1D Select Cfrom empD Select Cfrom empD .I*O) +pdate emp set salE1222 w$ere enoE2 No No No Soft )ing .read from disk) (alse )inging &lock )inging

(rom 3i S$ared lock at ta'le le%el; e-clusi%e lock at row le%elD w$ere as in ?i O)S t$e pro'lem is we canFt perform updation process from two instances.

3i 1"C "1C I6/C6+1/

Inter #rocess Comm$nication

CM + GSD
Global Service Daemon

CM + GSD
Global Service Daemon

DIAG
INSTANCE X

DIAG
INSTANCE Y

7aria'le )art of S8"

1edo 9og &uffer

#ata &ase &uffer Cac$e

7aria'le )art of S8"

1edo 9og &uffer

#ata &ase &uffer Cac$e

9,ON

9,#

9CG

98H1

9,Sn

#&H1

9,ON

9,#

9CG

98H1

9,Sn

#&H1

1edo 9og (iles

Voting Disk OC CON!IG


DATA"ASE C+ +D

1edo 9og (iles

6o o%ercome O)S disad%antages we use 1"C. ,andatory &ackground process from 3i 1) S,ON 2) ),ON 3) #&H1 !) 98H1 :) CG)6 4) 1/CO 1"C ,andatory processes 1) #I"8 2) 9,Sn 3) 9,# !) 9,ON :) 9CG 1) 9,#: 8lo'al /nAueue Ser%ice #aemon. 6o seriali@e t$e transaction it is allocating enAueues inside t$e 'uffer memory .#&&C) #etects dead locks and su'mitting to 81# .8lo'al 1esource #irectory) ,aintaining t$e enAueue reAuest status in t$e clustering en%ironment. 2) 9,Sn: 8lo'al /nAueue Cac$e Ser%ice. 9,S copy read consistent 'locks from $olding instance to reAuested instance 1esponsi'le for s$ifting of enAueues t$at is cac$e fusion and sync$roni@ation of 81# along wit$ li'rary cac$e and data dictionary cac$e from one node to anot$er. )erforming cac$e fusion .#&&C) and non cac$e fusion .81# and %aria'le part) 3) 9,ON: 8lo'al /nAueue Ser%ice ,onitor: ,onitor t$e entire cluster 1epairs enAueues 'ased on parity report of enAueue w$ic$ is a%aila'le in anot$er node taking $elp of cluster group ser%ice.C8S O*s Ser%ice).

!) #I"8: #iagnostic #aemon: 8i%es errors related to node $ealt$ inside alert log file. It also $elps 9,# to detect deadlocks. )u'lis$es deadlock information inside alert log. :) 9CG: 9OCG )1OC/SS: 6$e 9ock process .9CG) manages noncac$e fusion resource reAuests suc$ as li'rary; row cac$e; and lock reAuests t$at are local to t$e ser%er 4) C,: Cluster ,anager: .(oreground process; OS 9e%el O1"C,) It will %ote t$e node $ealt$ wit$ in specific time period. . ang C$eck timer module) o ang c$eck tick o ang c$eck margin I) 8S#: 8lo'al Ser%ice #aemon: elps to perform admin tasks. ?) S17C69: Ser%er control utility we can manage all remote nodes. 3) G D %Global eso$rce Director&'( It resides in S8". In 81# it maintains status of &uffer Information. Status is not$ing 'ut current location of &uffers. H$ic$ role t$at &uffer consist w$et$er t$at is 8lo'al 1ole or 9ocal 1ole. H$ic$ mode t$at &uffer consists of o /-clusi%e .<) o S$ared .S) o Null .N) H$ic$ Image o )ost Image .I) o Current Image .O) Instea) o* cl$ster manager +e ,ave -.g s/eci*ic )aemons( C1S#0Cluster 1eady Ser%ice #aemon /7,#0/%ent ,anager #aemon. ONS0Oracle Notification Ser%ice. 7I)07irtual I) CSS# = Cluster Sync$roni@ation ser%ice daemon 3i J+O1+, #ISG; OC1 CON(I8 12g7O6IN8 #ISG; OC1

SHI6C SHI6C /t$o )I7 /t$1 S C S I S C S I /t$o )I7 /t$1

2:28

1 8&

2:28

1 8&

SCSI #ISG

OS: 9IN+< "S !.2 Software: 12.2.2.1.2 )artitions: * 2222,& *'oot :22,& *opt 12222,& *oraeng I222,& *crs !222,& Not 1eAuired in 3i 1"C *tmp 2222,& *$ome !222,& *usr ?222,& *Swap #ou'le of t$e 1", *%ar 2222,&

O*s Installation:

Kdf =$ Kifconfig =aLmore Kcd *opt Kcd e1222 or Intel cd e12220I.3.22*src Kpwd *opt*e1222*e12220I.3.22*src Kmake install Ke-it "pplication System tool N*w de%ice contrl Configure New /t$ernet Connection (orward Ot$er /t$ernet card forward forward Statistically set I) address 132.14?.122.11I Su'net ,ask: 2::.2::.2::.2 (orward "pply. Click et$1 "cti%ate Sa%e Juit K init 4 K ifconfig =a et$20pri%ate .222.222.122.11I) et$10pu'lic .132.14?.122.11I) Note: (or c$aracter mode set IdE3 in *etc*initta' CSS#: .Cluster Sync$roni@ation Ser%ice daemon) 1) It will %ote t$e node $ealt$ and node name; node status inside t$e %oting disk. 2) If any node youFre adding to clustering /n%ironment or any node your deleting it will update t$at information inside %oting disk. 3) It estsa'lis$es t$e connection 'etween nodes using pri%ate interconnect. !) It master t$e first acti%e node. :) "ll t$e nodes sync$roni@es 'y registering wit$ t$e master node. C1S#: .Cluster 1eady Ser%ice #aemon) 1) Its going to register all t$e ser%ices i.e C1S#; CSS#; /7,); #NS; 7I); 9istener; Instance inside OC1 .Oracle Cluster 1egistry). 2) If any ser%ice is not up it will try for : times to 'ring t$e ser%ice up 'ut still ser%ice is not up it generates e%ent. /7,#: ./%ent ,anager #aemon)

1)The Event Manager Daemon (EVMD) is an event-forwarding daemon process that propagates events through the Oracle Notification ervice (ON )!

2)/%ent manager acts as a 'ridge 'etween CSS and C1S 3)"ll communications 'etween CSS and C1S $appens %ia /7,# 9og (ile 9ocation: *crs*oracle*log*node name*C1S# *CSS# *17,# ONS: .Oracle Notification Ser%ice) 1) It su'scri'es and pu'lis$es e%ent meant for ("N.(ast "pplication Notification) Note: 1"C is not supporting any kind of #,9 (ailure. ("N we need to configure at t$e front end application we need to configure call out script at ("N 9e%el. ("N Supported Connecti%ityFs: M#&C; OCI; and O#).Net 7I): .7irtual I)) ,andatory )rocess 1) Its not under control of O*s; ItFs under control of Cluster ware Software. 2) ItFs a &ackup for pu'lic I) 3) It is used to o%ercome t$e tcp wait outs in case of node failure. O)1OCd: ,onitors $ealt$ of eac$ process in t$e cluster in uni- ;in linu- same operation is performed 'y angc$eck timer module: ang c$eck0tickE42 ang c$eck0marginE1?2 7oting #isk: Consists Node Name; Node Status 2) 3) !) :) Contains Software %ersion He s$ould create in 1"H disk. ,in Space for 7oting #isk and OC1 2:2 ,& 7oting disk and OC1 s$ould 'e in a separate disk; ot$er t$an #& disk.

OC1:contains all t$e information a'out resources and %oting disk location

You might also like