Mapping the data warehouse in data warehousing and data mining.pdf

good for data mapping

1. Mapping the data warehouse
architecture to Multiprocessor
architecture
T.R.Lekhaa
AP –IT
SNSCE
7/16/2019 1
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE

2. •Thefunctionsofdatawarehousearebasedontherelational
databasetechnology.Therelationaldatabasetechnologyis
implementedinparallelmanner.Therearetwoadvantagesof
havingparallelrelationaldatabasetechnologyfordata
warehouse:
•LinearSpeedup:referstheabilitytoincreasethenumberof
processortoreduceresponsetime
•LinearScaleup:referstheabilitytoprovidesame
performanceonthesamerequestsasthedatabasesize
increases
7/16/2019 2
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE

3. Types of parallelism
•Therearetwotypesofparallelism:
•InterqueryParallelism:Inwhichdifferentserverthreadsorprocesses
handlemultiplerequestsatthesametime.
•IntraqueryParallelism:ThisformofparallelismdecomposestheserialSQL
•Intraqueryparallelismcanbedoneineitheroftwoways:
• Horizontalparallelism:whichmeansthatthedatabaseispartitioned
acrossmultipledisksandparallelprocessingoccurswithinaspecifictask
thatisperformedconcurrentlyondifferentprocessorsagainstdifferent
setofdata
7/16/2019 3
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE
query into lower level operations such as scan, join, sort etc.
• Vertical parallelism: This occurs amongdifferent tasks. All query
components such as scan, join, sort etc are executed in parallel in a
pipelined fashion.

4. 7/16/2019 4
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE

5. Data partitioning
•Datapartitioningisthekeycomponentforeffectiveparallel
executionofdatabaseoperations.Partitioncanbedonerandomly
orintelligently.
•Randomportioningincludesrandomdatastripingacrossmultiple
disksonasingleserver.Anotheroptionforrandomportioningis
roundrobinfashionpartitioninginwhicheachrecordisplacedon
thenextdiskassignedtothedatabase.
•IntelligentpartitioningassumesthatDBMSknowswhereaspecific
recordislocatedanddoesnotwastetimesearchingforitacrossall
disks.
•Thevariousintelligentpartitioninginclude:
•Hashpartitioning:Ahashalgorithmisusedtocalculatethe
partitionnumberbasedonthevalueofthepartitioningkeyfor
eachrow
7/16/2019 5
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE

6. •Userdefinedportioning:Itallowsatabletobepartitionedon
thebasisofauserdefinedexpression.
7/16/2019 6
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE
•Key range partitioning: Rows are placed and located in the
partitions accordingto the value ofthe partitioning key.
•Schema portioning: an entire table is placed on one disk;
anothertableisplacedondifferentdisketc.

7. Data base architectures of parallel
processing
•There are three DBMS software architecture
styles for parallel processing:
•1. Shared memory or shared everything
Architecture
•2. Shared disk architecture
•3. Shared nothing architecture
7/16/2019 7
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE

8. Shared Memory Architecture
•Tightlycoupledsharedmemorysystems,illustratedinfollowingfigure
havethefollowingcharacteristics:
7/16/2019 8
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE
•
Sym
MultiplePUssharememory.
• EachPUhasfullaccess toallsharedmemory throughacommonbus.
• Communicationbetween nodesoccursviasharedmemory.
• Performance islimited bythebandwidthofthememorybus.
• metric multiprocessor (SMP) machines are often nodes in a cluster.
•

9. 7/16/2019 9
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE

10. •Parallelprocessingadvantagesofsharedmemory
systemsarethese:
• Sharedmemorysystemsareeasiertoadministerthan
acluster.
•Adisadvantageofsharedmemorysystemsforparallel
processingisasfollows:
• Scalabilityislimitedbybusbandwidthandlatency,
andbyavailablememory.
7/16/2019 10
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE
• Memory access is cheaper than inter-node
communication.

11. Shared Disk Architecture
•Shareddisksystemsaretypicallylooselycoupled.Such
systems,illustratedinfollowingfigure,havethefollowing
characteristics:
• EachnodeconsistsofoneormorePUsandassociated
memory.
• Memoryisnotsharedbetweennodes.
• Communicationoccursoveracommonhigh-speedbus.
• Eachnodehasaccesstothesamedisksandother
resources.
• AnodecanbeanSMPifthehardwaresupportsit.
• Bandwidthofthehigh-speedbuslimitsthenumberof
nodes(scalability)ofthesystem.
7/16/2019 11
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE

12. 7/16/2019 12
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE

13. greaterdependencyonhigh-speedinterconnect.
• Iftheworkloadisnotpartitionedwelltheremaybehighsynchronization
overhead.
• Thereisoperatingsystemoverheadofrunningshareddisksoftware.
7/16/2019 14
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE
ksystemsareasfollows:•Parallelprocessingadvantages ofshareddis
• Shared disk systems permit high availability.
• •Shareddisksystems provideforincremental growth.
•Parallelprocessing disadvantages ofshareddisksystems arethese:
• Inter-node synchronization is required, involving DLM overhead and

14. Shared Nothing Architecture
•Sharednothingsystemsaretypicallylooselycoupled.In
sharednothingsystemsonlyoneCPUisconnectedtoagiven
disk.Ifatableordatabaseislocatedonthatdisk,access
dependsentirelyonthePUwhichownsit.Sharednothing
systemscanberepresentedasfollows:
7/16/2019 15
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE

15. 7/16/2019 16
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE

16. •Sharednothingsystemshaveadvantagesanddisadvantagesforparallel
processing:
•Advantages
• Sharednothingsystemsprovideforincrementalgrowth.
• Systemgrowthispracticallyunlimited.
• MPPsaregoodforread-onlydatabasesanddecisionsupport
applications.
• Failureislocal:ifonenodefails,theothersstayup.
•Disadvantages
• Morecoordinationisrequired.
• Moreoverheadisrequiredforaprocessworkingonadiskbelongingto
anothernode.
• Ifthereisaheavyworkloadofupdatesorinserts,asinanonline
transactionprocessingsystem,itmaybeworthwhiletoconsiderdata-
dependentroutingtoalleviatecontention.
7/16/2019 18
UNIT -1 MAPPING THE DW ARCHITECTURE
INTO MULTIPROCESSOR ARCHITECTURE