oracle io topics

8/8/2019 Oracle IO Topics

1/64

Oracle I/O TopicsOracle I/O Topics

Oracle and StorageOracle and Storage


2/64

Objectives for this unitObjectives for this unitAt the end of this module the student will understand theAt the end of this module the student will understand the

following tasks and concepts.following tasks and concepts. Understand how a disk drive worksUnderstand how a disk drive works

Understand disk drive performanceUnderstand disk drive performance

Understand RAIDUnderstand RAID

Understand disk controllersUnderstand disk controllers Understand disk configuration optionsUnderstand disk configuration options

Understand the read and write mechanisms of the SANUnderstand the read and write mechanisms of the SANsystemsystem

Understand the read and write mechanisms of OracleUnderstand the read and write mechanisms of Oracle

Understand how these mechanisms affect performanceUnderstand how these mechanisms affect performance


3/64

Physical Parts in a DiskPhysical Parts in a DiskDriveDrive

Disk Platters tracks and sectorsDisk Platters tracks and sectors Disk Heads read/write heads thatDisk Heads read/write heads that

perform reads and writes to theperform reads and writes to thetracks on the platterstracks on the platters


4/64

Disk Drive ComponentsDisk Drive Components

Disk

Platter


5/64


Disk

Platter

Track


6/64


Disk

Platter

Track

Sector


7/64


DiskPlatter

Track

Sector

Armature

Head


8/64


Cylinder


9/64

How a Disk WorksHow a Disk Works

RPM disks spin at a given RPM, suchRPM disks spin at a given RPM, suchas 15,000 rpmas 15,000 rpm

Seek Time the time it takes for theSeek Time the time it takes for the

head to move to the trackhead to move to the track Rotational Latency the time it takesRotational Latency the time it takes

for the disk to rotate (spin) to thefor the disk to rotate (spin) to thesector where the needed data issector where the needed data is


10/64

Disk Drive Seek andDisk Drive Seek andRotationRotation


11/64

Disk Drive Seek and RotationDisk Drive Seek and Rotation

Step 1: Seek to the DesiredStep 1: Seek to the DesiredTrackTrackTime = Seek TimeTime = Seek Time


12/64

Disk Drive Seek and RotationDisk Drive Seek and Rotation

Step 2: Wait for Sector toStep 2: Wait for Sector toRotateRotate

Time = Rotational LatencyTime = Rotational Latency


13/64

Disk PerformanceDisk PerformanceSpecificationsSpecifications Typical 18 Gbyte SCSI-3Typical 18 Gbyte SCSI-3

Rotational Speed - 15,000 RPMRotational Speed - 15,000 RPM

Avg. Seek for random I/Os 3.9 msAvg. Seek for random I/Os 3.9 ms

read, 4.5ms writeread, 4.5ms write Transfer Rate 40MB/secTransfer Rate 40MB/sec

Track to Track Seek for sequential I/OsTrack to Track Seek for sequential I/Os0.5ms read, 0.7 ms write0.5ms read, 0.7 ms write

Rotational Latency - 2.0 msRotational Latency - 2.0 ms


14/64

Calculating Max RandomCalculating Max RandomSeeks/SecSeeks/Sec Maximum Random Seeks / secMaximum Random Seeks / sec

max random seeks/sec = 1 / (sec per seek)max random seeks/sec = 1 / (sec per seek)

sec per seek = (3.9ms+2.0ms)(1sec/1000ms)sec per seek = (3.9ms+2.0ms)(1sec/1000ms)

1 / 0.0059 (sec/seek) = 169.5 Seeks/sec1 / 0.0059 (sec/seek) = 169.5 Seeks/sec

What about queuing?What about queuing?


15/64

I/O Request QueuingI/O Request QueuingGraphGraph

QUEUE LENGTH VS. UTILIZATION

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

16.000

18.000

20.000

5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70% 75% 80% 85% 90% 95%

UTILIZATION

QUEUELEN

GT


16/64

Maximum Utilization for BestMaximum Utilization for BestPerformancePerformance Maximum Seeks per second = 169.5Maximum Seeks per second = 169.5 Knee of Curve at 80%Knee of Curve at 80%

Configure for 125 I/Os per second perConfigure for 125 I/Os per second per

disk for random I/Osdisk for random I/Os This is 75% of maximum capacityThis is 75% of maximum capacity


17/64

Sequential vs. RandomSequential vs. RandomI/OsI/Os Sequential I/O is much fasterSequential I/O is much faster

Seek time 5.5 ms 0.7 msSeek time 5.5 ms 0.7 ms

Same calculation yields 370 I/Os per secSame calculation yields 370 I/Os per sec

or 277 I/Os per sec @ 75%or 277 I/Os per sec @ 75% > 300+ I/Os per sec is common for> 300+ I/Os per sec is common for

sequentialsequential

As I/Os increase so does LatencyAs I/Os increase so does Latency


18/64

Disk Layout for SequentialDisk Layout for SequentialI/OsI/Os Isolate Sequential I/OsIsolate Sequential I/Os

Sequential Disk I/Os achieve maximumSequential Disk I/Os achieve maximumperformanceperformance 255+ I/Os per second per disk255+ I/Os per second per disk

Sequential Access in OracleSequential Access in Oracle Redo LogRedo Log

Archive Log filesArchive Log files


19/64

Disk Layout for Random I/OsDisk Layout for Random I/Os OLTP OLTP OLTP (online transaction processing)OLTP (online transaction processing)

I/Os are fairly randomI/Os are fairly random

Determine the frequency of accessDetermine the frequency of access

Spread out the dataSpread out the data BalanceBalance


20/64

Disk Layout Random I/Os -Disk Layout Random I/Os -DSSDSS DSS (Decision support system)DSS (Decision support system)

I/Os are made up of many sequentialI/Os are made up of many sequentialaccessesaccesses

I/Os may be isolated to certain tablesI/Os may be isolated to certain tables

Analyze the access patternsAnalyze the access patterns

Use range partitioningUse range partitioning


21/64

I/O BalancingI/O Balancing Disk Arrays Provide Optimal DistributionDisk Arrays Provide Optimal Distribution

Use Disk Striping (through RAID)Use Disk Striping (through RAID)

Balance I/Os among logical volumesBalance I/Os among logical volumes

Dont exceed disk throughput limitsDont exceed disk throughput limits Pushing too hard increases latenciesPushing too hard increases latencies

Use all of the disks availableUse all of the disks available

Disk Drives can do 125 IOPS randomDisk Drives can do 125 IOPS random 10 Disks can do 1,250 IOPS random10 Disks can do 1,250 IOPS random

100 Disks can do 12,500 IOPS random100 Disks can do 12,500 IOPS random


22/64

Oracle Data Files andOracle Data Files andTablespacesTablespaces Use files and tablespaces in order toUse files and tablespaces in order to

spread data across multiple disk arraysspread data across multiple disk arrays(LUNs)(LUNs)

Spreading all your user database dataSpreading all your user database data

across all the data disks usually gives theacross all the data disks usually gives thebest I/O performancebest I/O performance

Separate the log disks from the data disksSeparate the log disks from the data disks


23/64

Create Database DiskCreate Database DiskViewView

C:\C:\oracle\data\redo1.dbf

E:\ E:\oracle\data\ts1a.dbf

F:\ F:\oracle\data\ts1b.dbf

G:\ G:\oracle\data\ts1c.dbf

H:\ H:\oracle\data\redo2.dbf

Data_TS


24/64

RAID Storage SubsystemsRAID Storage Subsystems RAID (Redundant Array of InexpensiveRAID (Redundant Array of Inexpensive

Disks)Disks) RAID can provide Fault ToleranceRAID can provide Fault Tolerance

Hardware Fault Tolerance is more efficientHardware Fault Tolerance is more efficient

and faster than software fault toleranceand faster than software fault tolerance RAID LevelsRAID Levels

RAID PerformanceRAID Performance


25/64

Fault tolerant hintsFault tolerant hints OS and binaries are hardest to replaceOS and binaries are hardest to replace

Transaction log and transaction log backupsTransaction log and transaction log backupsare essential for recoveryare essential for recovery

Database recovery can be time consumingDatabase recovery can be time consuming

RAID will only protect you against disk failuresRAID will only protect you against disk failures

RAID is not a substitute for backupsRAID is not a substitute for backups


26/64

RAID-0RAID-0 Striping across disksStriping across disks

No fault toleranceNo fault tolerance Use all of the disk spaceUse all of the disk space Best PerformanceBest Performance Lose one disk and lose the dataLose one disk and lose the data

1

5

9

13

2

6

10

14

4

8

12

16

3

7

11

15


27/64

RAID-1RAID-1 Mirroring provides a copy or mirrorMirroring provides a copy or mirror

Good fault toleranceGood fault tolerance Disk Space Available = Disk Drives / 2Disk Space Available = Disk Drives / 2

RAID-1 does not have stripesRAID-1 does not have stripes

Disk 1 Disk 1

Mirror


28/64

2b

4b

6b

8b

1b

3b

5b

6b

RAID-0+1 (a.k.a. RAID 10)RAID-0+1 (a.k.a. RAID 10)

Mirroring with StripingMirroring with Striping

Good fault toleranceGood fault tolerance

Disk Space Available = Disk Drives / 2Disk Space Available = Disk Drives / 2

1a

5a

9a

13a

2a

7a

10a

14a

2b

4b

6b

8b

3a

7a

11a

15a

2b

4b

6b

8b

4a

8a

12a

16a

d 0 O


29/64

RAID-1 and RAID-10 I/ORAID-1 and RAID-10 I/OCharacteristicsCharacteristics

Physical Reads = Logical readsPhysical Reads = Logical reads

Physical Writes = Logical Writes x 2Physical Writes = Logical Writes x 2

Good read performance canGood read performance cansimultaneously read from both diskssimultaneously read from both disksin the mirrored pairin the mirrored pair


30/64

RAID-5RAID-5 StripingStriping

Distributed ParityDistributed Parity

Disk space = Disk Drives - 1Disk space = Disk Drives - 1

1

4

7

Parity

2

5

Parity

10

Parity

6

9

12

3

Parity

8

11

RAID 5 W iRAID 5 W it


31/64

RAID-5 WriteRAID-5 WriteStep 1: Data and Parity are ReadStep 1: Data and Parity are Read

1

4

7

Parity

2

5

Parity

10

Parity

6

9

12

3

Parity

8

11

iRAID 5 W it


32/64

RAID-5 WriteRAID-5 WriteStep 2: New Parity isStep 2: New Parity is

CalculatedCalculated

1

4

7

Parity

2

5

Parity

10

Parity

6

9

12

3

Parity

8

11

XOR Engine Calculates

parity

RAID 5 W iRAID 5 W it


33/64

RAID-5 WriteRAID-5 WriteStep 3: Data and Parity areStep 3: Data and Parity are

WrittenWritten

1

4

7

Parity

2

5

Parity

10

Parity

6

9

12

3

Parity

8

11

RAID 5 I/O Ch i iRAID 5 I/O Ch t i ti


34/64

RAID-5 I/O CharacteristicsRAID-5 I/O Characteristics

Reading simply reads off of the diskReading simply reads off of the diskthat contains the datathat contains the data

WritingWriting Reads the data stripeReads the data stripe

Reads the parity stripeReads the parity stripe Xors the data and writes it outXors the data and writes it out

Calculates a new parity and writes it outCalculates a new parity and writes it out


35/64

Raid Cost SummaryRaid Cost Summary

P

ro

tec

tio

n

P

ro

tec

tio

n

Cost

RAID-1

RAID-5

RAID-0

R id P fR id P f


36/64

Raid PerformanceRaid PerformanceSummarySummary

P

ro

tec

tio

n

P

ro

tec

tio

n

Performance

RAID-1

RAID-5

RAID-0


37/64

RAID SummaryRAID Summary RAID-0RAID-0

No ProtectionNo Protection Best PerformanceBest Performance Least CostLeast Cost

RAID-1RAID-1 Best ProtectionBest Protection Good PerformanceGood Performance Most ExpensiveMost Expensive

RAID-5RAID-5 Good ProtectionGood Protection Worst PerformanceWorst Performance Least Expensive Fault TolerantLeast Expensive Fault Tolerant

F lt T l TiF lt T l Ti


38/64

Fault Tolerance TipsFault Tolerance Tips

Apply Fault tolerance to selectedApply Fault tolerance to selectedComponentsComponents

RAID-1 on OS and Transaction LogRAID-1 on OS and Transaction Log

RAID-5 on data drives if mostly reads or read-onlyRAID-5 on data drives if mostly reads or read-only

RAID-1 or RAID-10 on data drives if heavy updateRAID-1 or RAID-10 on data drives if heavy updateactivity (lots of writes occur)activity (lots of writes occur)

RAID-5 on backup disks (RAID 1 or RAID 10 ifRAID-5 on backup disks (RAID 1 or RAID 10 ifbackup is too slow)backup is too slow)

C t ll O tiC t ll O ti


39/64

Controller OptionsController Options

Read and Write CachesRead and Write Caches Read aheadRead ahead

Stripe SizeStripe Size

R d d W it C hR d d W it C h


40/64

Read and Write CachesRead and Write Caches

Memory cache on the controllerMemory cache on the controller For performanceFor performance

Cache ModesCache Modes

ReadRead WriteWrite

R d Ah d C hR d Ah d C h


41/64

Read Ahead CacheRead Ahead Cache

Used when sequential access is detectedUsed when sequential access is detected Will read more data than is requested andWill read more data than is requested and

put it in cacheput it in cache

If the data is requested, the response isIf the data is requested, the response is

immediateimmediate

R d h d C t llRead ahead Controller


42/64

Read-ahead - ControllerRead-ahead - ControllerRead Cache OptionRead Cache Option

May or may not be usefulMay or may not be useful Oracle does its own read-aheadOracle does its own read-ahead

Excessive read-ahead by controller canExcessive read-ahead by controller cancause unnecessary I/Oscause unnecessary I/Os Higher performance can be achievedHigher performance can be achieved

through reading from cache, but wontthrough reading from cache, but wontusually help Oracle performanceusually help Oracle performance 500 GB database500 GB database 7 GB Oracle buffer cache7 GB Oracle buffer cache 1 GB of SAN Controller cache1 GB of SAN Controller cache What do you think?What do you think?

C t ll W it C hController Write Cache


43/64

Controller Write CacheController Write Cache Write is acknowledged as soon as it getsWrite is acknowledged as soon as it gets

to the cacheto the cache Write performance is increasedWrite performance is increased

If cache fills, write performance degradesIf cache fills, write performance degrades Hides the RAID 5 latency penaltyHides the RAID 5 latency penalty

Back end must keep up or cache will fillBack end must keep up or cache will fill The controller gives priority to writing outThe controller gives priority to writing out

the data in write cache over performingthe data in write cache over performingreads from diskreads from disk

Therefore, read performance sometimesTherefore, read performance sometimessuffers at the expense of writessuffers at the expense of writes

This is true for data files, NOT for log filesThis is true for data files, NOT for log files

O l W it t D t FilOracle Writes to Data Files


44/64

Oracle Writes to Data FilesOracle Writes to Data Files

How an Oracle write occurs:How an Oracle write occurs: Page to be modified is read from disk into SQL bufferPage to be modified is read from disk into SQL buffer

cachecache

Data is modified (written to) in the buffer cacheData is modified (written to) in the buffer cache

Oracle user needing that write can now continue withOracle user needing that write can now continue with

other work considers the write finishedother work considers the write finished Since memory is a limited quantity, eventually modifiedSince memory is a limited quantity, eventually modifieddata pages must be written to disk (by the Databasedata pages must be written to disk (by the DatabaseWriter Process)Writer Process)

When write to disk is made, the controller write cacheWhen write to disk is made, the controller write cachethen comes into play the user does not knowthen comes into play the user does not know

Si ifi f O lSignificance of Oracle


45/64

Significance of OracleSignificance of OracleWritesWrites

Since Oracle writes occur in theSince Oracle writes occur in thebuffer cache, no user waits on databuffer cache, no user waits on datafile writes to disk!!!file writes to disk!!! Thus read performance for data files isThus read performance for data files is

more important than write performancemore important than write performance For log files, write performance is moreFor log files, write performance is more

importantimportant

When to Use the ControllerWhen to Use the Controller


46/64

When to Use the ControllerWhen to Use the ControllerWrite CacheWrite Cache

The controller write cache is notThe controller write cache is notuseful for database files and may hurtuseful for database files and may hurtread performanceread performance

The controller write cache is usefulThe controller write cache is usefulfor the redo log logwriter processfor the redo log logwriter process

Write caching can be configured on aWrite caching can be configured on a

per-LUN basisper-LUN basis



47/64


How much data is on each diskHow much data is on each disk Important for OracleImportant for Oracle Seek time is part of total transfer timeSeek time is part of total transfer time Configure stripe size large enough toConfigure stripe size large enough to

avoid causing extra I/Osavoid causing extra I/Os You only want one seek per requestYou only want one seek per request You want to only go to one disk driveYou want to only go to one disk drive

Default of 64K is usually pretty goodDefault of 64K is usually pretty good

If you want to try different stripe sizes,If you want to try different stripe sizes,increase, dont decrease the stripe sizeincrease, dont decrease the stripe size



48/64

Stripe SizeStripe Size The data is distributed across all of the diskThe data is distributed across all of the disk

drives in 64K piecesdrives in 64K pieces

1

5

9

13

2

6

10

14

4

8

12

16

3

7

11

15

64K Pieces

Review of Oracle StorageReview of Oracle Storage


49/64

Review of Oracle StorageReview of Oracle StorageConceptsConcepts

Oracle performance is extremelyOracle performance is extremelydependent on the performance of thedependent on the performance of theI/O subsystemI/O subsystem Redo Log performance limits transactionRedo Log performance limits transaction

processingprocessing Read performance of data files limitsRead performance of data files limits

query response timequery response time

Write performance of data files slowsWrite performance of data files slowsdown the DBWR processesdown the DBWR processes

Redo Log PerformanceRedo Log Performance


50/64

Redo Log PerformanceRedo Log Performance

Redo log entries must be written whenever aRedo log entries must be written whenever achange is made to the data files or wheneverchange is made to the data files or whenevera commit is issueda commit is issued

Redo log entries are written into the redo logRedo log entries are written into the redo log

bufferbuffer The LGWR process takes redo log entries andThe LGWR process takes redo log entries and

writes them out to the redo logwrites them out to the redo log

If the LGWR cannot keep up, transactions waitIf the LGWR cannot keep up, transactions wait

Read PerformanceRead Performance


51/64


Most criticalMost critical Logical reads (to the buffer cache)Logical reads (to the buffer cache)

takes CPU and bus resourcestakes CPU and bus resources

Each physical read takes time andEach physical read takes time andresourcesresources Query processing time is directlyQuery processing time is directly

related to I/O latencyrelated to I/O latency

Each individual I/O operation takes aEach individual I/O operation takes afinite timefinite time

Many operations are serial in natureMany operations are serial in nature



52/64

Read PerformanceRead PerformanceExampleExample An index lookup might take 50 individual readsAn index lookup might take 50 individual readsof data from the data filesof data from the data files

Each I/O takes 10 msEach I/O takes 10 ms Total index read time would be 500 msTotal index read time would be 500 ms

What if I/O latency were 50 msWhat if I/O latency were 50 ms Total index read time would be 2.5 secondsTotal index read time would be 2.5 seconds

This is totally unacceptableThis is totally unacceptable

Oracle I/O performanceOracle I/O performance


53/64

Oracle I/O performanceOracle I/O performanceissues on SANissues on SAN

All I/O operations are going through one 100All I/O operations are going through one 100MByte linkMByte link Or 200 MByte link for FC2Or 200 MByte link for FC2

Other systems in the SAN might be causingOther systems in the SAN might be causingload on the storage arrayload on the storage array

SAN software may be causing additionalSAN software may be causing additionaloverheadoverhead

Performance monitoring of I/O operations onPerformance monitoring of I/O operations oneach system only gives you part of the storyeach system only gives you part of the story

Fibre Channel BandwidthFibre Channel Bandwidth


54/64

Fibre Channel BandwidthFibre Channel Bandwidth

Depending on the type of I/O operations beingDepending on the type of I/O operations beingdone you may be disk bound or Fibre Channeldone you may be disk bound or Fibre ChannelBandwidth boundBandwidth bound

Other systems load may be taking up yourOther systems load may be taking up yourbandwidthbandwidth

Switches help, but shared storage in a singleSwitches help, but shared storage in a singlestorage unit can be a bottleneckstorage unit can be a bottleneck

Use multiple paths to the storage if necessaryUse multiple paths to the storage if necessary

SAN I/O PerformanceSAN I/O Performance


55/64

SAN I/O PerformanceSAN I/O Performance

Fibre Channel bus is 100 MB/sec (FC1)Fibre Channel bus is 100 MB/sec (FC1) Theoretical maximum of 351 GB/hrTheoretical maximum of 351 GB/hr

Storage Processor can do 50 MB/secStorage Processor can do 50 MB/sec Theoretical maximum of 175 GB/hrTheoretical maximum of 175 GB/hr

Throw in software (varies by OS) overheadThrow in software (varies by OS) overhead More like 25 MB/sec = 87 GB/hrMore like 25 MB/sec = 87 GB/hr

SAN I/O Performance (cont )SAN I/O Performance (cont )


56/64

SAN I/O Performance (cont.)SAN I/O Performance (cont.)

NetworkNetwork GigabitGigabit

125 MB/sec = 439 GB/hr125 MB/sec = 439 GB/hr

100baseT100baseT

12.5 MB/sec = 44 GB/hr12.5 MB/sec = 44 GB/hr 10baseT10baseT

1.25 MB/sec = 4.4 GB/hr1.25 MB/sec = 4.4 GB/hr

Tape can do 4 MB/secTape can do 4 MB/sec

Theoretical Maximum of 14 GB/hrTheoretical Maximum of 14 GB/hr

Transfer SpeedTransfer Speed


57/64

Transfer SpeedTransfer SpeedComparisonComparison

050100150

200250300350400450500

Gig

abit

Fibr

eCha

nnel

Storag

ePr

oce

ssor

NTF

S

100b

aseT

T

ape

10bas

eT

G

B/hr

SAN PerformanceSAN Performance


58/64

SAN PerformanceSAN Performance

20 drive SAN might perform to 2,50020 drive SAN might perform to 2,500IOPS at 10 msIOPS at 10 ms

20 drive SAN is capable of 5,000 IOPS,20 drive SAN is capable of 5,000 IOPS,

but at 100 ms latencybut at 100 ms latency Latency is much more important thanLatency is much more important than

throughputthroughput

The performance of the SAN at theThe performance of the SAN at thelowest level relates to the performancelowest level relates to the performanceof the disk drives and SAN configurationof the disk drives and SAN configuration

What About RAID?What About RAID?


59/64

What About RAID?What About RAID?

RAID overhead can be significantRAID overhead can be significant

RAID 0 offers no protection and noRAID 0 offers no protection and nooverheadoverhead

RAID 1 or 0+1 offers good protectionRAID 1 or 0+1 offers good protection

but with a 2x write overheadbut with a 2x write overhead RAID 5 offers moderate protectionRAID 5 offers moderate protection

but with a 4x write overheadbut with a 4x write overhead

RAID overhead can significantlyRAID overhead can significantlyaffect performanceaffect performance

Storage Consolidation MightStorage Consolidation Might


60/64

Storage Consolidation MightStorage Consolidation MightCause Subsystem OverloadCause Subsystem Overload

Since several or many systems may share the sameSince several or many systems may share the samestorage the processor managing that storage mightstorage the processor managing that storage mightbecome overloadedbecome overloaded

The CPUs in the storage unit is finite and can only processThe CPUs in the storage unit is finite and can only processa finite amount of dataa finite amount of data

Monitoring your storage CPUs is usefulMonitoring your storage CPUs is useful

RAID overhead = more CPU processingRAID overhead = more CPU processing

SAN Software Could CauseSAN Software Could Cause


61/64

SAN Software Could CauseSAN Software Could CauseAdditional LatenciesAdditional Latencies

Many SAN systems require additionalMany SAN systems require additionalsoftware that might increase latenciessoftware that might increase latencies

The more layers the I/Os have to go throughThe more layers the I/Os have to go throughthe more overhead involvedthe more overhead involved

Oracle is very sensitive to I/O latencyOracle is very sensitive to I/O latency

I/O Monitoring Might BeI/O Monitoring Might Be


62/64

I/O Monitoring Might BeI/O Monitoring Might BeInaccurate or WrongInaccurate or Wrong

Each system can only see the I/OsEach system can only see the I/Osthat it generatesthat it generates

The load on the storage system is theThe load on the storage system is thecumulative I/O operations of all thecumulative I/O operations of all theservers that access itservers that access it

Additional monitoring is necessaryAdditional monitoring is necessary

ReviewReview


63/64

ReviewReview Since most transactional databases use an 8K block size,Since most transactional databases use an 8K block size,

the optimum stripe depth for a striped RAID group is 8K.the optimum stripe depth for a striped RAID group is 8K.

What is wrong with this statement?What is wrong with this statement? What is the maximum I/Os per second that you should try toWhat is the maximum I/Os per second that you should try to

push through a single 15,000 rpm disk?push through a single 15,000 rpm disk?

True or False: Slow data writes to disk are a significantTrue or False: Slow data writes to disk are a significantcause of Oracle user waitscause of Oracle user waits

What is the difference between seek time and rotationalWhat is the difference between seek time and rotational

latency?latency? What types of Oracle files are sequentially accessed?What types of Oracle files are sequentially accessed?

Under what conditions would RAID-5 be undesirable? UnderUnder what conditions would RAID-5 be undesirable? Underwhat conditions would RAID-5 be acceptable?what conditions would RAID-5 be acceptable?

True or False: Controller write cache is important in tuningTrue or False: Controller write cache is important in tuning

Oracle data write performance.Oracle data write performance.

SummarySummary


64/64

SummarySummary Sizing is very important - # of disksSizing is very important - # of disks

Choose the right RAID levelChoose the right RAID level Choose the right stripe sizeChoose the right stripe size

Dont always count on the SANDont always count on the SAN

features helpingfeatures helping A poorly designed SAN will performA poorly designed SAN will perform

poorlypoorly

The performance of the disk drivesThe performance of the disk drivesdetermines the performance of thedetermines the performance of theI/O subsystemI/O subsystem

oracle io topics

Documents