data guard performance

Upload: srevenco-srevenco

Post on 06-Jul-2018

221 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/18/2019 Data Guard Performance

    1/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.1

    Oracle Active Data Guard

    Performance

    Joseph Meeks

    irector, !rod"ct Manage#ent

    Oracle $igh Availa%ility &yste#s

  • 8/18/2019 Data Guard Performance

    2/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.2

    Note to viewer 

    'hese slides provide vario"s aspects of perfor#ance da

    ata ("ard and Active ata ("ard ) *e are in the proc

    "pdating for Oracle ata%ase 12c.

    +t can %e shared *ith c"sto#ers, %"t is not intended to %

    canned presentation ready to go in its entirety

    +t provides &Cs data that can %e "sed to s"%stantiate

    perfor#ance or to provide foc"sed ans*ers to partic"la

    that #ay %e e-pressed %y c"sto#ers.

  • 8/18/2019 Data Guard Performance

    3/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.

    Note to viewer 

    &ee this A for #ore c"sto#er and sales co ) http//data%ase."s.oracle.co#/pls/ht#ld%/f

    p301451016516710677!458+,!458A9:A+214

  • 8/18/2019 Data Guard Performance

    4/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.6

    Agenda – Data Guard Performance

    ailover and &*itchover 'i#ings

    &;

  • 8/18/2019 Data Guard Performance

    5/39Copyright © 2012, Oracle and/or its affiliates. All rights reserved.5

     ata ("ard 12.1 :-a#ple = aster ailover 

    # of database

    sessions onprimary and

    6 seconds2,000 sessions

    on %oth pri#ary

    and stand%y

      !  r e  l  i  #  i

      n a  r  y

  • 8/18/2019 Data Guard Performance

    6/39Copyright © 2012, Oracle and/or its affiliates. All rights reserved.7

    ata ("ard 12.1 :-a#ple ) aster &*itchove

    # of databasesessions on

    primary andstandby

    > seconds500 sessions on

    %oth pri#ary and

    stand%y

    42 seconds1,000 sessions on%oth pri#ary and

    stand%y

      !  r e  l  i  #  i

      n a  r  y

  • 8/18/2019 Data Guard Performance

    7/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.4

    Agenda – Data Guard Performance

    ailover and &*itchover 'i#ings

    &;

  • 8/18/2019 Data Guard Performance

    8/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.>

    Synchronous Redo Transport

    !ri#ary data%ase perfor#ance is i#pacted %y the total ro"nd=

    ackno*ledge#ent to %e received fro# the stand%y data%ase

     ) ata ("ard

  • 8/18/2019 Data Guard Performance

    9/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.E

    Test ! Synchronous Redo Transport

    orkload ) 9ando# s#all inserts ?O@'! to E ta%les *ith 4>4 co##its

     ) 12 B redo siFe, 17> logical reads, 7E2 %lock changes pe

    &"n ire G6>00 M2 ?:-adata G2=>

     )1 'H 9AM, 76 Cores, Oracle ata%ase 11.2.0., Oracle @i

     ) +nfiniHand, seven :-adata cells, :-adata &oft*are 11.2..2

    :-adata art lash, art lash @ogging and rite=H

    ena%led provided significant gains

    O@'! *ith 9ando# all +nsert I 1#s 9''

  • 8/18/2019 Data Guard Performance

    10/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.10

    Test ! Synchronous Redo Transport

    @ocaI1#

    EEM

    I1

    data

    thro" 1 i

    trans

    O@'! *ith 9ando# all +nserts and I 1#s 9''

  • 8/18/2019 Data Guard Performance

    11/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.11

    Test "! Synchronous Redo Transport

    :-adata G2=>, 2=node 9AC data%ase ) s#art flash logging, s#art *rite %ack flash

    &*ing%ench O@'! *orkload

     ) 9ando# M@s, 1 #s think ti#e, 600 "sers, 7000K transac

    second, 0MH/s peak redo rate ?different fro# test 2

    'ransaction profile

     ) 5B redo siFe, 120 logical reads, 0 %lock changes per tran

    1 and 5#s 9'' net*ork latency

    &*ing%ench O@'! orkload *ith Metro=Area

  • 8/18/2019 Data Guard Performance

    12/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.12

    Test "! Synchronous Redo Transport

    0 M

    i

    1#s

    5 i

    5#s

    &*ing%ench O@'! orkload *ith Metro=Area

  • 8/18/2019 Data Guard Performance

    13/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.1

    Test *! Synchronous Redo Transport

    :-adata G2=>, 2=node 9AC data%ase ) s#art flash logging, s#art *rite %ack flash

    @arge insert O@'! *orkload

     ) 1>0K transactions per second, >MH/s peak redo rate, ran

    'ransaction profile ) 660B redo siFe, 7000 logical reads, 2100 %lock changes pe

    1, 2 and 5#s 9'' net*ork latency

    @arge +nsert O@'! orkload *ith Metro=Area

  • 8/18/2019 Data Guard Performance

    14/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.16

     0

    50

    100

    150

    200

    Test *! Synchronous Redo Transport

    > M I1

    at 1

    4

    2#s

    12

    5#s

    @arge +nsert O@'! orkload *ith Metro=Area

  • 8/18/2019 Data Guard Performance

    15/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.15

    Test 0! Synchronous Redo Transport

    :-adata G2=>, 2=node 9AC data%ase ) s#art flash logging, s#art *rite %ack flash

    Mi-ed *orkload *ith high '!&

     ) &*ing%ench pl"s large insert *orkloads

     ) 27000K t-n per second and 112 MH/sec peak redo rate

    'ransaction profile

     ) 6B redo siFe, 51 logical reads, 22 %lock changes per trans

    1, 2 and 5#s 9'' net*ork latency

    Mi-ed O@'! *orkload *ith Metro=Area

  • 8/18/2019 Data Guard Performance

    16/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.17

  • 8/18/2019 Data Guard Performance

    17/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.14

    Additional S$N% %onfiguration Details

  • 8/18/2019 Data Guard Performance

    18/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.1>

    %ustomer References for S$N% Transp

    annie Mae Case &t"dy that incl"des perfor#ance data

    Other &;

  • 8/18/2019 Data Guard Performance

    19/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.1E

    Synchronous Redo Transport

    9edo rates achieved are infl"enced %y net*ork latency, siFe, and co##it conc"rrency ) in a dyna#ic relationsh

    other that *ill vary for every environ#ent and application

    'est res"lts ill"strate ho* an e-a#ple *orkload can sca

    i#pact to pri#ary data%ase perfor#ance

     Act"al #ileage *ill vary *ith each application and enviro

    Oracle reco##ends c"sto#ers cond"ct their o*n tests,

    *orkload and environ#ent. Oracle tests are not a s"%sti

    Caveat that Applies to A@@ &;

  • 8/18/2019 Data Guard Performance

    20/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.20

    Agenda

    ailover and &*itchover 'i#ings &;

  • 8/18/2019 Data Guard Performance

    21/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.21

    Asynchronous Redo Transport

     A&;

  • 8/18/2019 Data Guard Performance

    22/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.22

    Asynchronous Test %onfiguration

    100(H online redo logs

     @og %"ffer set to the #a-i#"# of 257MH

    O& #a- 'C! socket %"ffer siFe set to 12>MH on pri#ary and s

    Oracle

  • 8/18/2019 Data Guard Performance

    23/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.2

    AS$N% Redo Transport Performance T

    &ingle +nstance0

    100

    200

    00

    600

    500

    700   ata ("ard A&;6 MH/sec on single node

    ‒ Dero transport lag

     Add 9AC nodes to scale transpor

    ‒ :ach node generates its o*n redo t

    dedicated ata ("ard transport pro

    ‒!erfor#ance *ill scale as nodes areade"ate C!L, +/O, and net*ork re

     A 10(ig:

  • 8/18/2019 Data Guard Performance

    24/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.26

    Data Guard g Streaming Networ& Pro

    &trea#ing protocol is ne* *ith

    'est #eas"red thro"ghp"t *ith 0

     A&;

  • 8/18/2019 Data Guard Performance

    25/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.25

    Agenda

    ailover and &*itchover 'i#ings &;

  • 8/18/2019 Data Guard Performance

    26/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.27

    2ulti3Stand4y %onfiguration

     A gro*ing n"#%er of c"sto#ers "se #"lti=

    ("ard config"rations. Additional stand%ys are "sed for

     ) @ocal Fero data loss $A failover *ith re#ote

     ) 9olling #aintenance to red"ce planned do*

     ) Offloading %ack"ps, reporting, and recovery

     ) 9eader far#s ) scale read=only perfor#anc 'his leads to the "estion $o* is pri#ary

    perfor#ance affected as the n"#%er of re#

    destinations increases

    Primary 3 A 'ocal Stand4y 3 #

    Remote

    Stand4y 3 %

    S$N%

    AS$N%

  • 8/18/2019 Data Guard Performance

    27/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.24

    Redo Transport in 2ulti3Stand4y %onfi

    E4.0

    E>.0

    EE.0100.0

    101.0

    102.0

    10.0

    106.0

    105.0

     

    !ri#ary !erfor#ance +#pact 16 Asynchrono"s 'ransport

     E.0

    E6.0

    E5.0

    E7.0E4.0

    E>.0

    EE.0

    100.0

    101.0

    102.0

    +ncrease in C!L

    ?co#pared to %aseline

    Change in re

    ?co#pared to

    0 = 16 destinations 0 =16 desti

  • 8/18/2019 Data Guard Performance

    28/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.2>

    Redo Transport in 2ulti3Stand4y %onfi

    E7.0

    E>.0

    100.0

    102.0

    106.0

     

    !ri#ary !erfor#ance +#pact 1 &;

  • 8/18/2019 Data Guard Performance

    29/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.2E

    Redo Transport for Gap Resolution

    &tand%y data%ases can %e config"red to re"est log fileresolve gaps fro# other stand%ys in a #"lti=stand%y con

     A stand%y data%ase that is local to the pri#ary data%ase

    the preferred location to service gap re"ests

     ) @ocal stand%y data%ase are least likely to %e i#pacted %y n

     ) Other stand%ys are listed ne-t

     ) 'he pri#ary data%ase services gap re"ests only as a last

     ) LtiliFing a stand%y for gap resol"tion avoids any overhead

    data%ase

    A d

  • 8/18/2019 Data Guard Performance

    30/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.0

    Agenda

    ailover and &*itchover 'i#ings &;

  • 8/18/2019 Data Guard Performance

    31/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.1

    Redo Transport %ompression

    'est config"ration

     ) 12.5 MH/second %a

     ) 22 MH/second redo

    Lnco#pressed vol"#

    availa%le %and*idth

     ) 9ecovery !oint O%

    i#possi%le to achie

     ) perpet"al increase 50 co#pression rat

     ) vol"#e I %and*idt

     ) ratio *ill vary acros

    9e"ires Advanced

    Conserve Hand*idth and +#prove 9!O *hen Hand*idth C

      "" 2#-secuncompressed

    " 2#-seccompressed

    :lapsed 'i#e = Min"tes

    'ransport

    @ag = MH

    A d

  • 8/18/2019 Data Guard Performance

    32/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.2

    Agenda

    ailover and &*itchover 'i#ings &;

  • 8/18/2019 Data Guard Performance

    33/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.

    Stand4y Apply Performance Test

    9edo apply *as first disa%led to acc"#"late a large n"#%

    at the stand%y data%ase. 9edo apply *as then restarted to#a- apply rate for this *orkload.

     All stand%y log files *ere *ritten to disk in ast 9ecovery A

    :-adata rite Hack lash Cache increased the redo apply

    42MH/second to 146MH/second "sing test *orkload ?Orac

     )  Apply rates *ill vary %ased "pon platfor# and *orkload

     Achieved vol"#es do not represent physical li#its

     ) 'hey only represent the partic"lar test case config"ration and

    higher apply rates have %een achieved in practice %y prod"ct

    Apply Performance at Stand4y Data4as

  • 8/18/2019 Data Guard Performance

    34/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.6

    Apply Performance at Stand4y Data4as

    'est 1 no *rite

    cache On :-adata -2

    &*ing %ench O

    42 MH/second

     )+/O %o"nd d"

     ) 1,472#s for

    co#plete

     ) 110#s H

    Apply Performance at Stand4y Data4as

  • 8/18/2019 Data Guard Performance

    35/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.5

    Apply Performance at Stand4y Data4as

    'est 2 a repeat of

    test %"t *ith *rite=

    cache ena%led

    On :-adata -2=2

    &*ing %ench O@'

    146 MH/second ap

     ) Checkpoint co#7#s vs 1,472

     ) H ile !arallel

    21#s vs 110#s

    Two Production %ustomer 51amples

  • 8/18/2019 Data Guard Performance

    36/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.7

    Two Production %ustomer 51amples

    'ho#son=9e"ters ) ata areho"se on :-adata, prior to *rite=%ack flash cach

     ) hile resolving a gap of o%served an average apply rate o

     Allstate +ns"rance

     ) ata areho"se :'@ processing res"lted in average apply

    ho"r period of 77>MH/second, *ith peaks hitting E00MH/se

    ata ("ard 9edo Apply !erfor#ance

    Redo Apply Performance for Different R

  • 8/18/2019 Data Guard Performance

    37/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.4

    Redo Apply Performance for Different R

    0100200.00600

    500700400

    $ig

    $ig

    9ange of O%served Apply 9ates for Hatch and O@'!

    &tand%y

     Apply

    9ate

    MH/sec

  • 8/18/2019 Data Guard Performance

    38/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.>

  • 8/18/2019 Data Guard Performance

    39/39

    Copyright © 2012, Oracle and/or its affiliates. All rights reserved.E