Oracle Maximum Availability Architecture Best PracFces For Oracle Database 12c

Oracle  Maximum  Availability   Architecture  Best  PracFces  For   Oracle  Database  12c       CON6537   Michael  Smith   ConsulFng  Member  of  Te...
Author: Frederick Bell
8 downloads 1 Views 4MB Size
Oracle  Maximum  Availability   Architecture  Best  PracFces  For   Oracle  Database  12c       CON6537  

Michael  Smith   ConsulFng  Member  of  Technical  Staff   Oracle  Server  Technologies  /  MAA     Manish  Upadhyay   Lead  Oracle  Machine  Administrator   FIS         Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |   ConfidenFal  –  Oracle  Internal/Restricted/Highly  Restricted  

Safe  Harbor  Statement   The  following  is  intended  to  outline  our  general  product  direcFon.  It  is  intended  for   informaFon  purposes  only,  and  may  not  be  incorporated  into  any  contract.  It  is  not  a   commitment  to  deliver  any  material,  code,  or  funcFonality,  and  should  not  be  relied  upon   in  making  purchasing  decisions.  The  development,  release,  and  Fming  of  any  features  or   funcFonality  described  for  Oracle’s  products  remains  at  the  sole  discreFon  of  Oracle.  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |   ConfidenFal  –  Oracle  Internal/Restricted/Highly  Restricted  

3  

Announcing  Oracle  Database  12c  Release  2  on  Oracle  Cloud   •  Available  now     – Exadata  Express  Cloud  Service  

•  Coming  soon   – Database  Cloud  Services     – Exadata  Cloud  Machine  

Oracle  is  presenFng  features  for  Oracle  Database  12c  Release  2  on  Oracle  Cloud.      We  will  announce  availability   of  the  On-­‐Prem  release  someFme  a\er  Open  World.   Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

4  

Program  Agenda   1  

Oracle  MAA  

2  

MAA  Best  PracFces  

3  

FIS    

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

5  

Impact  of  Database  DownFme  

$350K

Average  cost  of   downFme  per  hour    

3  hours  

Average  resoluFon   Fme  per  outage  

$10M   80%

 

Average  cost  of  unplanned   data  center  outage  or   disaster  

Percentage  of  all  unplanned   outages  due  to  people  and   process  issues   Source:  Ponemon  InsFtute,  KPMG  Tech  Risk  Radar,  Forrester  Research    

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

6  

Oracle  MAA  Reference  Architecture   Aligns  Oracle  Solu?ons  with  Customer  Business  Objec?ves  

GOLD   SILVER   BRONZE  

Prod/Departmental  

Development  &  Test  

Bronze  +   •  Database  Clusters    

•  Oracle  Database   •  Backups  

Business  Cri?cal   Silver  +   •  Simple  Physical   ReplicaFon  

PLATINUM   Mission  Cri?cal   Gold  +   •  Advanced  Logical  ReplicaFon    

www.oracle.com/goto/maa   Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

7  

Program  Agenda   1  

Oracle  MAA  

2  

MAA  Best  PracFces  

3  

FIS  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

8  

MAA  Best  PracFces   Always  Moving  Forward  

•  Well  known,  proven,  and  tested   •  Engineered  into  all  Oracle  products   – So\ware  and  Engineered  Systems   •  Validated  with  every  new  release   •  Documented  in  MAA  whitepapers  and  key  Oracle  Support  (MOS)  notes  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

9  

MAA  Best  PracFces   Ac?ve  Data  Guard  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

10  

AcFve  Data  Guard   Best  Disaster  Protec?on,  Real-­‐?me  Failover,  High  ROI     Primary  Database   Open  Read-­‐Write   Primary   Oracle   Instance  

Ac?ve  Standby  Database   Open  Read-­‐Only  

Oracle-­‐aware  Replica?on   Con?nuous  Oracle  Data  Valida?on  

Database   Files  

Standby     Oracle   Instance  

Database   Files  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

11  

Data  Guard  ConfiguraFon  Best  PracFces   Data  Guard  Crea?on  Made  Easy!  

•  At  the  Standby  server  install  the  Oracle  So\ware  and  start  a  listener   •  Execute  a  single  line  command   dbca -silent -createDuplicateDB -primaryDBConnectionString myprimary.domain:1523/chicago.domain -gdbName chicago.domain -sid boston -initParams instance_name=boston -createAsStandby -customScripts /tmp/test.sql

•  First  release  caveats   – SIDB  only,  does  not  support  MulFtenant  Primary  databases,  yet!   – Cannot  directly  create  RAC  Standby  databases,  but  they  can  be  converted  to  RAC.   Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

12  

Data  Guard  ConfiguraFon  Best  PracFces   Best  Prac?ces  for  11.2  /  12.1  –  ASYNC  Transport  

MAA  Best  Prac?ces:   Data  Guard   Asynchronous  Redo   Transport  

•  MAA  Best  PracFce  paper  for  both  11.2  and  12.1   •  How  to  accurately  determine  transport  lag   •  Tuning  ASYNC  transport   – Network  performance   – IdenFfy  bollenecks  

•  Diagnosing  reasons  for  ASYNC  transport  lag   – Using  AWR  to  assess  peak  redo  rate  can  be  misleading  due  to  averages  bring  down   the  rate  over  longer  period  of  Fme   – Examine  the  Fme  spent  in  each  log  to  determine  the  peak  redo  rate  on  a  finer  level  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

13  

Data  Guard  ConfiguraFon  Best  PracFces   Best  Prac?ces  for  11.2  /  12.1  –  SYNC  Transport  

MAA  Best  Prac?ces:   Data  Guard   Synchronous  Redo   Transport  

•  New  MAA  Best  PracFce  paper  for  both  11.2  and  12.1   •  Deep  dive  on  SYNC  performance  tuning   – Test  results  that  illustrate  performance  gains  when  using  best  pracFces   – For  example,  proper  online  log  file  sizes  with  a  large  banking  customer  improved   performance  by  30%   – Frequent  log  switches  force  a  checkpoint  on  the  standby  which  results  in  increased  I/ O  thereby  affecFng  performance   – Single  member  standby  redo  log  placed  on  fast  storage  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

14  

Data  Guard  ConfiguraFon  Best  PracFces  

MAA  Best  Prac?ces:   Data  Guard  Redo   Apply  

Best  Prac?ces  for  11.2  /  12.1  –  Redo  Apply  

•  Whitepaper  covers  redo  apply  performance  for  both  11.2  and  12.1   •  How  to  tune  with  examples  for  various  scenarios   – Tuning  using  top  five  wait  events   – Test  results  that  illustrate  performance  gains  when  using  best  pracFces  

•  New  installaFon  and  usage  instrucFons  for  standby  statspack   – RAC  consideraFons  documented  in  the  paper   – Using  in-­‐memory  ASH  report  when  you  can’t  install  standby  statspack  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

15  

MulF-­‐Instance  Redo  Apply   •  Parallel,  mulF-­‐instance  recovery  :  standby  will  keep  up   – Standby  recovery  -­‐  uFlizes  CPU  and  IO  across  all  nodes  of  RAC  standby   – Some  of  our  OLTP  workload  tests  on  Exadata  show  great  scalability   1400   Standby   1200   Apply   1000   Rate   800   MB/sec   600   400   200   0  

OLTP  Workload  

1  Instance  

2  Instances  

4  Instances  

8  Instances  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |      

16  

AcFve  Data  Guard  Enhancements   Manageability  

•  Supports  DiagnosFc  Pack   –  Take  AWR  snapshots  from  standby(s),  store  on  primary   –  Create  AWR  reports,  and  AWR  Extract  &  Load  of  Standby  AWR  Snapshots   –  Run  ADDM  on  primary  using  AWR  snapshots  from  standby  

•  Supports  Tuning  Pack  and  SQL  Plan  Analyzer   –  Tune  Standby  SQL  workoad  (on  standby)   •  Create  and  execute  SQL  Tuning  tasks  on  ADG  (reports  wrilen  to  primary  over  dblinks)  

–  Offload  tuning  of  primary  SQL  workload  to  standby   •  Create  remote  SQL  Performance  Analyzer  trials  to  offload  test  workloads  

–  Create  SQL  profiles  on  primary  that  can  be  applied  on  standby    

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |   ConfidenFal  –  Oracle  Internal  

Data  Guard  ConfiguraFon  Best  PracFces   Best  Prac?ce  for  11.2  /  12.1  –  Role  Transi?on  

MAA  Best  Prac?ces:   Data  Guard   Switchover  and   Failover  

•  MAA  Best  PracFces  for  both  11.2  and  12.1   •  Discussion  on  how  switchovers  have  changed  for  12.1   –  Single  command   –  New  validate  opFon  prior  to  switchover  

•  How  to  assess  your  role  transiFon  Fmings  and  where  the  Fme  is  being  spent   •  OpFmizaFons  to  make  role  transiFons  faster  with  11.2.0.2  and  higher   –  _defer_eor_orl_arch_for_so"=TRUE     –  _switchover_to_standby_opFon"=  OPEN_ALL_IGNORE_SESSIONS   –  Default  behavior  for  11.2.0.4  and  12.1   Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

18  

12.2  AcFve  Data  Guard  Enhancements     Fast  Failover  

Read/Write  

Read  

Standby  

Read/Write  

•  Read-­‐only  sessions  connected  to  AcFve   Data  Guard  will   – Remain  connected  during  the  failover/ switchover   – Become  read/write  a\er  AcFve  Data  Guard   becomes  the  primary  

•  DramaFcally  reduces  brownout  due  to   reconnect  storm  a\er  failover   –  Another  reason  to  use  AcFve  Data  Guard  (instead  of  Data   Guard)  

Standby  

Primary   Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |   ConfidenFal  –  Oracle  Internal  

Data  Guard  ConfiguraFon  Best  PracFces   Best  Prac?ce  for  11.2  /  12.1  –  Outage  Repair  

MAA  Best  Prac?ces:   Automa?c  Resolu?on  of   Outages  to  Restore  Zero   Data  Loss  Protec?on  

•  Whitepaper  covers  outage  repair   for  both  11.2  and  12.1   •  Being  updated  for  12.2   improvements   •  Detailed  discussion  of  how  Data   Guard  handles  various  outages   – Test  results  that  show  outage  repair   Fmes   – How  to  perform  the  test  and  what  to   expect  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

20  

MAA  Best  PracFces   Applica?on  Con?nuity  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

21  

End-­‐to-­‐end  ApplicaFon  Availability   Applica?on  Con?nuity  for  Unplanned  Outage   1.  Database   Request   5.  Response   ApplicaFon   Server   2.  DB  Calls  

3.  Errors,     Timeouts  

4.  Applica?on          Con?nuity  

DB  Servers  or     Data  Centers  

•  Replay  in-­‐flight  work  on  recoverable  errors   •  Masks  hardware,  so\ware,  network,  storage  errors   and  outages   •  12.1  supports  Java  (JDBC-­‐Thin,  UCP,  WebLogic  Server,   3rd  Party  Java  applicaFon  servers)   •  12.2  adds  support  for  OCI,  ODP.NET  unmanaged,   JDBC  Thin  on  XA,  Tuxedo,  SQL*Plus   •  Acchk  (part  of  Orachk)  Replay  Coverage  Analysis  –   idenFfies  %  of  SQL  calls  that  are  [not]  covered  by   automaFc  replay  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |   ConfidenFal  –  Oracle  Internal  

End-­‐to-­‐end  ApplicaFon  Availability   Masking  Planned  DB  Outage   •  When  patching  DB,  want  to  migrate  exisFng  connecFons  to  another  DB  instance   •  Works  well  in  12.1  with  Oracle  connecFon  pools   ApplicaFon  Server   srvctl  relocate  service  

–  WebLogic  Ac3ve  GridLink,  UCP,    OCI,  ODP.NET  managed  and  unmanaged,  OCI  Session   Pool,  Tuxedo),  and  3rd  party  App  Servers  using  UCP:  IBM  WebSphere,  Apache  Tomcat,   NEC  WebOTX   –  New  work  directed  to  new  connecFons;  exisFng  sessions  released  when  returned  to   pools  

•  12.2  makes  this  work  even  when  Oracle  connecFon  pools  are  not  used   –  JDBC  driver  automaFcally  reclaims  connecFons  at  safe  places   •  Also  available  with  other  drivers  (OCI,  ODP.NET,  …)  

–  Works  against  11.2  and  12.1  DB   DB  Servers  or     Data  Centers    

•  12.2  makes  this  easier  by  supporFng  relocate  or  stop  of  all  services  associated   with  a  PDB,  instance,  node,  or  DB   Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |   ConfidenFal  –  Oracle  Internal  

MAA  Best  PracFces   Mul?tenant  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

24  

MAA  Best  Prac?ces:   Database  Consolida?on   with  Oracle  Mul?tenant  

Oracle  MulFtenant   Architecture  for  consolida?ng  databases  and  simplifying  opera?ons   AP   GL  

OE  

Self-­‐contained  PDB  for  each  applica?on   •  •  •  • 

Portability  (via  pluggability)   Rapid  provisioning  (via  clones)   ApplicaFons  run  unchanged   PDB  upgrades  via  plug/unplug  

PDBs  

Common  opera?ons  performed  at  CDB  level  

Root  

  Shared  memory  and  background  processes   •   More  applicaFons  per  server  

CDB  

•  Manage  many  as  one  (upgrade,  backups,  HA)   •  Granular  control  when  appropriate   •  Simple  DR  

 

MAA  and  Mul?tenant  

•  SoluFons  for  planned  /  unplanned  outages  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

25  

Single  PDB  Failover/RelocaFon   •  Loss  of  single  PDB  at  primary   •  All  other  PDBs  operaFng  normally   •  Standby  copy  of  PDB  prisFne   •  Current  soluFons?     •  Failover  enFre  CDB?   •  Restore/Recover  PDB?    

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

Single  PDB  Failover/RelocaFon   •  Use  standby  copy  to  “unplug”  PDB  (MOS  2088201.1)   •  Create  manifest  with  dbms_pdb.describe   •  Plug  in  PDB’s  standby  files  into  co-­‐located  CDB   with  NOCOPY  (quick)   •  Migrate  Services  to  new  CDB   •  Broker  enhanced  to  automate  process  (12.2)   •  Source  can  be  primary  or  standby   •  AutomaFon  will  not  create  the  desFnaFon  CDB  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

Normal  RunFme  

PDB1  

PDB2  

PDB3  

PDB1  

PDB2  

CDB  1  

Read-­‐Write  

Data  Guard  

PDB3  

CDB  1   Standby  

Read-­‐  Only  

PDB4  

CDB  2  

Read-­‐Write  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

PDB  2  Outage  

Unplug/plug  PDB2  from  CDB1  standby  to  CDB2  and  failover  applica?on  connec?ons  

PDB1  

PDB2  

PDB3  

PDB1  

PDB2   PDB3  

CDB  1  

Read-­‐Write  

Data  Guard  

CDB  1   Standby  

Read-­‐  Only  

PDB4  

PDB2   CDB  2  

Read-­‐Write  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

MAA  Best  PracFces   GoldenGate  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

30  

Oracle  GoldenGate   Flexible  Logical  Replica?on  

LAN / WAN / Internet Over TCP/IP

Source  &  Target   Oracle  &  Non-­‐Oracle  Database(s)  

Bi-­‐direcFonal  

Target  &  Source   Oracle  &  Non-­‐Oracle  Database(s)  

•  Zero-­‐downFme  maintenance  and  migraFons   •  AcFve-­‐AcFve  high  availability   •  Heterogeneous  replicaFon,  data  distribuFon  and  integraFon  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

31  

MAA  Best  Prac?ces:   Transparent  Role   Transi?ons  with   Data  Guard  and   Oracle  GoldenGate  

Oracle  GoldenGate   MAA  Best  Prac?ces  

•  Transparent  Role  TransiFons  in  a  Data  Guard  ConfiguraFon   – No  manual  intervenFon  is  required  with  FSFO  and  DG  Broker  

•  ConfiguraFon  makes  use  of:   •  Oracle  Grid  Infrastructure  Bundled  Agent  (XAG)   •  DBFS  for  shared  GoldenGate  files  (trails  and  checkpoint  files)   •  Role  based  services     Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

32  

Sample  Deployment   Observer  

 

Standby  Database  

Primary  Database   Integrated   Extract  

LogMining   Server   Redo  Transport   (SYNC  or  ASYNC)  

BidirecFonal   GoldenGate  ReplicaFon  

Trail  and  other  OGG  Files   In  DBFS  

Redo  Transport   OCI  ConnecFon     File  I/O   Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

Warehouse   33  

Sample  Deployment  –  Post  Role  TransiFon   Observer  

 

(OLD)  Primary  Database  

(NEW)  Primary  Database   LogMining   Server  

LogMining   Server  

Integrated   Extract  

Redo  Transport   (SYNC  or  ASYNC)  

Trail/Checkpoint/BR  Files   In  DBFS  

BidirecFonal   GoldenGate  ReplicaFon  

Redo  Transport   OCI  ConnecFon     File  I/O  

Warehouse   Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

34  

MAA  Best  PracFces   Sharding  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

35  

Oracle  Sharding  –  A  Database  Architecture  Palern     •  Horizontal  parFFoning  of  data  across  a   farm  of  independent  databases  using  a   sharding  key  (i.e.  account_id)  

One  giant  database  parFFoned  into     many  small  databases  (shards)  

•  Some  Web-­‐scale  OLTP  applicaFons  use   database  sharding   –  Avoid  scalability  or  availability  edge  cases  of  a   single  large  system  image  database   –  Willing  to  customize  data  model  and   applicaFons  to  enable  transacFons  to  be   automaFcally  routed  to  the  right  shard  

•  NoSQL  databases  made  it  easy  to  deploy   sharding,  now  Oracle  Database  Na?ve   Sharding  also  makes  it  easy   Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

36  

Oracle  Database  Sharding  –  Benefits   Linear  Scalability  

…  

Add  shards  online  to  scale   transac?ons  and  concurrent  users.   Online  rebalance.  

Fault  Tolerant  

Geographic  Distribu?on  

…  

No  shared  hardware  or   solware    to  isolate  faults.   Shards  may  run  different    Oracle   releases.    

User  defined  data  placement  for   performance,  availability,  DR  or  to   meet  regulatory  requirements.  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

37  

Deployment  of  a  System-­‐Managed  SDB  with  Data  Guard   Shard  Director   shdir1,2  

Region   Availabilty_Domain1  

Shardgroup   shgrp1  

Shard  Catalog   shardcat  

Primaries   Clients  

Region   Availability_Domain2  

ConnecFon   Pools  

…  

HA  Standbys  

ConnecFon   Pools  

…  

Shard  Director   shdir3,4

Shard  Catalog  

Shardgroup   shgrp2  

shardcat_stdby  

Data  Guard   Fast-­‐Start  Failover     Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

38  

Oracle  Sharding  –  MAA  Outage  TesFng   •  Outage  of  Shard  Catalog  has  no   effect  on  applicaFon   performance   •  Shard  Keys  are  cached  within  the   shard  directors   •  MAA  Best  PracFce  is  to  protect   catalog  with  Data  Guard   Maximum  Availability  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

39  

Oracle  Sharding  –  MAA  Outage  TesFng   •  Outage  of  shard  directors  does   not  affect  a  running  connecFon   pool   •  ConnecFon  pool  caches  range  of   shard  keys  /  shards   •  MAA  best  pracFce  to  have  3   shard  directors  per  region  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

40  

Oracle  Sharding  –  MAA  Outage  TesFng   Failover  Performance  

•  Failover  of  an  individual  shard   does  not  affect  applicaFon   performance  for  remaining   shards  

4500   4000  

3000   2500   Read/Write  

2000  

Read  Only  

1500  

18:13:53  

18:13:50  

18:13:47  

18:13:44  

18:13:41  

18:13:38  

18:13:35  

18:13:32  

18:13:29  

18:13:26  

18:13:23  

18:13:20  

18:13:17  

18:13:14  

18:13:11  

18:13:08  

18:13:05  

18:13:02  

18:12:59  

18:12:56  

18:12:53  

500   0  

•  Fast  failover  for  both  read  /  write   and  read  only  connecFons   •  Generic  MAA  best  pracFces  apply   for  sharded  environments  

1000  

18:12:50  

Transac?ons  Per  Second  

3500  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

41  

Demo   Sharding  on  Oracle  Bare  Metal    Cloud  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

42  

MAA  Best  PracFces   Oracle  Cloud  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

43  

Oracle  MAA  for  Oracle  BMC  Database  Service   • Bare  Metal  Cloud  (BMC)  and  MAA  teams  collaborate  to  validate   all  MAA  soluFons  in  BMC  environment   • New  BMC/MAA  best  pracFce  paper  that  outlines  soluFons   – BMC  highly  redundant  infrastructure   – Data  Guard  for  near  or  zero  data  loss   – Golden  Gate   – Sharding  

• Coming  soon  –  MAA  validated  managed  services  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |   Oracle  ConfidenFal  –  Internal/Restricted/Highly  Restricted  

44  

Zero  Data  Loss  with  Minimal  Performance  Impact   SYNC  Performance  Impact   2500  

   

1500   async   fastsync   1000  

500  

0  

FAST  SYNC  

SYNC  

Redo  Rate  (MB/sec)  

14.93  

14.48  

14.39  

Block  Changes/sec     (KB/sec)  

96.92  

94.3  

93.86  

Txn  Rate    

2082  

2025  

2018  

%  Difference  from   ASYNC  

N/A  

97%  

97%  

sync  

1   14   27   40   53   66   79   92   105   118   131   144   157   170   183   196   209   222   235   248   261   274   287   300  

Transac?ons  Per  Second  

2000  

ASYNC  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |   ConfidenFal  –  Oracle  Internal    

Failover  Performance  

•  Swingbench  OLTP  applicaFon   performing  mixture  of  inserts,   updates,  and  deletes  

Database  Failover  -­‐  8  Seconds  

Database  Failover  with  Minimal  DownFme  

•  ApplicaFon  redo  rate  of  15MB/ sec  

4500   4000  

FSFO  Threshold  -­‐  6  Seconds  

3000   2500   2000   1500   1000  

Read/Write   Read  Only  

18:13:53  

18:13:50  

18:13:47  

18:13:44  

18:13:41  

18:13:38  

18:13:35  

18:13:32  

18:13:29  

18:13:26  

18:13:23  

18:13:20  

18:13:17  

18:13:14  

18:13:11  

18:13:08  

18:13:05  

18:13:02  

18:12:59  

18:12:56  

0  

18:12:53  

500  

18:12:50  

Transac?ons  Per  Second  

3500  

•  Fast  Start  failover  in  Maximum   Availability  mode,  FSFO  threshold   configured  for  6  seconds   •  Database  failover  Fme  of  8   seconds  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |   ConfidenFal  –  Oracle  Internal    

Planned  Maintenance  with  Minimal  DownFme   Switchover  Performance  

•  Swingbench  OLTP  applicaFon   performing  mixture  of  inserts,   updates,  and  deletes  

4500   4000  

3000   2500   Read  Write  

2000  

Read  Only  

1500   1000  

18:46:53  

18:46:50  

18:46:47  

18:46:44  

18:46:41  

18:46:38  

18:46:35  

18:46:32  

18:46:29  

18:46:26  

18:46:23  

18:46:20  

18:46:17  

18:46:14  

18:46:11  

18:46:08  

18:46:05  

18:46:02  

18:45:59  

18:45:56  

0  

18:45:53  

500  

18:45:50  

Transac?ons  Per  Second  

3500  

•  ApplicaFon  redo  rate  of  15MB/ sec   •  ApplicaFon  outage  of  12  seconds   during  the  swFchover  process   •  Total  switchover  Fme  of   approximately  40  seconds  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |   ConfidenFal  –  Oracle  Internal    

Oracle  Sharding  Demo  on  Bare  Metal  IaaS  Cloud   •  ElasFcally  scaled-­‐out  to  200  shards  on  Oracle  Bare  Metal  Cloud   – Demonstrated  linear  scalability  of  RelaFonal  transacFons     – Demonstrated  11  Million  user  calls  per  sec  that  includes:   •  4.5  Million  Database  TransacFons  per  sec  across  all  100  Primary  shards   •  6.5  Million  Queries  per  sec  across  all  100  AcFve  Standby  shards  

•  MAA  Sharding  provides  highest  availability     – Each  shard  is  protected  by  Data  Guard  Fast-­‐Start  Failover  across  Availability  Domains   – Single  Shard  Failure  resulted  in  100%  availability  for  99%  of  the  applicaFon   •  1%  of  the  applicaFon  experienced  only  15  seconds  blackout  

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

Program  Agenda   1  

Oracle  MAA  

2  

MAA  Best  PracFces  

3  

FIS    

Copyright  ©  2016,  Oracle  and/or  its  affiliates.  All  rights  reserved.    |  

49  

FIS and Oracle Exadata FIS Maximum Availability Architecture

Manish Upadhyay ,Lead Oracle Machine Administrator 09/20/2016

About Me •  4+ years with FIS •  8 years at hedge funds •  20+ years working with Oracle technologies and databases •  Team lead of database appliance •  Oracle Certified Professional •  Supporting over 500 databases across a variety of platforms and applications

About FIS •  Fortune 500 company •  World's largest global provider dedicated to banking and payments technologies •  20,000 clients in over 130 countries •  Powering billions of transactions annually that move over $9 trillion around the globe •  55,000 Employees / HQ Jacksonville, FL

Challenge •  Long time and business delay to stand up remote site to become active on failover - Start database - Re-map storage volumes - Application reconnect

Symantec VCS Cluster

•  SUN M5000 Server •  IBM XIV Storage

•  SUN M5000 Server •  IBM XIV Storage

•  Unreliable HA •  Delayed recovery with tape archive •  Reliance on backup team and concern over backup reliability

Primary Site

Failover Site

ACTIVE

PASSIVE

Migration – Oracle GoldenGate •  “GoldenGate is your friend” •  Powerful database replication technology •  Hardware independent; cross platforms Oracle or non-Oracle •  Database agnostic

Success with Exadata

Exadata & MAA at FIS for most recent project

Active Data Guard

•  No unplanned downtime •  Active Data Guard •  Ability to redirect application tier quickly

Exadata 1

Exadata 2

Exadata 1

Exadata 2

•  Database protection •  DR to remote site to become active on failover in real-time

ZFS Appliance-backup

ZFS Appliance-backup

Primary Site

Remote Site

ACTIVE

ACTIVE

Suggest Documents