sci lab test validation report: netapp storage efficiency

21
Silverton Consulting, Inc. StorInt™ Briefing SCI Lab Test Validation Report: NetApp Storage Efficiency Written by: Ray Lucchesi Published: July 2012

Upload: netapp

Post on 20-Aug-2015

168 views

Category:

Technology


4 download

TRANSCRIPT

Page 1: SCI Lab Test Validation Report: NetApp Storage Efficiency

   

   

Silverton Consulting, Inc. StorInt™ Briefing

               

SCI Lab Test Validation Report: NetApp Storage Efficiency

                 Written by: Ray Lucchesi Published: July 2012

Page 2: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  1     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

Executive  Summary  Silverton  Consulting  tested  a  number  of  NetApp’s  widely  used  software  storage  efficiency  features  on  a  FAS3240  storage  system  using  a  mix  of  data  types.  The  testing  was  designed  to  measure  the  cumulative  impact  of  multiple  efficiency  technologies  when  used  together,  and  as  a  result,  several  test  phases  were  required.    The  first  phase  tested  three  NetApp  storage  efficiency  features.    Specifically,  the  storage  system  was  thick  provisioned  to  create  a  baseline  and  then  cumulatively  thin  provisioned,  deduplicated  and  compressed,  all  using  the  same  data.    Storage  capacity  requirements  were  assessed  after  each  step  to  measure  any  savings  that  occurred.    At  the  end  of  this  phase,  the  thinly  provisioned,  deduplicated  and  compressed  storage  saved  an  impressive  79%  when  compared  with  the  baseline  capacity  requirements.    The  second  phase  examined  two  NetApp  copy  services:  Snapshot™  and  FlexClone™.  During  this  phase,  the  storage  at  the  end  of  the  prior  phase  was  first  Snapshot  copied  and  then  FlexClone  copied.    Capacity  requirements  were  again  evaluated  after  each  NetApp  copy  had  completed  to  compare  against  the  baseline.  Once  again,  storage  capacity  requirements  for  both  copies  were  substantially  less  than  baseline.    The  final  phase  ran  a  mixed  workload  against  the  FlexClone  copies  of  the  previous  phase  to  determine  how  capacity  efficiency  and  copy  services  impacted  performance  for  the  storage  system  under  test.    In  the  first  phase  above,  after  creating  the  storage  baseline,  a  set  of  database,  email  and  file  system  workloads  were  run.    For  this  phase  those  workloads  were  run  once  more,  only  this  time  against  the  deduplicated,  compressed,  and  FlexClone  copied  data.    This  final  step  showed  little  to  no  performance  degradation  when  using  deduplicated,  compressed  and  cloned  data  as  compared  against  baseline  performance.        

Table  of  Contents  

Executive  Summary  Introduction  Test  Step  1:  Baseline  

Cumulative  storage  efficiency  Test  Step  2:  Thin  provisioning  Test  Step  3:  Data  deduplication  Test  Step  4:  Data  compression  

Copy  services    Test  Step  5:  Snapshot  copy  Test  Step  6:  FlexClone  copy  

Performance  Test  Step  7:  Performance  

Summary  Appendices    

Page 3: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  2     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

SUMMARY:  Cumulative  Effect  of  NetApp  Storage  Efficiency  Features  

SUMMARY:  Savings  from  NetApp  Space  Efficient  Copy  Features  

 

Page 4: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  3     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

Introduction    NetApp  recently  contracted  with  Silverton  Consulting  Inc.  (SCI)  to  independently  validate  the  substantial  storage  capacity  savings  attainable  with  NetApp  systems.    The  three  storage  efficiency  features  rigorously  tested  included  thin  provisioning,  data  deduplication  and  data  compression.    NetApp  further  engaged  SCI  to  verify  the  storage  copy  efficiency  capabilities  of  NetApp’s  Snapshot  and  FlexClone  features.1    In  a  final  corollary  phase  of  testing,  SCI  was  asked  to  independently  authenticate  the  I/O  performance  of  the  storage  system  with  data  already  subjected  to  the  efficiency  features  under  trial.    

NetApp  storage  system  test  environment  One  NetApp  FAS3240  storage  system  running  Data  ONTAP  8.1  operating  in  7-­‐Mode,  with  ~20TB  of  SAS  disk  and  10GbE  interfaces  was  installed  at  SCI’s  lab.    The  storage  was  arranged  as  a  RAID-­‐DP  configuration  (19-­‐data  and  2-­‐parity,  using  450  GB  SAS  disk  drives)  with  two  (2TB)  aggregates  and  five  user  volumes:    

• One  volume  with  629GB  of  storage  for  a  file  system,    • Two  volumes  configured  as  a  629GB  iSCSI  LUN  for  SQL  Server  database  

tables  and  the  second  as  a  SQL  log  volume  of  ~105GB  iSCSI  LUN,    • Two  more  volumes  to  hold  iSCSI  LUNs,  one  configured  with  629GB  for  a  

Microsoft  Exchange  database  and  the  other  LUN  with  ~42GB  for  an  Exchange  log  file.      

 The  storage  system  was  equipped  with  8GB  of  system  RAM/cache  and  was  connected  to  lab  servers  using  three  separate  10GbE  interfaces,  one  for  the  file  workload  and  two  for  the  SQL  database  and  email.    No  FlashCache  or  Fibre  Channel  interfaces  were  used  during  the  testing.    The  specific  NetApp  system  options  employed  to  store  this  data  were  as  follows:    

• Volume  Space  Guarantee  =  volume  • LUN  Set  Reservation  =  enable  • Fractional  Reserve  =  100%  • Snapshot  Reserve  (Aggregate  and  Volume)  =  5%.    

These  system  storage  parameters  were  selected  to  establish  a  thickly  provisioned  baseline  and  insure  that  any  space  savings  or  consumption  afforded  by  the  storage  efficiency  features  under  further  evaluation  would  be  more  accurately  assessed.    That  is,  the  performance  and  capacity  results  experienced  would  be  due  solely  to  the  storage  efficiency  feature  under  test.  

                                                                                                               1  For  a  customer  case  study  on  NetApp  storage  efficiency  features  see  our  report  on  Achieving  Exceptional  Storage  Efficiency  with  NetApp  Storage  available  at  http://media.netapp.com/documents/ar-­‐exceptional-­‐data-­‐density.pdf  

Page 5: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  4     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

Actual  test  process  The  actual  test  was  a  multi-­‐step  task  where  data  was  loaded  to  the  storage  then  capacity  measurements  using  NetApp/Windows  facilities  were  taken  initially  to  establish  baseline.  The  same  measurements  were  then  taken  again  after  thin  provisioning,  data  deduplication  and  data  compression  were  progressively  enabled  and  the  resulting  transformation  process  was  complete.  Additionally,  capacity  measurements  were  taken  after  the  workload  was  run  to  isolate  and  identify  the  data  growth  due  solely  to  the  workload  process.            The  actual  workload  mixture  used  in  the  testing  process  was  specifically  designed  to  more  closely  emulate  and  approximate  realistic  operating  conditions  for  a  storage  system.    In  addition,  the  three  types  of  workloads  were  run  simultaneously  to  better  mirror  real  shared  storage  use  operations;  running  any  one  of  these  workloads  in  isolation  would  have  generated  significantly  different  throughput.          For  performance,  measurements  were  taken  only  after  the  baseline  data  was  loaded  and  then  again  after  all  storage  efficiency  features  were  enabled.    The  measurements  were  reported  on  minutely  over  the  concurrently  working  simulated  file,  SQL  database  and  email  workload  runs.    These  measurements  were  derived  by  using  Window’s  Perfmon,  running  on  each  of  the  three  VMs,  executing  the  different  workloads.    Using  this  approach,  individual  performance  measurements  for  each  of  the  three  workloads  was  determined.    Because  of  the  realistic  workload  design,  high  variability  of  performance  measurements  was  expected  and  in  fact,  experienced  during  evaluation.    As  such,  average  performance  was  used  to  compare  throughput  operations  between  the  baseline  and  final  all  features  test  step.          

Test  Step  1:  Baseline  To  measure  subsequent  capacity  savings  and  performance,  the  measurement  tools  were  used  to  establish  both  baseline  numbers  after  the  data  had  been  loaded  onto  the  storage  as  well  as  after  a  simulated  workload  run.  That  is,  the  data  was  initially  loaded  and  a  workload  processed  with  no  storage  efficiency  features  enabled;  the  resulting  measurements  were  the  baseline  numbers  for  subsequent  comparisons.      Figure  1  Baseline  capacity  requirements  summary  chart  

Page 6: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  5     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

Baseline  capacity  parameters    After  the  initial  data  was  loaded  but  before  any  storage  efficiency  features  were  enabled,  the  capacity  measurements  reported  by  the  Windows  host  validated  the  NetApp  storage  system  measurements.    In  fact,  the  reported  measurements  were  identical  and  as  follows:      

• File  system  storage    -­‐  629.1GB  • SQL  DB  storage  –  629.1GB  • SQL  Log  storage  –  104.9GB  • Email  DB  storage  –  629.1GB  • Email  log  storage  –  41.9GB  • Total  baseline  storage  capacity  –  2.0TB.  

 

Baseline  performance    

Figure  2  Baseline  performance  run    

 Figure  2  graphs  the  performance  achieved  by  the  NetApp  storage  without  enabling  any  capacity  efficiency  features.    As  can  be  seen,  each  type  of  workload,  experienced  wide  variability  as  follows:      

Page 7: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  6     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

• The  file  workload  varied  between  a  high  of  ~70  MB/sec.  to  a  low  of  ~3  MB/sec.,  

• The  SQL  Server  workload  varied  between  ~134  and  ~54  MB/sec.,  and  • The  email  workload  varied  between  ~37  and  ~28  MB/sec.    

 However,  average  baseline  performance,  also  depicted  in  Figure  2,  showed  mean  throughput  as  follows:    

• File  services  average  performance  was    ~25  MB/sec.  • SQL  Server  DB  average  performance  was    ~97  MB/sec.  • Email  average  performance  was    ~33MB/sec.  

Cumulative  storage  efficiency  tests  During  this  phase  of  the  testing,  we  enabled  NetApp  thin  provisioning,  data  deduplication  and  data  compression  features  against  the  test  data  created  during  the  baseline  test  step  above.    The  intent  of  this  phase  of  the  testing  was  to  determine  what  if  any  storage  capacity  requirements  could  be  saved  by  an  aggressive  use  of  these  features.  

Test  Step  2:  Thin  provisioning  Although  not  required  to  apply  thin  provisioning,  the  data  was  reloaded  in  order  to  start  from  the  same  conditions,  then  thin  provisioning  was  enabled  in  the  next  trial  iteration  by  setting  “Vol  options  guarantee=none”  and  “LUN  set  reservation=disable”  for  each  volume  and  LUN.  The  thin  provisioning  feature  saved  capacity  by  freeing  up  unused  space  in  partially  used  volumes  and  LUNs.    Thin  provisioning  also  allowed  the  creation  of  many  more  file  systems  and  LUNs  on  the  storage  system    (‘oversubscription’).    Substantial  savings  were  anticipated  but  were  

dependent  on  how  much  empty,  yet  reserved  space  had  been  allocated  to  each  volume  as  a  result  of  using  thick  provisioning.  

Thin  provisioning  capacity  requirement  savings  

Figure  3  clearly  shows  the  dramatic  reduction  in  storage  capacity  requirements  that  enabling  thin  provisioning  afforded.    However,  this  savings  was  entirely  dependent  on  the  amount  of  allocated  and  never  written  space  available.      Comparing  the  baseline  capacity  to  

Figure  3  Thin  provisioning  capacity  requirements    

Page 8: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  7     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

the  pre-­‐workload  capacity  requirements  using  thin  provisioning,  the  following  available  storage  capacity  requirements  savings  percentages  were  derived  on  this  pass  of  the  test:    

• File  system  storage:  211.7GB,  a  substantial  66%  savings  over  baseline  capacity.    With  thin  provisioning,  the  file  system  reserved  only  as  much  space  as  data  written,  releasing  significant  storage  capacity  for  other  use.  

• SQL  DB  storage:  473.6GB,  a  moderate  savings  of  25%  over  baseline  capacity.    Thin  provisioning  freed  up  all  of  the  SQL  DB  LUN’s  reserved  space  that  had  yet  to  be  written.  

• SQL  log  storage:  1.2GB,  an  outstanding  savings  of  99%  over  baseline  capacity.    Much  if  not  all  of  the  log  space  had  never  been  written.  

• Email  database  storage:  414.0GB,  a  significant  savings  of  34%  over  baseline  capacity.    Similarly,  thin  provisioning  freed  up  all  email  database  reserved  space.  

• Email  log  storage:  0.1GB,  another  outstanding  savings  of  over  99%  from  baseline  capacity.    Again,  the  same  as  that  described  above.  

 Overall,  thin  provisioning  saved  a  remarkable  45+  percent  of  the  capacity  used  in  the  baseline  step.    It  should  be  noted  that  actual  storage  efficiency  measurements  for  this  and  all  remaining  steps  was  calculated  solely  from  internal  NetApp  storage  commands.    The  Windows  command  that  normally  displays  storage  capacity  does  not  recognize  thin  provisioning,  deduplication  or  compression  and  thus,  does  not  report  on  capacity  savings  or  any  measurement  to  derive  capacity  savings.    In  this  step,  the  NetApp  CLI  “df  -­‐k”  command  was  used.        

Test  Step  3:  Data  deduplication  The  next  storage  efficiency  feature  enabled  in  the  trial  was  data  deduplication  on  top  of  the  already  thinly  provisioned  storage.    This  feature  was  enabled  and  then  run  by  issuing    “sis  on”  and  “sis  start  –s”  commands  at  the  volume  level.    NetApp’s  deduplication  feature  was  expected  to  reduce  storage  used  by  eliminating  duplicate  4KB  data  blocks  within  a  volume.    However,  the  anticipated  savings  were  expected  to  vary  significantly  depending  on  the  amount  of  duplicate  blocks  present  from  volume-­‐to-­‐volume.    Storage  efficiency  was  calculated  using  the  NetApp  CLI  “df  –S”  command.2  

Data  deduplication  capacity  requirement  savings  

As  shown  in  Figure  4  below,  data  deduplication  resulted  in  additional  storage  capacity  savings.    The  significant  reduction  in  used  storage  space  was  realized  

                                                                                                               2  NetApp  has  written  a  guide  to  implementing  deduplication  that  can  be  found  at  http://media.netapp.com/documents/tr-­‐3958.pdf  

Page 9: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  8     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

almost  entirely  due  to  the  email  and  file  system  data  being  responsive  to  the  dedupe  process.    Actual  capacity  savings  after  data  deduplication  were  as  follows:    

• File  system  storage:  118.3GB  of  data  stored,  a  44%  incremental  savings  over  thin  provisioning  capacity  requirements.  

• SQL  DB  storage:  452.4GB  of  data  stored,  a  slight,  4%  incremental  savings  over  thin  provisioning  capacity.  

• SQL  log  storage:  18MB  of  data  stored,  a  99%  incremental  savings  over  thin  provisioning  capacity.  

• Email  database  storage:  87.0GB  of  data  stored,  an  impressive  79%  incremental  savings  over  thin  provisioning  capacity.  

• Email  log  storage:  2MB  of  data  stored,  a  97%  incremental  savings  over  thin  provisioning  capacity.  

 Overall,  data  deduplication  saved  an  additional  40+  percent  of  the  capacity  used  in  the  thin  provisioning  step.  

Test  Step  4:  Data  compression  Data  compression,  a  compute  intensive  efficiency  feature,  was  enabled  for  the  thinly  provisioned  and  deduplicated  storage  for  the  fourth  pass  by  issuing  a  “sis  config  –C  TRUE”  command  followed  by  initiating  compression  using  the  “sis  start  –S  –C”  command  at  the  volume  level.    This  command  scanned  all  current  volume  and  LUN  data  and  automatically  compressed  it.    This  compression  activity  of  the  original  data  was  completed  prior  to  any  further  testing  steps.    However,  by  not  using  the  “-­‐I”  option  in  the  command  above,  offline  compression  was  activated.    NetApp  does  offer  inline  compression  but  offline  was  used  to  more  closely  emulate  a  customer  that  wanted  the  space  savings  of  compression  but  executed  off  hours  to  minimize  the  impact  on  daily  IO  activity.    The  data  compression  feature  was  expected  to  increase  free  capacity  by  reducing  repeating  patterns  of  data  within  the  volume.  Storage  efficiency  was  calculated  using  the  NetApp  CLI  “df  –S”  command.    

Data  compression  capacity  requirement  savings  

As  shown  below  in  Figure  5,  storage  savings  were  moderate  using  the  data  compression  feature  against  previously  deduplicated  and  thinly  provisioned  data.    However,  these  realized  savings  were  only  modest  due  to  the  inherent  compressibility  of  the  data.    That  is,  image  and  zipped  or  archive  (already  compressed)  files  did  not  further  compress  well  whereas  Microsoft  Office  files  were  

Figure  4  Data  deduplication  capacity  requirements  

Page 10: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  9     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

compressed  by  50  percent  or  more.    Database  and  email  compressibility  rates  also  varied  considerably.      

 In  this  step,  actual  capacity  savings  after  data  compression  were  as  follows:    

• File  system  storage:  106.1GB  of  data  stored,  a  10%  incremental  savings  over  capacity  present  for  the  data  deduplication  step.    This  only  modest  savings  was  primarily  due  to  the  nature  of  

the  test  file  data,  which  consisted  of  email  and  incompressible,  image  data.    • SQL  DB  storage:229.2GB  of  data  stored,  a  49%  incremental  savings  over  

data  deduplication  capacity,  primarily  due  to  the  amount  of  text  and  web  log  data  present  in  the  tables.  

• SQL  log  storage:  17.8MB  of  data  stored,  a  slight  2%  incremental  savings  over  data  deduplication  capacity.    

• Email  database  storage:  87.0GB  of  data  stored,  a  minimal  <1%  incremental  savings  over  data  deduplication  capacity  due  to  the  nature  of  the  test  data  used  for  email  data.  

• Email  log  storage:  1.8MB  of  data  stored,  a  13%  incremental  savings  over  data  deduplication  capacity.    

 Overall,  compression  saved  an  additional  36  percent  of  the  capacity  used  in  the  deduplication  step.    

Copy  services  tests  After  storage  capacity  measurements  for  thin  provisioning,  data  deduplication  and  data  compression  were  established,  a  single  set  of  Snapshot  and  FlexClone  copies  were  taken  of  the  test  data.    This  was  done  to  ascertain  capacity  requirement  savings  provided  by  NetApp’s  point-­‐in-­‐time  volume  and  LUN  storage  copies,  i.e.  read-­‐only  Snapshot  copies  and  read-­‐write  FlexClone  copies.  Then  the  workload  was  run  against  the  FlexClone  copies.      Both  Snapshot  and  FlexClone  copy  capacity  requirements  were  expected  to  be  significantly  smaller  than  source  data  capacity  requirements.    However,  this  presented  a  significant  dilemma  as  to  when  to  measure  Snapshot  and  FlexClone  

Figure  5  Data  compression  capacity  requirements    

Page 11: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  10     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

capacity  requirements.    There  are  at  least  two  very  different  alternatives:  1)  Measure  copy  capacity  requirements  before  a  performance  workload  was  run  against  the  source  data  and  2)  Measure  copy  capacity  requirements  after  a  performance  workload  was  run  against  the  source  data.    Pre-­‐workload  Snapshot  and  FlexClone  copies  only  store  meta-­‐data  to  describe  the  data  being  copied  and  points  to  the  original  source  data.    In  contrast,  a  post-­‐workload  Snapshot  and  FlexClone  copies  must  store  this  meta-­‐data  plus  any  original  data  that  was  modified,  thus  consuming  more  storage  capacity.    As  a  result,  post-­‐workload  copy  capacity  requirements  were  measured  and  compared  with  the  post-­‐workload  baseline  capacity  measured  in  Test  Step  1  (see  p.  4).    

Test  Step  5:  Snapshot  copy  In  this  step,  Snapshot  copies  were  taken  of  the  data  by  using  the  “snap  create”  NetApp  command.      In  Figure  6  below,  post-­‐workload  Snapshot  copies  capacity  requirements  were  measured  and  compared  against  the  baseline  capacity  after  the  workload  was  run.    The  Snapshot  copies  were  expected  to  be  significantly  smaller  than  source  data  as  any  storage  capacity  consumed  should  only  represent  data  modified  from  the  original.  

Snapshot  copy  capacity  requirement  savings  

Figure  6  clearly  shows  that  the  capacity  requirements  for  the  post-­‐workload  set  of  Snapshot  copies  were  significantly  smaller  than  the  post-­‐workload  baseline  source  data.    The  capacity  consumed  by  Snapshot  copies  only  slightly  registered  on  the  chart  as  it  represented  the  incremental  space  required  to  store  any  changes  to  the  source  data.    Actual  post-­‐workload  capacity  measurements  for  the  Snapshot  copies  were  as  follows:    

• File  system  Snapshot  storage:  6.8GB  of  data  stored,  an  outstanding  99%  savings  over  the  capacity  present  for  the  baseline  data  file  system.    

• SQL  DB  Snapshot  storage:  90.1GB  of  data  stored,  an  86%  savings  over  the  capacity  present  in  the  baseline  SQL  data.  

• SQL  log  Snapshot  storage:  7MB  of  data  stored,  a  100%  savings  over  the  capacity  present  in  the  baseline  email  log  data.    

• Email  database  Snapshot  storage:  7.5GB  of  data  stored,  a  99%  savings  over  

Figure  6  Snapshot  copy  capacity  requirements  

Page 12: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  11     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

the  capacity  present  in  the  baseline  email  database.  • Email  log  Snapshot  storage:  <1MB  of  data  stored,  a  100%  savings  over  the  

capacity  present  in  the  baseline  email  log  data.      Of  note,  NetApp  Snapshot  capacity  is  entirely  contingent  upon  the  amount  of  data  modified  since  the  original  Snapshot  copies  were  taken.    Thus,  heavily  modified  data  will  consume  more  Snapshot  space  and  may  grow  over  time  as  the  source  data  is  updated.      

Test  Step  6:  FlexClone  copy  As  the  next  step  in  the  rigorous  testing  of  NetApp’s  storage  features,  measurements  were  derived  after  taking  NetApp  FlexClone  copies,  another  type  of  space  efficient,  point-­‐in-­‐time  copy  of  source  data.    These  copies  differed  from  NetApp  Snapshot  copies  because  they  could  be  written  as  well  as  read.    Once  again  post-­‐workload  FlexClone  capacity  requirement  measurements  were  measured  and  compared  to  post-­‐workload  baseline  numbers  to  determine  the  capacity  requirement  savings.    Once  more,  significant  storage  capacity  requirement  savings  were  anticipated  for  these  copies  of  the  source  data.  

FlexClone  copy  capacity  requirement  savings  

As  expected,  the  numbers  generated  in  the  trial  and  depicted  in  Figure  7,  shows  the  significant  storage  capacity  requirement  savings  available  by  taking  a  FlexClone  copy  of  the  source  data.    In  this  step,  actual  post-­‐workload  FlexClone  capacities  were  as  follows:    

• File  system  FlexClone  storage:  66.6GB,  an  89%  savings  over  the  capacity  present  in  the  baseline  data.    

• SQL  DB  FlexClone  storage:  200.9GB,  a  68%  savings  over  the  capacity  present  in  baseline  SQL  data.  

• SQL  log  FlexClone  storage:  65.2GB,  a  34%  savings  over  the  capacity  present  in  the  baseline  SQL  log  data.  

• Email  database  FlexClone  storage:  115.8GB,  an  82%  savings  over  the  capacity  present  in  the  baseline  email  data.      

• Email  log  FlexClone  storage:  349MB,  an  89%  savings  over  the  capacity  present  in  the  baseline  email  log  data.  

 

Figure  7  FlexClone  copy  capacity  requirements  

Page 13: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  12     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

Similar  to  Snapshot  copies,  space  savings  from  FlexClone  copies  depended  on  the  amount  of  data  modified  from  the  original  source  storage.    However,  calculating  the  space  used  by  FlexClone  copies  was  more  complex.  In  this  case,  the  NetApp  “vol  clone  split  estimate”  command  was  relied  on  to  provide  the  amount  of  space  shared  between  the  source  data  and  its  clone.    The  space  consumed  by  the  clones  was  then  calculated  as  the  difference  between  the  capacity  used  by  the  FlexClone  data  and  the  estimate  of  shared  storage.  

Performance  testing  

Test  step  7:  Thin  provisioning,  deduplication,  compression,  Snapshot  and  FlexClone  performance  results  After  all  capacity  efficiency  features  and  copy  services  discussed  above  were  enabled,  baseline  workloads  were  rerun  to  determine  their  impact  on  storage  system  performance.    As  discussed  above,  all  the  workloads  were  run  against  FlexClone  copies  with  thin  provisioning,  deduplication,  compression  and  Snapshot  copy  enabled  and  compared  against  a  similar  workload  run  against  the  original  baseline  data  to  test  how  these  features  and  copy  services  would  impact  storage  performance.    System  capacity  requirements  did  not  change  from  previous  steps  and  have  thus,  not  been  reported  on  again  (see  pp.  9,  10  &  11).  

Performance  results  after  capacity  efficiency  and  copy  services  were  enabled  

In  Figure  8  below  both  the  baseline  and  the  capacity  efficiency  and  copy  services  run  results  were  shown  side-­‐by-­‐side  to  facilitate  easy  comparison.    Some  impact  from  all  the  storage  features  was  expected,  but  system  performance  significantly  exceeded  predictions.    

Page 14: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  13     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

 Figure  8  Baseline  vs.  all  features  performance  comparison  chart  

 In  fact,  enabling  NetApp’s  space  saving  features  of  thin  provisioning,  data  deduplication,  compression  along  with  Snapshot  and  FlexClone  copy  actually  had  a  positive  effect  on  storage  performance  during  some  of  our  testing.  Specifically,  no  negative  performance  was  seen.    Performance  of  the  all  features  enabled  workloads  were  as  follows:    

• Average  SQL  DB  performance:  118  MB/sec.,  an  improvement  of  22%  versus  the  baseline  performance.      

• Average  email  performance:  40  MB/sec.,  an  improvement  of  24%  over  baseline  performance.  

• Average  file  system  performance:  only  a  slight  negative  performance  impact,  24  MB/sec.,  for  only  a  minor,  <1%  degradation  over  baseline  performance,  which  could  arguably  be  considered  noise  in  the  performance  run.      

Overall,  total  median  performance  also  improved  incrementally  when  all  of  the  storage  efficiency  features  were  enabled.      Also  evident  in  Figure  8  is  the  increased  variability  of  the  all  features  run,  i.e.,  the  peak  minus  the  minimum  performance  for  each  workload  increased.    However,  most  of  this  range  difference  was  attributable  to  the  higher  performance  of  each  workload.      

Page 15: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  14     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

 

Summary  

 In  conclusion,  the  storage  capacity  savings  gained  from  NetApp’s  thin  provisioning,  data  deduplication  and  data  compression  were  truly  remarkable.    As  shown  in  Figure  9,  thin  provisioning  alone  provided  a  sizable  46  percent  capacity  savings.    

But  enabling  data  deduplication  provided  even  more  overall,  a  68  percent  savings  as  compared  to  baseline  capacity  used.    Data  compression  added  still  more,  such  that  when  all  tested  features  were  activated,  the  size  of  the  original  storage  was  reduced  by  an  impressive  79  percent.      

…  when  all  tested  features  were  activated,  the  size  of  the  original  storage  was  reduced  by  an    

impressive  79  percent    

Figure  9  Overall  capacity  requirements  

Page 16: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  15     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

 In  comparison,  NetApp  Snapshot  and  FlexClone  copies  did  not  have  any  impact  on  capacity  requirements  for  source  data.    As  both  are  only  point-­‐in-­‐time  copies,  their  post-­‐workload  capacity  was  compared  simply  with  the  baseline  capacity  in  the  above  chart.  Thus,  as  seen  in  Figure  10,  both  facilities  provided  impressive  point-­‐in-­‐time  copies  greater  than  95  and  78  percent  smaller  for  Snapshot  and  FlexClone  copies  respectively  than  baseline  data.        

Figure  10  Capacity  savings  for  data  copy  facilities  chart  

Page 17: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  16     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

 Figure  11  Baseline  vs.  all  tested  features  performance  comparison  chart  

 Besides  the  tremendous  capacity  savings  achieved  using  thin  provisioning,  deduplication  and  compression,  enabling  these  storage  efficiencies  had  no  negative  impact  on  the  overall  performance  of  the  NetApp  storage  system.    Moreover,  when  comparing  overall  median  performance,  NetApp’s  operational  throughput  also  exhibited  no  negative  impact.        The  ultimate  decision  to  use  any  or  all  of  vendor’s  storage  capacity  saving  features  or  their  point-­‐in-­‐time  copy  capabilities  can  be  a  complex  decision  and  often  involves  a  tradeoff  with  performance.    However,  NetApp  thin  provisioning,  data  deduplication  and  compression  can  potentially  provide  overwhelming  storage  capacity  savings  with  little,  if  any,  overall  performance  degradation  and  thus,  deserve  strong  consideration  for  any  data  center  environment.         Silverton Consulting, Inc. is a Storage, Strategy & Systems consulting services company, based in the USA offering products and services to the data storage community.  

Page 18: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  17     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

Appendix  1  SCI  Lab  resources  and  workload  details  SCI’s  lab  uses  enterprise-­‐class  server  and  networking  resources  to  support  hardware  and  software  validation  activities  including:    

• One  Westmere  class,  dual  processor,  six-­‐core  server  with  144GB  of  memory  and  an  SSD  for  internal  storage  

• One  Nehalem  class,  dual  processor,  quad-­‐core  server,  with  48GB  of  memory  and  an  SSD  for  internal  storage  

• One  SandyBridge  class,  single  processor,  quad-­‐core  server,  with  32GB  DRAM  and  an  SSD  for  internal  storage  

• Six  Xeon  class,  dual  processor,  quad  core  servers  with  five  having  48GB  of  DRAM,  using  internal  SAS  drives  for  local  storage  

• Three  FC  SAN  switched  fabrics  supporting  2GFC,  4GFC,  and  8GFC,  and  • Two  Ethernet  fabrics  supporting  both  1GigE  as  well  as  10GbE,  providing  

FCoE,  iSCSI  and  normal  LAN  traffic.    Although  all  the  above  were  available  for  testing,  the  Nehalem  class  server  running  VMware  with  3  virtual  machines  (VMs)  each  having  16GB  of  DRAM  was  utilized  for  this  test.    All  the  data  was  accessed  over  10Gb/sec  Ethernet  (10GbE)  interfaces.    The  server  had  two  standard  Intel  10GbE  XF  SR  NICs  teamed  together  used  for  iSCSI  and  a  single  Emulex  11101  NIC  used  for  CIFS  traffic.    No  attempts  were  made  to  optimize  system  or  storage  performance  but  rather  to  establish  a  baseline  level  of  performance  for  comparison  purposes.        

Workloads  used  to  measure  performance    To  measure  system  performance,  a  typical  workload  was  generated  against  the  previously  acquired  data  using  the  SCI  lab  server.    One  VM  was  dedicated  to  each  workload  as  follows:      

• File  system  workload:  A  CIFS  file  share  was  created  and  accessed  by  one  VM.    Then,  a  simulated  file  workload  was  constructed  which  wrote  and  read  data  concurrently  using  an  automated  copy  script.  

• SQL  DB  workload:  A  SQL  Server  was  configured  and  a  simulated  workload  was  created  consisting  of  changing  and  modifying  column  values  in  the  relational  tables.  

• Email  workload:  Microsoft’s  Exchange  2010  Jetstress  tool  was  run  for  1150  mailboxes  producing  ~0.18  I/O  per  mailbox  per  second,  i.e.  a  normal  email  workload.  

Data  used  in  test  Test  data  was  taken  from  a  number  of  sources  including  publicly  available  email  data,  internal  file  data  from  SCI’s  lab  and  office  environment  and  text/image/PDF  data  obtained  from  the  web.    Specifically,    

Page 19: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  18     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

• File  system:  ~211GB  consisting  of  48%  email  data  (.pst  files/email  data),  21%  Perfmon  data,  15%  text,  7%  image  data,  5%  Office/PDF  data,  and  4%  DB/SQL  data.  

• SQL  database  (DB)  data:  ~474GB  of  data  spread  across  18  tables  containing  text  and  web  server  log  data.  

• Email  data:  ~414GB  of  email  with  88MB  of  log  data  created  by  the                                                                                    Microsoft  Jetstress  tool.  

 The  testing  used  a  variety  of  data  types  to  simulate  the  diversity  of  data  found  in  many  customer  environments  and  to  reduce  the  potential  for  non-­‐standard  results  based  on  “artificial”  data.  However,  the  testing  is  not  intended  to  represent  best  practice  guidelines  for  any  specific  application  or  environment.    Readers  are  encouraged  to  consult  NetApp  documentation  and  personnel  directly  for  the  best  practice  recommendations  for  their  specific  application  requirements.      Additionally,  the  performance  testing  was  designed  to  measure  before  and  after  results  to  assess  any  potential  impact  of  implementing  multiple  storage  efficiency  and  copy  technologies.  These  results  are  not  intended  to  be  used  for  performance  sizing  and  do  not  reflect  possible  throughput  results  outside  of  the  specific  test  environment.  Readers  are  encouraged  to  consult  NetApp  documentation  and  personnel  directly  for  performance  recommendations  for  their  specific  requirements.      

Page 20: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  19     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

 

Appendix  2  NetApp  CLI  commands  used  and  results  summary   Feature Commands to enable Commands to measure

savings Savings

Baseline df –k Thin provisioning

lun set reservation path disable; vol options volname guarantee=none

df –k; df -A

Moderate

Data deduplication

sis on path; sis start –S path;

df –k; df –S

Moderate

Data compression

sis config –C TRUE path; sis start –S -C path;

df –k; df –S

Moderate

Snapshot snap create volname snapvolname

df –k; snap list

Outstanding3

FlexClone Vol clone create clonename –s volume volname

df –k; vol clone split estimate clonename; snap list;

Significant4

All features (As indicated above) (As indicated above) Substantial Table  1  Command  and  results  summary  table  

                                                                                                               3  For  original  source  data  there  were  no  savings  but  for  snapshot  copies  there  were  outstanding  savings  4  For  original  source  data  there  were  no  savings  but  for  FlexClones  there  were  significant  savings  

Page 21: SCI Lab Test Validation Report: NetApp Storage Efficiency

  SCI  Lab  Validation  Report:  NetApp  Storage  Efficiency  

  ©  2012  Silverton  Consulting,  Inc.   Page  20     All  Rights  Reserved  twitter.com/RayLucchesi|RayOnStorage.com

+1-720-221-7270|SilvertonConsulting.com

Appendix  3  Summary  of  Capacity  Test  Results  Cumulative  Storage  Efficiency  Test  Results  (GB)         Test  Step  1   Test  Step  2   Test  Step  3   Test  Step  4  

   

Post-­‐workload  Baseline    

 Net  After  Thin  Provisioning  

 Net  After  Thin  Provisioning  &  Deduplication  

 Net  After  Thin  Provisioning,    

Deduplication  &  Compression  

File  System  Storage   629.15   211.72   118.27   106.12  SQL  DB  Storage   629.15   473.60   452.42   229.20  SQL  Log  Storage   104.86   1.20   0.02   0.02  Email  DB  Storage   629.15   414.03   87.04   87.03  Email  Log  Storage   41.94   0.09   0.00   0.00  Total  Capacity   2,034.24   1,100.64   657.76   422.37  Table  2  Cumulative  storage  efficiency  test  results    Copy  Services  Test  Results  (GB)           Test  Step  1   Test  Step  5   Test  Step  6  

   Post-­‐workload  

Baseline      Net  After  

Snapshot  Copy    Net  After  

FlexClone  Copy  File  System  Storage   629.15   6.81   66.63  SQL  DB  Storage   629.15   90.07   200.87  SQL  Log  Storage   104.86   0.01   65.15  Email  DB  Storage   629.15   7.48   115.80  Email  Log  Storage   41.94   0.00   0.35  Total  Capacity   2,034.24   104.37   448.80  Table  3  Copy  services  test  results