How to use Software-‐Defined Networking to Improve ...

24 downloads 15629 Views 312KB Size Report
Keywords: Software Defined Network, SDN, Network Security, IP spoofing ... gathering and security reaction mechanisms, especially when compared with the narrow local view .... some SDN benefits added on top (Levin 2013), with SDN policies being ..... Kwon, J., Seo, D., Kwon, M., Lee, H., Perrig, A., & Kim, H. (2015).
How  to  use  Software-­‐Defined  Networking  to  Improve  Security  –  a  Survey   Jorge  Proença,  Tiago  Cruz,  Edmundo  Monteiro,  Paulo  Simões   Universidade  de  Coimbra,  Portugal   [email protected]   [email protected]   [email protected]   [email protected]     Abstract:   Despite   being   a   relatively   recent   development,   the   SDN   paradigm   has   already   challenged   the   established   network   design,   management   and   operation   concepts.   SDN   is   the   result   of   a   number   of   studies   and   ideas   on   network   programming,   oriented   towards   the   improvement   of   the   traditional   network   functionality   and   management,   due   to   its   unique   levels   of   flexibility.   One   of   the   main   characteristics   of   this   paradigm  is  the  breakout  of  the  functionalities  of  traditional  network  elements  (e.g.,  routers),  moving  part  of   them  to  a  centralized  controller.  Network  elements  such  as  routers  become  simpler  forwarding  devices,  and   the   controller   becomes   responsible   for   routing   decisions.   This   allows   the   controller   to   have   complete   perspective  and  control  over  the  network.  Moreover,  network  changes  become  seamless,  as  the  controller  can   change  the  network  behaviour  online.       In   this   paper   we   present   an   overview   and   survey   on   the   usage   of   the   Software   Defined   Networking   (SDN)   concept   for   security   purposes,   discussing   how   to   use   to   increase   network   security   awareness,   as   well   as   for   reacting   to   ongoing   attacks.   This   survey   analyses   research   studies   that   are   representative   of   a   variety   of   security  measures  that  can  be  improved  by  the  use  of  SDN  enabled  technologies.  We  address  network  security   in   four   distinct   fields:   SDN-­‐enabled   honeypots,   Denial-­‐of-­‐Service   (DoS)   mitigation,   source   address   spoofing   countermeasures,  and  network  scanning  avoidance.     Keywords:  Software  Defined  Network,  SDN,  Network  Security,  IP  spoofing  mitigation,  DoS  Spoofing  Mitigation,   Network  Scan  Avoidance,  Honeypot         1. Introduction   Software   defined   networking   (SDN)   is   an   architectural   approach   to   network   configuration   and   management   which   allows   for   unprecedented   levels   of   flexibility,   also   enabling   the   implementation   of   evolved   network   virtualization   scenarios.   The   SDN   paradigm   aggregates   a   number   of   studies   and   ideas   about   network   programmability   (Feamster   2013).   In   SDN   routers   and   switches   (forwarding   elements)   the   control   plane   is   decoupled  from  the  forwarding  plane,  with  the  former  moved  to  a  central  controller.  Packet  forwarding  is  flow   oriented,  meaning  that  both  origin  and  destination  are  taken  into  account,  instead  of  just  packet  destination   (as  in  traditional  networking).  Flow  policies  are  granted  by  a  centralized  controller,  which  manages  the  policies   of   a   range   of   forwarding   elements   in   a   given   network.   Using   this   controller,   control   plane   functions   can   be   moved  out  of  the  forwarding  element.  For  this  reason,  the  controller  will  have  a  broader  view  of  the  domain,   contrasting  with  the  narrow  (localized)  view  that  an  individual  forwarding  element  has  in  traditional  networks.   As  a  result,  this  paradigm  enables  the  creation  of  more  flexible  management  mechanisms,  especially  suited  for   complex   scenarios.   The   interface   between   the   central   controller   and   the   forwarding   elements   (routers,   switches)  is  usually  based  on  the  OpenFlow  protocol  (Greene  2009).     While   SDN   is   frequently   leveraged   to   address   the   operation   and   management   needs   of   applications   such   as   on-­‐the-­‐fly  network  virtualization,  it  can  also  provide  an  effective  mechanism  for  security  applications.  In  fact,   SDN  not  only  improves  network  deployment  and  management,  but  also  simplifies  network  security  tasks  in  an   efficient  manner.  Part  of  this  is  due  to  the  fact  that  a  centralized  element  with  a  global  view  of  all  the  network   entities   –   such   as   devices,   flows   and   network   elements   –   is   able   to   provide   more   efficient   information   gathering  and  security  reaction  mechanisms,  especially  when  compared  with  the  narrow  local  view  individually   provided   by   each   forwarding   element   in   traditional   IP   networks,   where   constant   data   pooling   is   needed   to  

monitor  the  whole  network  state.  This  broad  view  of  the  network  and  ability  to  easily  receive  information  from   all   forwarding   elements   can   be   used   to   provide   better   security   methods   without   adding   new   and   specific   protocols   (which   can   be   a   deterrent   for   their   adoption)   such   as   in   VAVE   source   address   spoofing   countermeasure  (Yao  2011).  Moreover,  flow-­‐based  forwarding  rules  used  in  SDN  networks  can  help  detecting   suspicious   connections   and   increase   the   efficiency   of   a   reaction.   With   its   flexible   nature,   the   controller   can   seamlessly  isolate  or  divert  flows.  Instead  of  blocking  an  attack,  SDN  allows  an  easier  redirection  of  traffic.  This   is   useful   to   extend   existing   techniques,   improving   their   effectiveness,   such   as   adding   a   layer   of   anomaly   detection   to   remote   triggered   black   holes   (RTBH).   This   flexibility   can   also   be   used   to   improve   a   honeypot’s   efficiency   which   may   become   active   –   meaning   the   whole   system   has   a   higher   probability   of   detecting   the   attack  and  then  to  easily  redirect  it  to  the  honeypot  (rather  than  having  a  static  honeypot  passively  listening   on  the  empty  scope  of  the  network).  SDN  can  also  help  handling  Denial  of  Service  (DoS)  and  Distributed  DoS   (DDoS)  attacks,  by  providing  more  effective  detection  and  reaction  mechanisms.       In  this  paper  we  discuss  a  representative  sample  of  proposals  of  using  SDN  to  improve  a  number  of  security   fields   ranging   from   avoiding   attacks   to   their   detection,   namely:   SDN-­‐enabled   honeypots,   Denial-­‐of-­‐Service   (DoS)   mitigation,   source   address   spoofing   countermeasures,   and   network   scanning   avoidance.   We   also   compare   these   samples   with   traditional   techniques,   which   do   not   use   an   SDN   approach.     This   is   not   an   exhaustive   analysis,   but   rather   a   survey   of   representative   use   cases   related   to   the   use   of   SDN   technologies   for   security  purposes.     The  remaining  sections  are  ordered  as  follows.  Section  2  explains  the  conceptual  vision  of  a  network  into  three   separate   planes.   Section   3   gives   an   overview   of   SDN   networking   and   compares   it   to   traditional   networking   approaches.   Section   4   discusses   security   improvements   that   SDN   can   bring   to   networking   and   Section   5   concludes  the  paper.     2. Software  Defined  Networking  Overview   This   section   intends   to   provide   a   technical   background   on   SDN   technologies   and   its   related   concepts,   with   the   main  purpose  of  supporting  the  survey  study  that  will  be  introduced  in  the  next  chapter.  Hereby  we  describe   the   main   characteristics   of   SDN,   including   its   conceptual   division   of   networking   planes   and   how   it   compares   with  traditional  networking.     2.1 The  three  planes  of  networking   From   a   functional   standpoint,   network   architectures   can   be   split   across   three   fundamental   planes:   Management,  Control  and  Data  (Kreutz  2014,  Ellanti  2005).  Each  plane  has  a  well-­‐defined  purpose  and  specific   set  of  characteristics  (Figure  1):   § The  management  plane  relates  to  traffic  used  by  software  services,  for  provision,  maintenance,  and   monitoring.   Traffic   related   to   this   plane   can   be   either   transported   through   in-­‐band   (using   the   same   connections   as   user   traffic)   or   out-­‐of-­‐band   (OOB)   connections   (a   separate   connection,   exclusively   dedicated  to  management  operations)  (Schudel  2007).     § The  main  role  of  the  control  plane  (also  known  as  signalling  plane)  relates  with  the  setup  of  the  data   plane,   also   encompassing   the   traffic   between   network   elements.   For   example,   routers   use   routing   protocols  to  exchange  network  information,  so  that  each  router  knows  where  to  forward  a  packet  for   it  to  reach  its  destination.  Control  plane  traffic  includes  signalling,  routing  information,  and  link-­‐state   protocols,  among  other  types  of  traffic  (Schudel  2007).   § The  data  plane  (also  known  as  user  plane,  forwarding  plane,  carrier  plane  or  bearer  plane)  is  in  charge   of   transporting   user   data.   This   means   that   the   traffic   belonging   to   the   data   plane   never   has   source   or   destination  IP  addresses  belonging  to  the  network  elements  (e.g.,  switches  and  routers).  It  will  have   addresses   belonging   to   end   devices,   such   as   computers   and   servers.   Traffic   in   this   data   plane   uses  

network  as  transport  only.  The  main  objective  of  the  other  conceptual  network  planes  is  to  support   the  operation  of  the  data  plane  (Schudel  2007).    

                                                       Management  Plane

                                                                 Control  Plane

                                                                       Data  Plane

  Figure  1:  Network  Planes.  Adapted  from  (Kreutz  2014)     This  functional  division  is  traditionally  handled  in  such  a  way  that  control  and  data  planes  are  frequently  kept   in   close   coupling,   within   the   scope   of   network   equipment.   For   instance,   in   a   conventional   Ethernet   switch,   the   forwarding  plane  is  integrated  with  the  control  plane  –  the  MAC  address  table  is  maintained  by  the  embedded   electronics   that   constitute   the   packet   forwarding   engine.   This   corresponds   to   the   conventional   networking   paradigm,  which  is  challenged  by  SDN.     2.2 Software  Defined  Networking  Overview   The   term   SDN   (for   “Software   Defined   Networking")   was   first   introduced   in   (Greene   2009),   referring   to   the   OpenFlow  protocol  being  developed  at  the  University  of  Stanford  (ONF  2014)  ,  which  eventually  became  one   of  the  first  SDN-­‐enabling  standards.     SDN  breaks  with  the  existing  vertical  integration  that  is  characteristic  of  the  traditional  networking  model,  by   decoupling   of   the   control   and   data   planes   (Kreutz   2014).   In   SDN,   the   control   plane   is   moved   away   from   the   forwarding   elements   (i.e.,   routers),   and   placed   in   a   logically   centralized   controller.   This   does   not   imply   a   physical   concentration   of   functionality,   as   the   controller   functions   can   be   performed   by   several   controller   instances   to   improve   scalability   and   resilience   (Yeganeh   2013).   The   data   plane   remains   in   the   forwarding   elements,  whose  task  becomes  focused  only  on  traffic  forwarding.  The  forwarding  elements  can  have  multiple   purposes   and   can   perform   the   role   of   several   devices   (e.g.   router,   switch,   and   firewall)   depending   on   the   configuration   provided   by   the   controller.   These   changes   allow   the   controller   to   have   a   broader   view   of   the   network,   comparing   with   the   narrow   view   each   individual   router   has   in   traditional   networks   –   where   each   router  has  limited  scope  of  visibility,  provided  by  their  closest  neighbouring  routers  and  other  elements  (e.g.   static  routes).       The   broader   view   of   the   SDN   controller   allows   networks   to   become   more   flexible   and   more   manageable,   with   network   changes   becoming   a   simpler   task.   Additionally,   packet   forwarding   is   flow   oriented,   meaning   both   origin  and  destination  are  taken  into  account,  instead  of  just  packet  destination  as  in  traditional  networking.   Figure  2  illustrates  the  flow  rule  table  of  the  OpenFlow  protocol.    

Flow   policies   are   granted   by   a   centralized   controller,   which   manages   the   policies   of   a   range   of   forwarding   elements  in  a  given  network.  Using  this  controller,  control  plane  functions  can  be  moved  out  of  the  forwarding   element.  For  this  reason,  the  controller  will  have  a  broader  view  of  the  domain,  contrasting  with  the  narrow   view  that  an  individual  forwarding  element  has  in  a  traditional  IP  network.      

Rule

Action

Statistics Packet+  byte  counters  

-­‐Forward  packets  to  ports -­‐Encapsulate  and  forward  to  controller -­‐Drop  the  packet -­‐Send  to  normal  processing  pipeline

Switch Port

MAC Src

MAC Dst

Eth Type

VLAN ID

IP Src

IP Dst

IP Prot

IP IP Sport Dport

  Figure  2:  OpenFlow  flow-­‐rule  table.  Adapted  from  (SDX  Central  2014).     The   SDN   paradigm   allows   for   easier   and   more   flexible   network   management,   especially   on   complex   scenarios.   Moreover,  migrating  to  SDN  does  not  require  a  sudden  change:  it  can  be  done  incrementally,  using  switches   that   are   not   SDN   exclusive.   This   way,   some   elements   of   the   network   can   have   a   traditional   behaviour   but   with   some  SDN  benefits  added  on  top  (Levin  2013),  with  SDN  policies  being  deployed  gradually.  Additionally,  this   allows  the  hardware  to  be  gradually  replaced  taking  into  account  its  normal  life  expectancy.  As  a  result,  costs   can  be  lowered  in  the  deployment  and  transition  between  these  scenarios  (Jarraya  2014).     This  paradigm  not  only  allows  for  more  manageable  networks,  but  can  also  improve  network  security.  In  the   next  section  we  will  describe  some  of  the  possible  uses  for  enhancing  network  security  with  the  use  of  SDN.     3 Software  Defined  Network  Uses  in  Security   The   benefits   of   SDN   are   not   limited   to   network   management   and   improved   flexibility,   as   they   can   also   be   leveraged   in   the   scope   of   network   security   mechanisms.   This   is   mainly   due   to   the   fact   that   SDN   enables   network   operators   to   manipulate   traffic   flows   seamlessly   without   service   interruptions   (as   previously   discussed  in  Chapter  2),  a  capability  that  can  be  explored  to  implement  and/or  improve  security  measures.       In  this  section,  we  detail  some  network  security  aspects  that  can  be  enhanced  or  mitigated  using  SDN.  More   specifically,  we  address  security  proposals  in  four  different  topics:  honeypots,  DoS  mitigation,  source  address   spoofing,  and  network  scans.     3.1 SDN-­‐Enabled  Honeypots   One   of   the   potential   applications   of   SDN   relates   with   honeypots.   By   definition,   a   Honeypot   is   a   decoy   or   dummy   target   set   up   to   attract   and   detect/observe   attacks.   By   being   exposed   to   probing   and   attack,   its   purpose   is   to   lure   and   track   intruders   as   they   advance   (Simões   2013),   revealing   any   scouting   activities.   Traditionally,  honeypot  systems  live  in  unused  address  space  in  the  system  (i.e.  IP  addresses  not  used  by  the   production  system),  passively  waiting  for  attackers  to  “find  them”  (Spitzner  2003),  but  their  operation  can  be   greatly  improved  by  SDN,  which  ads  the  possibility  of  turning  them  into  a  more  proactive  defence  layer.     Using  the  SDN's  capability  of  changing  network  flows,  it  becomes  possible  to  change  the  honeypot's  behaviour   from  passive  to  active.  One  way  to  achieve  this  is  for  the  active  honeypot  to  work  together  with  other  security   mechanisms  such  as  network  intrusion  detection  systems  (NIDS).  These  are  systems  placed  in  strategic  points   in  the  network  to  monitor  traffic  from  the  devices  in  the  network.  They  can  detect  suspicious  activity  such  as   probe   scans,   DoS   attacks   or   MITM   attacks.   Snort,   for   instance,   is   one   of   the   most   popular   NIDS   (Roesch   1999).  

When  an  unauthorized  activity  is  detected  by  the  NIDS,  the  SDN  controller  can  divert  the  suspicious  traffic  to   the  honeypot,  even  when  it  was  originally  intended  for  production  system.  The  attacker  would  be  not  aware  of   this  diversion  and  would  continue  the  attack,  while  the  honeypot  would  log  its  activity  for  forensics  analysis.   Figure  3  illustrates  an  example  of  an  active  honeypot  architecture.     Compared  to  traditional  honeypots,  SDN-­‐enabled  honeypots  provide  a  key  advantage:  while  the  effectiveness   of  traditional  honeypots  depends  on  the  attacker  directly  contacting  the  Honeypot  on  the  addresses  it  is  using   (typically   by   randomly   or   systematically   probing   the   network   under   attack   in   order   to   discover   its   resources   and   layout),   SDN-­‐enabled   honeypots   can   proactively   “capture”   any   suspicious   traffic,   even   when   this   traffic   directly  targets  the  production  systems  (either  by  chance  or  because  the  attacker  had  some  prior  knowledge   of  the  system  and  knows  the  IP  addresses  of  the  production  systems,  therefore  circumventing  the  honeypot   trap).       Data  Path Failed    Connection Normal  Traffic

GuardiCore Defense  Suite Control  Logic

Active   Honeypot

HP  VAN  SDN Controller

VM              VM              VM

VM              VM              VM

Hypervisor

Hypervisor

  Figure  3:  Active  honeypot  (Hewlett-­‐Packard  2014).     3.2 DoS  Mitigation   Denial   of   service   (DoS)   and   Distributed   DoS   (DDoS)   attacks   have   the   goal   of   disabling   legitimate   communications   and   activities   such   as   web   page   browsing   or   online   streaming,   disabling   the   source   of   the   service  (Mirkovic  2004).  This  can  be  done  in  a  number  of  ways  such  as  by  exploiting  vulnerabilities  or  sending   an   excessive   amount   of   connections   in   order   to   consume   all   the   resources   such   as   CPU   time   or   memory.   Moreover,  the  number  of  DoS  attacks  has  been  increasing,  as  well  as  their  complexity  (Yu  2014,  Apps  2014).   These   attacks   are   unusually   difficult   to   detect   because   of   the   similarities   between   real   traffic   generated   by   the   users  and  false  traffic  generated  to  disable  the  targeted  service.       One   method   of   reducing   the   impact   of   a   DoS   attack   in   traditional   non-­‐SDN   networks   is   by   using   remote   triggered   black-­‐holes   (RTBH)   to   reduce   the   impact   of   a   DoS   attack   (Kumari   2009).   This   technique   consists   of   creating   routing   entries   to   drop   undesirable   traffic   before   it   reaches   an   unprotected   network.   Usually,  this   has   a  secondary  effect  –  the  attacked  host  becomes  unreachable,  as  traffic  destined  to  it  is  dropped.  Instead,  by   using  SDN  it  becomes  possible  to  prevent  collateral  damage  on  the  network  and  hosts  surrounding  the  victim.       There   are   several   proposals   to   mitigate   these   attacks   using   SDNs   technologies.   For   instance,   (Giotis   2014)   proposed   an   OpenFlow-­‐based   architecture   which   is   capable   of   mitigating   a   DoS   attack,   while   keeping   the   victim   available   and   preventing   collateral   damage   to   the   remaining   network.   This   proposal   uses   SDN   to   enhance  the  abovementioned  RTBH  method.  Instead  of  the  RTBH  classical  approach,  in  which  packets  are  sent   to  a  sinkhole  interface  of  the  edge  router,  network  traffic  flagged  as  unauthorized  or  malicious  is  sent  to  an   OpenFlow  switch.  Using  the  Layer  2-­‐4  matching  capabilities  of  the  OpenFlow  protocol,  malicious  and  benign  

traffic   are   separated.   The   former   is   dropped   and   the   latter   is   sent   back   to   the   edge   router,   where   it   will   be   forwarded  to  its  intended  destination.       Another  SDN  enabled  DoS  mitigation  proposal  uses  an  unsupervised  neural  network  (self-­‐organizing  maps)  to   achieve  detection  with  low  processing  requirements  (Braga  2010).  The  detection  of  DoS  attacks  is  done  using   NOX  (Gude  2008)  and  OpenFlow.  This  proposal  takes  advantage  of  the  controller's  overall  view  of  the  network   to   collect   flow   information   from   the   switches   to   feed   the   classifier.   The   detection   is   performed   on-­‐line   and   uses  a  combination  of  six  metrics:     § Packets  per  flow:  as   DDoS   relies   on   a   transmission   of   a   small   number   of   packets   from   a   large  number   of  sources,  flows  with  a  small  number  of  packets  are  possible  DDoS  sources.   § Average   bytes   per   flow:   small   payload   size   is   common   in   DDoS   packets   to   increase   the   attack   efficiency.   § Average   of   duration   per   flow:   an   SDN   flow   is   deleted   from   its   flow   table   if   left   inactive   (no   packets   received)  for  a  period  of  time.  If  a  flow  remains  in  the  flow  table  for  a  short  period  of  time,  it  can  be  a   DDoS  source.   § Percentage   of   pair-­‐flows:   the   asymmetry   between   flows   coming   into   an   out   of   the   network   can   be   an   indicator   of   DDoS   attack,   as   sometimes   it   increases   the   number   of   incoming   connections   without   a   corresponding  outgoing  one  (Kreibich  2005).   § Growth   of   single-­‐flows:   it   is   possible   that   the   number   of   flows   increases   dramatically   in   the   beginning   of  a  flood  attack.  This  growth  is  calculated  with  the  number  of  pair-­‐flows  left  out  during  a  certain  time   interval.     § Growth  of  different  ports:  Used  to  calculate  the  number  of  ports  used  in  the  attack  (port  spoofing),   since  it  can  also  have  a  dramatic  increase  similar  to  the  number  of  flows.     Using  the  well-­‐known  KDD-­‐99  dataset,  the  authors  report  good  detection  performance  with  a  low  percentage   of  false  positives.  Moreover,  they  require  substantially  less  CPU  time  to  perform  the  detection  comparing  with   other  kinds  of  detection.     3.3 Source  Address  Spoofing  Countermeasures   Source   address   spoofing   is   a   technique   that   can   enable   a   number   of   attacks,   such   as   SYN   Flood,   man   in   the   middle  (MITM),  Smurf,  and  DNS  amplification,  among  others  (Bi  2010).  One  of  the  most  common  and  accepted   practices   to   mitigate   address   spoofing   is   ingress   filtering   (Liu   2011).   However,   its   deployment   is   slower   than   the   internet   growth   and   will   not   be   effective   unless   fully   deployed   (Kwon   2015).   Source   Address   Validation   Improvement   (SAVI)   is   an   IETF   effort   to   improve   ingress   filtering   (McPherson   2013).   SAVI   does   address   spoofing   detection   by   binding   the   switch   port   where   a   host   is   connected   to   its   IP   address   prefix   (Bagnulo   2013).   However,   it   has   some   limitations.   A   host   attached   to   a   SAVI   device   can   be   attacked   by   spoofing   the   source  address  with  one  bound  to  another  SAVI  device  (Yao  2001).       Virtual   source   Address   Validation   Edge   (VAVE)   is   an   enhanced   proposal   for   address   spoofing   detection.   It   is   designed  to  protect  the  important  addresses  from  the  network  from  being  spoofed  (e.g.,  servers  and  routers)   and  it  can  prevent  flows  from  between  two  SAVI  switches  from  being  spoofed  (Yao  2011).  It  uses  OpenFlow   switches   and   NOX   to  improve  source  address  verification.  VAVE  conceptually  separates  the  flows  coming  from   hosts,   other   OpenFlow   switches   and   legacy   (non   OpenFlow)   routers.   When   a   new   flow   is   detected,   the   application   checks   it   for   spoofing   addresses.   If   source   address   spoofing   is   detected,   it   changes   the   flow   rule   table  to  drop  packet.  If  a  spoofing  flow  is  inactive  (no  new  packets)  for  some  time,  it  is  deleted.  Seeing  that   packets   are   checked   when   they   entered   an   OpenFlow   switch   from   a   legacy   device,   packets   transmitted   between  neighbouring  OpenFlow  switches  are  not  checked  for  spoofed  addresses.  

NOX Filter Generator

Validation Module

Filter Adapter

VAVE  App Hit  Filtering  Rule

LD

VAVE

Flow  Path (A,  F)

OP  A

OP  C

LD

Spoofing Flow (S=A,  D=F)

OP  D

OP  F

LD

OP

OP  B

Spoofing Flow (S=A,  D=F) OpenFlow   Device

LD

OP  E

OP  G

LD

LD

LD

Legacy Device

LD

  Figure  4:  VAVE  architecture  (Yao  2011).     3.4 Network  Scanning  Avoidance   Network   security   improvements   can   be   done   proactively:   it   is   better   to   avoid   an   attack   than   to   detecting   it.   The  initial  stage  of  several  attacks  consists  of  network  scans  with  the  objective  of  gathering  information.       To   reduce   the   probability   of   a   successful   scan,   a   random   host   mutation   is   proposed   in   (Jafarian   2012)   using   OpenFlow  Remote  Host  Mutation  (OF-­‐RHM).  The  reason  behind  the  proposal  is  that  static  configurations  are   beneficial  for  attackers.  If  their  targets  are  fixed  it  is  easier  to  find  and  attack  than  if  the  hosts'  IP  addresses  are   constantly   changing.   The   two   main   objectives   of   OF-­‐RHM   are:   transparent   IP   address   changing   to   the   end-­‐ hosts;   and   highly   unpredictable   IP   changes   and   frequent   enough   to   distort   the   attacker's   view   of   the   network.   IP  address  changing  is  achieved  by  using  the  entire  unused  space  to  assign  virtual  IP  addresses.  Each  host  has  a   real   IP   address   and   is   assigned   a   virtual   (short-­‐lived)   IP   addressed   that   will   be   translated   to   the   real   IP   right   before  the  host.  There  are  other  proposals  of  host  mutation  (Atighetchi  2003,  Kewley  2001,  Antonatos  2005),   but   have   some   implementation   limitations   such   as   requiring   cooperation   from   end-­‐hosts,   and   disrupting   existing  connections.       A  host  can  be  reached  by  hostname  or  by  its  real  IP  address.  When  reached  by  hostname,  the  first  stage  is  to   query  the  DNS  server  to  get  the  IP  address  associated  with  the  hostname.  As  soon  as  a  DNS  query  is  made,  the   NOX   controller   changes   fields   in   the   DNS   response.   It   replaces   the   real   IP   for   the   virtual   IP   and   sets   a   small   value  to  the  time  to  live  (TTL)  field.  The  host  will  then  send  the  first  packet  and  because  the  OpenFlow  switch   does  not  have  a  matching  rule  to  that  connection,  the  controller  updates  its  rule  table.     The   virtual   IP   is   assigned   from   the   pool   of   unused   IP   addresses   and   can   be   chosen   by   two   ways:   1)   blind   mutation  where  the  new  IPs  are  chosen  randomly  with  a  uniform  probability;  2)  weighted  mutation  where  the   weight  is  associated  to  each  virtual  IP  address.  This  weight  is  based  on  how  many  times   an  unused  IP  has  been   scanned.  The  more  times  an  IP  has  been  scanned,  the  less  likely  it  is  to  be  scanned  again  by  a  malicious  host.    

According  to  the  authors,  OF-­‐RHM  can  reduce  the  efficiency  of  network  scan  attacks  up  to  99%.  This  figure  was   achieved   by   running   100   scans   on   the   network   and   comparing   their   result   with   the   ground   truth   (the   first   scan).  Moreover,  up  to  90%  of  the  network  hosts  are  saved  from  vicious  scanning  worms.  This  was  tested  with   worms   with   low   IP   repetition,   which   use   number   generators   to   reduce   the   possibility   of   hitting   the   same   IP   several   times.   Although   named   hosts   can   still   be   reached   via   DNS,   most   scanners   use   IP   address   to   collect   information  in  order  to  avoid  too  many  DNS  queries  and  detection.     Another  method  to  avoid  network  scans  attacks  is  proposed  in  (Song  2014).  The  authors  propose  a  mitigation   strategy  by  providing  false  information  with  the  use   of  a   honeynet.  Their   system   uses   OpenFlow   to   detect   the   scan   attacks   by   looking   at   incoming   packets   towards   a   closed   or   unused   ports,   or   if   the   packets   are   corrupted.   A   corrupted   packet   is   one   that   does   not   follow   the   network   protocol   standard.   For   example,   if   a   TCP   session   is   initiated  with  a  RST  packet,  that  packet  is  considered  corrupted.       After  a  successful  detection,  the  packet  and  successive  ones  in  that  flow  are  redirected  to  the  honeynet.  There   are  two  different  systems:  the  All-­‐Alive  Network  and  the  Phantom  Network.  In  the  former,  the  honeypots  have   all  ports  open.  The  honeypot  simulates  popular  network  services  (e.g.  using  port  80,  445,  and  8080)  and  for   the  remaining  ports,  the  authors  use  a  program  open  them,  but  network  services  are  not  emulated,  it  will  only   reply   with   simple   pre-­‐defined   data.   For   the   Phantom   Network,   some   network   services   are   chosen   for   emulation.  This  network  will  not  be  the  same  as  the  real  one  that  is  being  protected.     4 Discussion  and  Conclusion   The  emerging  concept  of  Software  Defined  Networking  is  expected  to  strongly  change  the  way  we  design  and   manage  communications  networks.  While  not  specifically  targeting  network  security,  adoption  of  SDN  opens   the   door   for   novel   approaches   to   network   and   systems   security.   In   this   paper   we   discuss   some   of   the   more   promising   security   management   approaches   enabled   by   SDN,   including   SDN-­‐enabled   honeypot   scenarios,   defence   mechanisms   for   denial   of   service   attacks,   for   spoofing   source   addresses   and   for   network   scans.   Although  SDN  is  a  key  element  in  these  proposals,  the  authors  use  SDN  in  a  number  of  different  ways.  Not  only   is   it   used   to   implement   novel   security   techniques   such   as   seamless   random   host   mutation,   it   also   leverages   and   enhances   well   known   techniques   such   as   RTBH   and   SAVI.   Moreover,   SDN   provides   flexibility   to   use   techniques   such   as   anomaly   detection   or   to   forward   the   traffic   in   a   seamless   way   that   is   now   feasible   with   traditional  non-­‐SDN  techniques.     Albeit   all   proposals   have   security   benefits,   some   may   have   limitations   in   real   world   deployment.   Some   proposals  such  as  (Braga  2010),  assume  a  full  SDN  deployment  while  other  proposals  like  (Giotis  2014)  use  SDN   as  an  add-­‐on  to  a  traditional  network.  Despite  the  fact  that  SDN  deployments  are  increasing,  most  networks   are   still   traditional,   and   the   change   to   a   programmable   networking   paradigm   will   be   gradual.   This   can   be   a   leverage  to  solutions  that  use  SDN  in  combination  with  the  traditional  network,  without  the  need  to  rebuild   the  network.     As  SDN  technologies   gradually  reach   production  networks,  these  and  forthcoming  approaches   are  expected   to   strongly  improve  network  security.     5 Acknowledgements   This   work   was   partially   funded   by   PT   Inovação   (in   the   scope   of   Project   HolisticSDN)   and   by   project   iCIS   (Intelligent  Computing  in  the  Internet  of  Services  –  CENTRO-­‐07-­‐0224-­‐FEDER-­‐002003).     References   Antonatos,  S.  et  al.,  2005.  Defending  Against  Hitlist  Worms  Using  Network  Address  Space  Randomization.  In   Proceedings  of  the  2005  ACM  Workshop  on  Rapid  Malcode.  WORM  ’05.  New  York,  NY,  USA:  ACM,  pp.  30–40.    

Apps,  P.,  2014.  DDoS  cyber  attacks  get  bigger,  smarter,  more  damaging.  Available  at:   http://www.reuters.com/article/2014/03/05/us-­‐cyber-­‐ddos-­‐idUSBREA240XZ20140305.   Atighetchi,  M.  et  al.,  2003.  Adaptive  use  of  network-­‐centric  mechanisms  in  cyber-­‐defense.  In  Object-­‐Oriented   Real-­‐Time  Distributed  Computing,  2003.  Sixth  IEEE  International  Symposium  on.  pp.  183–192.   Bagnulo,  M.  &  García-­‐Martínez,  A.,  2013.  SAVI:  The  IETF  standard  in  address  validation.  Communications   Magazine,  IEEE,  51(4),  pp.66–73.   Bi,  J.,  Hu,  P.  &  Li,  P.,  2010.  Study  on  Classification  and  Characteristics  of  Source  Address  Spoofing  Attacks  in  the   Internet.  In  Networks  (ICN),  2010  Ninth  International  Conference  on.  pp.  226–230.   Braga,  R.,  Mota,  E.  &  Passito,  A.,  2010.  Lightweight  DDoS  flooding  attack  detection  using  NOX/OpenFlow.  In   Local  Computer  Networks  (LCN),  2010  IEEE  35th  Conference  on.  pp.  408–415.   Ellanti,  M.N.  et  al.,  2005.  Next  generation  transport  networks:  data,  management,  and  control  planes,   Springer.   Feamster,  N.,  Rexford,  J.  and  Zegura,  E.  (2013).  The  Road  to  SDN.  Queue,  11(12),  pp.20-­‐40.   Finkle,  J.,  2014.  Sony  works  for  3rd  day  to  restore  PlayStation  after  attack.  Reuters.  Available  at:   http://www.reuters.com/article/2014/12/27/us-­‐sony-­‐cybersecurity-­‐playstation-­‐idUSKBN0K50FV20141227.   Giotis,  K.,  Androulidakis,  G.  &  Maglaris,  V.,  2014.  Leveraging  SDN  for  Efficient  Anomaly  Detection  and   Mitigation  on  Legacy  Networks.  In  Software  Defined  Networks  (EWSDN),  2014  Third  European  Workshop  on.   pp.  85–90.   Greenberg,  A.  et  al.,  2008.  The  Cost  of  a  Cloud:  Research  Problems  in  Data  Center  Networks.  SIGCOMM   Comput.  Commun.  Rev.,  39(1),  pp.68–73.  Available  at:  http://doi.acm.org/10.1145/1496091.1496103.   Greene,  K.  (2009).  Tech  Review  10  Breakthrough  Technologies:  Software-­‐Defined  Networking.  [online]  MIT   Technology  Review.  Available  at:  http://Tech  Review  10  Breakthrough  Technologies:  Software-­‐Defined   Networking  [Accessed  12  Dez.  2014].   GuardiCore  (2014).  Data  center  security  redefined.  [online]  Technical  report,  Hewlett-­‐Packard.  Available  at:   http://h20195.www2.hp.com/V2/GetDocument.aspx?docname=4AA5-­‐1629ENW.  [Accessed  10  Dez.  2014].   Gude,  N.  et  al.,  2008.  NOX:  Towards  an  Operating  System  for  Networks.  SIGCOMM  Comput.  Commun.  Rev.,   38(3),  pp.105–110.  Available  at:  http://doi.acm.org/10.1145/1384609.1384625.   Hewlett-­‐Packard,  (2014).  Data  center  security  redefined.  Available  at:   http://h20195.www2.hp.com/v2/GetPDF.aspx%2F4AA5-­‐1629ENW.pdf.   Muelder,  C.,  Ma,  K.-­‐L.  &  Bartoletti,  T.,  2006.  Interactive  Visualization  for  Network  and  Port  Scan  Detection.  In   Proceedings  of  the  8th  International  Conference  on  Recent  Advances  in  Intrusion  Detection.  RAID’05.  Berlin,   Heidelberg:  Springer-­‐Verlag,  pp.  265–283.   Jafarian,  J.H.,  Al-­‐Shaer,  E.  &  Duan,  Q.,  2012.  Openflow  Random  Host  Mutation:  Transparent  Moving  Target   Defense  Using  Software  Defined  Networking.  In  Proceedings  of  the  First  Workshop  on  Hot  Topics  in  Software   Defined  Networks.  HotSDN  ’12.  New  York,  NY,  USA:  ACM,  pp.  127–132.  Available  at:     Jarraya,  Y.,  Madi,  T.  and  Debbabi,  M.  (2014).  A  Survey  and  a  Layered  Taxonomy  of  Software-­‐Defined   Networking.  IEEE  Commun.  Surv.  Tutorials,  16(4),  pp.1955-­‐1980.   Kewley,  D.  et  al.,  2001.  Dynamic  approaches  to  thwart  adversary  intelligence  gathering.  In  DARPA  Information   Survivability  Conference  amp;  Exposition  II,  2001.  DISCEX  ’01.  Proceedings.  pp.  176–185  vol.1.   Koomey,  J.G.,  2008.  Worldwide  electricity  used  in  data  centers.  Environmental  Research  Letters,  3(3),  p.34008.  

Kreibich,  C.  &  Warfield,  A.,  2005.  Using  packet  symmetry  to  curtail  malicious  traffic.  ACM  Hotnets  …,  200.   Available  at:  http://conferences.sigcomm.org/hotnets/2005/papers/kreibich.pdf  [Accessed  February  2,  2015].   Kreutz,  D.,  Ramos,  F.,  Verissimo,  P.,  Rothenberg,  C.,  Azodolmolky,  S.  and  Uhlig,  S.  (2014).  Software-­‐Defined   Networking:  A  Comprehensive  Survey.  Proc.  IEEE,  103(1),  pp.14-­‐76.   Kumari,  W.  &  McPherson,  D.,  2009.  Remote  Triggered  Black  Hole  Filtering  with  Unicast  Reverse  Path   Forwarding  (uRPF).  ,  (5635).  Available  at:  http://www.ietf.org/rfc/rfc5635.txt.   Kwon,  J.,  Seo,  D.,  Kwon,  M.,  Lee,  H.,  Perrig,  A.,  &  Kim,  H.  (2015).  An  incrementally  deployable  anti-­‐spoofing   mechanism  for  software-­‐defined  networks.  Computer  Communications.   Levin,  D.,  Canini,  M.,  Schmid,  S.,  &  Feldmann,  A.  (2013).  Panopticon:  Reaping  the  benefits  of  partial  sdn   deployment  in  enterprise  networks.  TU  Berlin/T-­‐Labs,  Tech.  Rep.   Liu,  B.,  Bi,  J.,  &  Zhu,  Y.  (2011).  A  deployable  approach  for  inter-­‐AS  anti-­‐spoofing.  In  Network  Protocols  (ICNP),   2011  19th  IEEE  International  Conference  on  (pp.  19–24).  doi:10.1109/ICNP.2011.6089052   Liu,  Y.  et  al.,  2014.  Research  of  Data  Center  Fresh  Air  Ventilation  Cooling  System.  In  A.  Li,  Y.  Zhu,  &  Y.  Li,  eds.   Proceedings  of  the  8th  International  Symposium  on  Heating,  Ventilation  and  Air  Conditioning  SE    -­‐  30.  Lecture   Notes  in  Electrical  Engineering.  Springer  Berlin  Heidelberg,  pp.  299–306.  Available  at:   http://dx.doi.org/10.1007/978-­‐3-­‐642-­‐39581-­‐9_30.     McPherson,  D.,  Baker,  F.  &  Halpern,  J.,  2013.  Source  Address  Validation  Improvement  (SAVI)  Threat  Scope.  ,   (6959).  Available  at:  http://www.ietf.org/rfc/rfc6959.txt.   Mirkovic,  J.  et  al.,  2004.  Internet  Denial  of  Service:  Attack  and  Defense  Mechanisms  (Radia  Perlman  Computer   Networking  and  Security),  Upper  Saddle  River,  NJ,  USA:  Prentice  Hall  PTR.   ONF,  Open  Networking  Foundation.  2014.  Available  at:  https://www.opennetworking.org/.   R.  Presuhn,  “Version  2  of  the  Protocol  Operations  for  the  Simple  Network  Management  Protocol  (SNMP),”  RFC   3416  (INTERNET  STANDARD),  Internet  Engineering  Task  Force,  Dec.  2002.  [Online].  Available:   http://www.ietf.org/rfc/rfc3416.txt   Roesch,  M.  &  others,  1999.  Snort:  Lightweight  Intrusion  Detection  for  Networks.  In  LISA.  pp.  229–238.     Schudel,  G.  &  Smith,  D.,  2007.  Router  Security  Strategies:  Securing  IP  Network  Traffic  Planes,  Pearson   Education.   Champagne,  D.  &  Lee,  R.B.,  2006.  Scope  of  DDoS  Countermeasures:  Taxonomy  of  Proposed  Solutions  and   Design  Goals  for  Real-­‐World  Deployment.  Available  at:   http://palms.ee.princeton.edu/PALMSopen/champagne06DDoS.pdf.   SDX  Central  (2014).  What  is  OpenFlow?.  Available  at:  https://www.sdxcentral.com/resources/sdn/what-­‐is-­‐ openflow/   Shin,  S.,  Porras,  P.,  Yegneswaran,  V.,  Fong,  M.,  Gu,  G.  and  Tyson,  M.  (2013).  FRESCO:  Modular  Composable   Security  Services  for  Software-­‐Defined  Networks.  In:  NDSS  Symposium.   Simões,  P.  et  al.,  2013.  On  the  use  of  Honeypots  for  Detecting  Cyber  Attacks  on  Industrial  Control  Networks.  In   12th  European  Conference  on  Information  Warfare  and  Security  (ECIW  2013).   Song,  Y.,  Shin,  S.  &  Choi,  Y.,  2014.  Network  Iron  Curtain:  Hide  Enterprise  Networks  with  OpenFlow.  In  Y.  Kim,  H.   Lee,  &  A.  Perrig,  eds.  Information  Security  Applications  SE  -­‐  14.  Lecture  Notes  in  Computer  Science.  Springer   International  Publishing,  pp.  218–230.  Available  at:  http://dx.doi.org/10.1007/978-­‐3-­‐319-­‐05149-­‐9_14.   Spitzner,  L.,  2003.  Honeypots:  Tracking  Hackers,  Addison-­‐Wesley.  

Stutter,  J.D.,  2009.  Twitter  hit  by  denial-­‐of-­‐service  attack.  CNN.  Available  at:   http://edition.cnn.com/2009/TECH/08/06/twitter.attack/index.html.   Wang,  B.  et  al.,  2014.  DDoS  Attack  Protection  in  the  Era  of  Cloud  Computing  and  Software-­‐Defined   Networking.  In  Network  Protocols  (ICNP),  2014  IEEE  22nd  International  Conference  on.  pp.  624–629.   Yao,  G.,  Bi,  J.  &  Xiao,  P.,  2011.  Source  address  validation  solution  with  OpenFlow/NOX  architecture.  In  Network   Protocols  (ICNP),  2011  19th  IEEE  International  Conference  on.  pp.  7–12.   Yeganeh,  S.H.,  Tootoonchian,  A.  &  Ganjali,  Y.,  2013.  On  scalability  of  software-­‐defined  networking.   Communications  Magazine,  IEEE,  51(2),  pp.136–141.   Yu,  S.,  2014.  An  Overview  of  DDoS  Attacks.  In  Distributed  Denial  of  Service  Attack  and  Defense  SE    -­‐  1.   SpringerBriefs  in  Computer  Science.  Springer  New  York,  pp.  1–14.  Available  at:  http://dx.doi.org/10.1007/978-­‐ 1-­‐4614-­‐9491-­‐1_1.