ALABAMO: A LoAd BAlancing MOdel for RPL

27 downloads 817 Views 2MB Size Report
Network control messages (DIO, DAO, DIS). – Each node ... Extend the network lifeSme avoiding holes in the network. 9 ... Rou ng tree. – Network delivery ra o.
ALABAMO:  A  LoAd  BAlancing  MOdel  for  RPL   Tarcísio  Bruno  C.  Oliveira,  Pedro  Henrique  Gomes,   Danielo  G.  Gomes,  Bhaskar  Krishnamachari      

1

Outline   • Mo8va8on     • Background  and  Problem  Statement   • Objec8ves   • Proposal   – ALABAMO:  A  LoAd  BAlancing  MOdel  for  RPL   • Performance  Evalua8on   • Related  Work   • Conclusion   2

Mo8va8on   •  Energy  is  a  constrained  resource  in  Low-­‐Power   and  Lossy  Networks  (LLNs)   •  RPL  became  the  standardized  IPV6  rou8ng   protocol  for  LLNs  in  2012   •  The  default  RPL  objec8ve  func8on  uses  link   quality  as  rou8ng  criterion  which  can  cause  an   unbalanced  distribu8on  of  nodes     3

Mo8va8on   •  Unbalanced  Network    

 

4

Background   •  RPL  -­‐  Rou8ng  Protocol  for  LLNs   – IPv6  support  

– Tree-­‐like  topology   – Network  control  messages  (DIO,  DAO,  DIS)  

– Each  node  has  a  rank  

– The  ranks  are  calculated  using  an  ObjecSve   Funcion  (OF)     5

Background   •  Objec8ve  Func8ons   – OF0  

• It  uses  hop-­‐count  as  a  rouSng  metric  

– Minimum  Rank  With  Hystheresis  Objec8ve   Func8on  (MRHOF)  

• It   chooses   routes   that   minimize   rouSng   metrics   as   latency,   hop-­‐count   and   ETX   (Expected   Transmission   Count)    

6

Background  

7

Problem  Statement   Sensor  nodes  that  have  links  with  beZer  quality   will  forward  more  packet  than  others.  Thus,    their   energy  deple8on  will  be  notably  faster.    

8

Objec8ves    

•  Develop  a  mechanism  based  on  RPL  to  force  a   homogeneous  energy  consump8on    among  the   nodes   •  Extend  the  network  life8me  avoiding  holes  in   the  network  

9

Objec8ves     •  Balanced  Network    

10

ALABAMO   •  It  takes  into  account  the  traffic  profile  to  avoid   overloading  the  nodes  with  high-­‐quality  links:   – Based  on  MRHOF;  

– All  nodes  broadcast  the  number  of  transmiWed   packets;  

– Hystheresis  mechanism  to  prevent  fast   fluctuaSons.    

11

ALABAMO   •  It  defines  two  auxiliary  constants  for  flexibility   purpose:   – ETX  

• Max  ETX  Ra8o   • ETX  RaSo  =  ETX1/ETX2  

– Sent  packets  

• Max  Workload  Ra8o   • Workload  RaSo  =  SENT_PKT1/SENT_PKT2  

 

12

ALABAMO  

13

ALABAMO  

14

Performance  Evalua8on   •  Alabamo  vs  MRHOF¹   •  41  motes  (Tutornet¹)   •  UDP  packets  every  30  seconds   •  Each  experiment  lasts  for  2  hours  and  it  was   repeated  12  8mes  

¹[Gnawali  and  Levis  2012]   ²hWp://anrg.usc.edu/www/tutornet/     15

Performance  Evalua8on  

16

Performance  Evalua8on   •  Parameters  

– Max  ETX  RaSo  equals  to  80  and  Max  ETX  RaSo   equals  to  90   – Max  Workload  RaSo  equals  to  70  

• Metrics  

– RouSng  tree   – Network  delivery  raSo   – Energy  consumpSon/network  lifeSme   – Parent  switching   17

Performance  Evalua8on   •  Rou8ng  tree   – A  snapshot  of  the  rouSng  tree  was  taken  every  30   minutes  

 

Std  devia8on  

Avg.  heaviest  sub-­‐ tree   size  

Std  devia8on  

4.00  

4.93  

18.00  

0.12  

ALABAMO-­‐80  

3.63  

3.79  

11.93  

1.67  

ALABAMO-­‐90  

3.64  

3.34  

10.33  

1.23  

Model  

Avg.  Sub-­‐tree   size  

MRHOF  

Table 1. Routing tree

18

Performance  Evalua8on   •  Network  delivery  ra8o  

Figura 6. Packet Delivery Ratio

19

Performance  Evalua8on   •  Energy  consump8on  

Std     devia8on  

Avg.     consump8on     (most  loaded  node  in   mW)  

Std     devia8on  

4.65

6.24

34.72

9.04

ALABAMO-­‐80  

4.07

3.08

16.69

3.70

ALABAMO-­‐90  

4.04

4.79

21.21

15.12

Model  

Avg.     consump8on     (per  node  in  mW)  

MRHOF  

Table 2. Energy Consumption

20

Performance  Evalua8on   •  Normalized  Network  life8me  

Figure 7. Network Lifetime

21

Performance  Evalua8on   •  Parent  switching   – MaxWorkloadRaSo   Model  

Average  number  of   parent  switches     (per  node)  

Std  devia8on  

MRHOF  

0.30  

0.69  

ALABAMO-­‐80  

1.24  

1.93  

ALABAMO-­‐90  

0.59  

1.25  

Table 3. Parent switching

22

Related  Work   •  Hierarchical  and  centralized  solu8ons   •  RPL-­‐based  solu8ons   – An  Energy  Efficient  and  Reliable  Composite  Metric  for   RPL  Organized  Networks  [Capone  et  al.  2014]   – Improving  the  network  lifeCme  with  energy-­‐   balancing  rouCng:  ApplicaCon  to  RPL  [Iova  et  al.  2014]   – Queue  UClizaCon  based  RPL  for  Load  Balancing  in   Large  Scale  Industrial  ApplicaCons  [Kim,  H.-­‐S.  2015]  

23

Conclusion   •  2-­‐fold  increase  in  network  life8me   •  More  balanced  trees   •  Trade-­‐off  between  network  life8me  and  PDR   •  Prac8cal  solu8on  tested  in  real  deployment   •  RPL-­‐compa8ble  and  totally  distributed     24

Thanks  !!!  

Contact   [email protected] r  

25