Docstoc

dsf

Document Sample
dsf Powered By Docstoc
					                               Yu Gu and Tian He
  Minnesota Embedded Sensor System (MESS)
Department of Computer Science & Engineering
                                               http://mess.cs.umn.edu

          This work is supported by National Science Foundation
                                                                     Yu Gu@SenSys’07




 Sleep Latency in Low Duty-Cycle
 Sensor Networks                                     Sleep now. Wake
                                                     up in 57seconds
Sleep now. Wake
up in 35 seconds
                                                                    D
        B


                                                      57s latency
                   35s latency
                                    A



                               13s latency   4s latency                 E
            C

            Sleep now. Wake                               Sleep now. Wake
            up in 13 seconds                              up in 4 seconds
                                Yu Gu@SenSys’07




Unreliable Radio Links
                                D
  B
                          70%


          90%
                      A


                          50%
                95%
      C                             E
                                                               Yu Gu@SenSys’07




State-of-the-art Solutions: ETX (MobiCom’03)
                                                ETX only considers link
            ETX = 1/0.5 + 1/0.5 = 4
                                                quality


                                   B       50%, 100s
        50%, 100s
                 Expected E2E
           quality based solutions
   Sole link     delay is 400s     cannot help
  Areduce E2E delay in extremely low-duty cycle
                                          D
   sensor networks!
                 Expected E2E
                    delay is 50s
                                           40%, 10s
         40%, 10s
                                       C

               ETX = 1/0.4 + 1/0.4 = 5
                                                              Yu Gu@SenSys’07




State-of-the-art Solutions: DESS (INFOCOM’05)
                    DESS = 10 + 10 = 20s        DESS only considers
                                                sleep latency

                                 B
                                           10%, 10s
         10%, 10s
                  Expected E2E
  Sole sleeplatency based solutions
                  delay is 200s     cannot help
                                          D
                   in extremely low-duty cycle
  reduce E2E delayExpected E2E
   A
  sensor networks!delay is 40s
                                           100%, 20s
         100%, 20s
                                     C

                     DESS = 20 + 20 = 40s
                                                                    Yu Gu@SenSys’07




State-of-the-art Solutions (2)
         80 fold performance                       20 fold performance
         degradation!                              degradation!




Only Consider impact of link qualities   Only Consider impact of Duty Cycling

     Intelligent MAC protocols (B-MAC, S-MAC, SCP-MAC …) provide
     significant performance improvement at the MAC layer.

     We focus on further performance improvement at the network layer.
                  Yu Gu@SenSys’07




Outline
 Motivation
 Network Model
 DSF Design
 Evaluation
 Conclusion
                                       Yu Gu@SenSys’07




Sensor States Representation
 Scheduling Bits
   (10110101)*             10110101



 Switching Rate            Off
   0.5HZ 16s round time   On
                                                                                        Yu Gu@SenSys’07




   Data Delivery Process
(1 0 0 0 0 0 0 0 0 0)*   (0 1 0 0 0 0 0 0 0 0)*   (0 0 0 1 0 0 0 0 0 0)*   (0 0 0 0 0 0 1 0 0 0)*

             1                       2                      3                      4


             Sleep latency is 1       Sleep latency is 2        Sleep latency is 3



                                         E2E Delay is 6
                                                                                  Yu Gu@SenSys’07




   Main Idea
                                     Sleep latency is 1


                       try a 0 0 0 0)* (0 of 0 0 0 0 0 0)* nodes
           We should 1 0 0 0 0sequence 0 0 1forwarding(0 0 0 0 0 0 1 0 0 0)*
                    (0
(1 0 0 0 0 0 0 0 0 0)*
           instead of a fixed forwarding node!
             1                      2                      3                 4
                         (0 0 1 0 0 0 0 0 0 0)*

                                                (DSF) is important
      Dynamic Switching-based Forwarding Sleep latency is 1
                         5         1st attempt:
      in extremely low duty-cyclethsensor networks. is 1 + 10 * (i-1)
                              nd i attempt: Sleep latency
                                          2       attempt: Sleep latency is 1 + 10 =11
2nd attempt: Sleep latency is 1 + 1 =2
                                                             Yu Gu@SenSys’07




Optimization Objectives
 EDR: Expected Delivery Ratio
         Assisted            Border              Environmental
         Living              Control             Monitoring



 EED: Expected End-to-End Delay
                                   Disaster              Traffic
                Target
                                   Response              Control
                Tracking


 EEC: Expected Energy Consumption
                Habit                  Space           Precision
                Monitoring             Monitor         Agriculture
                                                                  Yu Gu@SenSys’07




                     Forwarding Sequence
Optimization Objectives(1) : EDR
                                    EDR: Expected Delivery Ratio.
               2   (010)*
(100)*             EDR = 70%
         60%
                                      EDR for node 1 is (EDR1):
  1      50%
               3   (001)*
                   EDR = 80%
                                       0.6*0.7 + (1-0.6)*0.5*0.8
      40%                              + (1-0.6)*(1-0.5)*0.4*0.9
               4   (100)*
                   EDR = 90%
                                     Yu Gu@SenSys’07




Optimization Objectives(2)
 EDR: Expected Delivery Ratio



 EED: Expected End-to-End Delay



 EEC: Expected Energy Consumption
                                                                   Yu Gu@SenSys’07




Optimizing EDR
                                   Shall we try all available neighbors?
                    (010)*
                2   EDR = 70%      If both node 2 and node 3 are selected
(100)*                             as forwarding nodes:
         100%
                                   EDR a 1 * 0.7 of
                We should only choose 1 =subset = 0.7
  1             neighboring nodes as forwarding nodes!
         100%                      If only node 3 is selected as
                    (001)*         forwarding node:
                3   EDR = 80%
                                   EDR1 = 1 * 0.8 = 0.8
                                                                   Yu Gu@SenSys’07



Optimizing EDR with dynamic
programming
                   Try or skip

                                  Select only a subset of neighbors as
               2      (010)*
                                  forwarders
(100)*                EDR = 70%
         60%       Try or skip    Node 4 has to be selected
  1      50%
               3      (001)*
                      EDR = 80%
                                  Then we attempt to add more nodes
      40%          Try or drop    into the forwarding sequence
                                  backwardly.
               4      (100)*
                      EDR = 90%
                                                                           Yu Gu@SenSys’07




Distributed Implementation
     EDR = 99%, EED = 15, EEC = 2          EDR = 98%, EED = 2, EEC = 1



                1                                   3
                       EDR = 100%, EED = 0, EEC = 0


                                    sink


                2                                     4
  EDR = 97%, EED = 20, EEC = 5             EDR = 90%, EED = 90, EEC = 12
                                                   Yu Gu@SenSys’07




   Interesting Findings
     Temporary routing loops may be helpful on reducing
      E2E Delay
   (111111)*    (010000)*

       1   (90%,1)
                     2      (100%,1)

                                 (111111)*
(100%,1)
                                5
           (90%,1)          (100%,1)
       3             4
   (111111)*    (000010)*
                  Yu Gu@SenSys’07




Outline
 Motivation
 Network Model
 DSF Design
 Evaluation
 Conclusion
                                                    Yu Gu@SenSys’07




Evaluations
 Both testbed implementation and large-scale
  simulations

 Baseline solutions:
    ETX by Douglas S.J. De Couto et al. in Mobicom’03
    PRR*D by Karim Seada et al. in SenSys’04
    DESS by Gang Lu et al. in INFOCOM’05
                                                             Yu Gu@SenSys’07




Testbed Results




 20 MicaZ nodes, 27,398 bytes code memory and 1,137 bytes data memory
                         Yu Gu@SenSys’07




Simulation Results (1)
                DSF
                                          Yu Gu@SenSys’07




Simulation Results (2)

               DSF converges to DESS at
               perfect link
                         Yu Gu@SenSys’07




Simulation Results (3)
                                                    Yu Gu@SenSys’07




Conclusion
 A Dynamic Switch-based Forwarding (DSF) scheme
 for extremely low duty-cycle sensor networks
   Addressed both sleep latency and lossy radio links
   Dynamic switching is essential


 Distributed model for data delivery ratio (EDR), E2E
 delay (EED) and energy consumption (EEC).
   Optimal forwarding on these three metrics
   A generic metrics that converge to ETX (in always-awake
    networks) and DESS (in perfect-link networks)

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:1
posted:11/14/2012
language:English
pages:24