Presentation is loading. Please wait.

Presentation is loading. Please wait.

Interference-Aware Fair Control in Wireless Sensor Networks Present by Zhe Zhou.

Similar presentations


Presentation on theme: "Interference-Aware Fair Control in Wireless Sensor Networks Present by Zhe Zhou."— Presentation transcript:

1 Interference-Aware Fair Control in Wireless Sensor Networks Present by Zhe Zhou

2 2 Outline  Introduction  Related Work  Motivation and Definitions  IFRC Design  Parameter Selection In IFRC  Evaluation  Conclusions

3 3 Outline  Introduction  Related Work  Motivation and Definitions  IFRC Design  Parameter Selection In IFRC  Evaluation  Conclusions

4 4 Introduction  We need congestion control in wireless sensor network –Structural Health Monitoring –Flat sensor network for low-rate periodic sensing –Tiered sensor network for high data-rate applications : complicated topology makes congestion control more tricky

5 5 Introduction  How to ensure fair and efficient transmission rates for each nodes in a sensor network?  Interference-Aware Fair Rate Control (IFRC) –Transport layer, based on CSMA and routing layer (link quality based path selection) –Distributed –Use average queue length to detect congestion –Low-overhead congestion sharing –Signals all related nodes –Use AIMD to converge to fairness

6 6 Introduction  The challenge –Hard to determine the related nodes –Hard to rapidly signal them

7 7 Outline  Introduction  Related Work  Motivation and Definitions  IFRC Design  Parameter Selection In IFRC  Evaluation  Conclusions

8 8 Related Work  TCP Congestion Control  AQM (Active Queue Management)  TCP for ad-hoc wireless networks  Extension of RED on wireless networks  Congestion mitigation and congestion control  ……

9 9 Outline  Introduction  Related Work  Motivation and Definitions  IFRC Design  Parameter Selection In IFRC  Evaluation  Conclusions

10 10 Motivation and Definitions   r 20   r 21   r 16 +r 20 +r 21

11 11 Motivation and Definitions  Assumptions –TinyOS –CSMA (Carrier Sense Multiple Access) and RTS/CTS( Request to Send / Clear to Send) –Token-Based and TDMA MACs are not considered –Static Routing Tree in most experiments IFRC can adapt to changes in routing tree IFRC achieves higher overall throughput on routing protocols based on link-quality merics

12 12 Motivation and Definitions  Assumptions (continued) –Link-Layer Retransmissions IFRC performs well when link-layer retransmissions recover from most packet losses Impact of packet losses will be described later  Definitions –Fair and efficient Each flow fairly divides the channel capacity IFRC – Each flow receives at least the most congested fair share rate Not absolutely fair – Flows having fewer contenders can send at a higher rate to ensure overall efficiency

13 13 Motivation and Definitions  Definitions (continued) –Interfering Links A link l 1 interferes with a link l 2 if a transmission along l 1 prevents the initiation or the successful reception of a transmission along l 2 –Potential Interferer A node n 1 is a potential interferer of node n 2 if a flow originating from node n 1 uses a link that interferes with the link between n 2 and its parent

14 14 Motivation and Definitions

15 15 Motivation and Definitions  In tree-based communication, the potential interferer of a node include: –Its subtree –Its neighbor and parent’s subtree –Its parent’s neighbor’s subtree  Definition (again!) –F i – Set of flows routed through node i, including flows originating at i and its subtree

16 16 Motivation and Definitions  Definition (continued) –B : Nominal total bandwidth –F i = F i + F j, j is either a neighbor of i, or a neighbor of i ’s parent ( set of all potential interferers) –f l,i : the assigned rate of each flow in F i –f l : minimum of all f l,i

17 17 Motivation and Definitions  F 16 = {20, 21, 14, 16, 17, 13, 12, 15, 18, 19}  Nodes contribute to the arrival rate of 16 : 16, 20, 21  Nodes contend with 16: others

18 18 Outline  Introduction  Related Work  Motivation and Definitions  IFRC Design  Parameter Selection In IFRC  Evaluation  Conclusions

19 19 IFRC Design  Main Task –Congestion Detection –Signaling –Rate Adaptation  Congestion Detection –Channel Utilization –Queue size –With a MAC with carrier-sense, backoffs and retransmission, overloaded traffic will increase the queue size. Therefore, we simply use queue size to indicate congestion.

20 20  Congestion Detection (con.) –EWMA (Exponentially Weighted Moving Average) for estimating average queue size –Updated for each packet inserting –A node is congested if avg q > U, and returns to uncongested state if avg q < L –Sometimes a single halving is not enough. To determine if multiple halving should be executed, we need multiply U. IFRC Design

21 21 IFRC Design  Congestion Detection (con.) –We define multiply U as below ( k is a small integer and I is a constant increment of queue length) –So that as k increases, the difference between U(k) and U(k+1) decreases, resulting in more frequent rate halving which accelerates the draining of queue.

22 22 IFRC Design  Congestion Sharing –Insert congestion related information in header of each outgoing packet Current r i and average queue length A bit indicating whether any of its children is congested The smallest rate r l among all its congested children and l ’s average queue length –To this point, all neighbors of an arbitrary node can receive the congestion information of this node and the nodes in its subtree.

23 23 IFRC Design  Congestion Sharing (con.) –Two rules for implicitly notify all potential interferers –Child’s rate can never surpass parent –A node will adapt its rate when congestion occurs either at its neighbor or the neighbor’s subtree

24 24 IFRC Design  Rate Adaptation –Average value of r i is not the max rate by which i generate traffic –At the beginning, a node starts its sending rate at r init and add Φ to its rates every 1/ r i seconds. –The node continues to increase the rate until itself congested or the two rules satisfied; Then it adapts the rate accordingly. –After the adaptation, the node increases its r i by δ /r i every 1/ r i seconds.

25 25 IFRC Design  Base Station Behavior –Sets the initial rate r b to the nominal rate of the channel and do not increases it –If any of its children is congested, decreases its rate, and broadcasts it twice –After each adaptation, increments r b by δ/ r b every 1/ r b seconds. As the station itself has no data to send, it broadcasts its rate after at least m packet have been received from the fastest child.

26 26 IFRC Design  Extension to IFRC –Multiple Base Stations If one of the children of the base station is congested, the base station sends a control packet indicating that. –Weight Fairness –When only a subset of nodes transmit

27 27 IFRC Design  Discussion –IFRC can not implemented over an unreliable MAC layer –IFRC can not detect interference from non- neighboring nodes –IFRC can not work on cards turning off overhearing (Battery Killer!) –IFRC will work when intermediate nodes perform in-network aggregation

28 28 Outline  Introduction  Related Work  Motivation and Definitions  IFRC Design  Parameter Selection In IFRC  Evaluation  Conclusions

29 29 Parameter Selection In IFRC  Intensity in AIMD –Each node i increases its rate r i by δ /r i every 1/ r i seconds. –Namely, it follows a linear curve with slope δ. –For efficiency, δ should be as large as possible. However, for stability δ should be kept not too large. So, our task is to find its upper bound in terms of maintaining the stability

30 30 Parameter Selection In IFRC r st,i – r min,i > r max,i – r st,i r st,i > ( r max,i – r min,i ) / 2 r st,i > 3 * r max,i / 4 Equation 1

31 31 Parameter Selection In IFRC r min,i r max,i  To prevent r i ramping from r min,i to r max,i in one step (in 1/ r i seconds), we need δ /r min,i << r min,i, or –Where 0< ε <1 is a small positive number. We will derive its upper bound below.  The excess number of packets can be calculated as  If we focus on one congested node j, and I ij be the function that indicates whether packets from i traverse j. The total number of excess packets could then be denoted as:

32 32 Parameter Selection In IFRC  Taking the effect of contention into account, we substitute I ij with f ij.  We need to tune the value of to validate the following two equations:  Equation 1, 2, 3 guarantee system stability and only one signal is sent for one node when congestion occurs, which mitigates the reduce of efficiency. Equation 2 Equation 3

33 33 Parameter Selection In IFRC  By substituting r st in Equation 2 using Equation 1 and let F j = Σ i f ij, we get r st =1.5 * r min   (See the figure) As r st rises, the difference between the area of two triangles increase, thus the efficiency decreases.   As r st drops, the upper bound of εdrops, so we will get a smallerε.

34 34 Parameter Selection In IFRC  To prevent a node sending out congestion info in the duration of receiving other node’s congestion info, we have:  And consequently, we have:   The # of packet updates performed at node i before it receives the congestion info from j Average of s i

35 35 Parameter Selection In IFRC  In small network when is small, the first inequality determines.  In small network when F j is small, the first inequality determines ε.  In large network when is large, the second inequality determines.  In large network when F j is large, the second inequality determines ε.  Use for.  Use nlogn for F j (Intuitively, every node interferers with j for logn times).  should be something proportional to  r st should be something proportional to B/nlogn, so we set r init to B/10nlogn.   Φ is set to r init /8.   U(0) and U(1) are set to N/2 and N respectively.  So is restricted by these two equations:  So ε is restricted by these two equations:

36 36 Outline  Introduction  Related Work  Motivation and Definitions  IFRC Design  Parameter Selection In IFRC  Evaluation  Conclusions

37 37 Evaluation  Implementation and Methodology –40-node wireless sensor testbed –TinyOS 1.1 with IFRC plugged in –Two modules. Neighbor’s congestion table is stored. –Promiscuous mode enabled, which disables the chip-level ack, thus ack in MAC is added. –Each node: Moteiv Tmote with a 8MHz Texas Instruments MSP430 microcontroller, 10KB RAM and a 2.4GHz IEEE 802.15.4 Chipcon Wireless Transceiver with a nominal bit rate of 250 Kbps –Deployed over 1125 sqare meters of a large office floor –A USB backchannel for logging experiment data (which will have some problem later) –8 hops, all links have a loss rate lower that 40%, pretty uncomplaining

38 38 Evaluation Testbed connectivity graph

39 39 Evaluation Window based Pretty small! Really slow slow-start

40 40 Evaluation  A fixed tree to maintain a same environment for all experiments (modifies MultiHopLQI)  A hour at least for each experiment  Long experiments, run at usually late at night or in early morning  Every packet transmission, reception, and every change in rate at each node (including base station, although no transmission) is recorded.

41 41 Evaluation   Packet reception ratios range from 66% to 96%   9 hops deep   A good topology with all kinds of variance

42 42 Evaluation Packet transmitted Packet transmitted and received Packet transmitted but lost Base station control traffic   Every nodes receive approximately fair rates and goodput  Node 13 and 8 are congested (hard to perceive from the graph)  Hop-by-hop recovery resulted in fewer that 8% packet loss

43 43 Evaluation  Instantaneous goodput is stable, with minor variations attribute to AIMD.

44 44 Evaluation  Nodes adapt their rate nearly synchronically  Slow start and AIMD is clear visible  Some nodes adapt their rate slower due to network lantency (not shown)  Horizontal line caused by experiment data loss resulted from USB issues

45 45 Evaluation  Queue never builds up to higher than 25 -> no packet loss

46 46 Evaluation  The efficiency is pretty encouraging. 80% efficiency 60% efficiency

47 47 Evaluation Two successive decreases

48 48 Evaluation

49 49 Evaluation

50 50 Evaluation

51 51 Evaluation

52 52 Evaluation

53 53 Evaluation

54 54 Outline  Introduction  Related Work  Motivation and Definitions  IFRC Design  Parameter Selection In IFRC  Evaluation  Conclusions

55 55 Conclusions  Conclusion –IFRC is the first practical interference-aware rate control mechanism for WSN –IFRC is fair –In terms of efficiency, IFRC is questionable  Future work –Implement reliability in IFRC –A more rigorous proof of the choice of IFRC parameters –A complete analysis of the effects of other factors on IFRC

56 56 Thank you!


Download ppt "Interference-Aware Fair Control in Wireless Sensor Networks Present by Zhe Zhou."

Similar presentations


Ads by Google