A Glimpse at Three Wireless Networking Problems


Published on

1 Like
  • Be the first to comment

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

A Glimpse at Three Wireless Networking Problems

  1. 1. A Glimpse at Three Wireless Networking Problems Bob Kinicki Computer Science Department [email_address] Colloquium October 5, 2007
  2. 2. Outline <ul><li>Thoughts and Mini-Motivation </li></ul><ul><li>Wireless Networking Primer </li></ul><ul><li>#1 Dynamic Rate Adaptation </li></ul><ul><ul><li>Performance problems with ARF </li></ul></ul><ul><ul><li>Rate Adaptation Algorithms </li></ul></ul><ul><ul><ul><li>RBAR, CARA, RFT and CARAF </li></ul></ul></ul><ul><li>Wireless Sensor Networks (WSNs) </li></ul><ul><li>#2 Dynamic Cluster Formation </li></ul><ul><li>#3 Power-Aware MAC Protocols </li></ul><ul><ul><li>SMAC, TMAC, WiseMAC, SCP-MAC and </li></ul></ul><ul><ul><li>Crankshaft </li></ul></ul>
  3. 3. My Research Space Networking Wireless Networking Wireless Sensor Networks 1983 2003 2006
  4. 4. The Future of Sensor Networks?
  5. 5. Wireless Primer
  6. 6. LAN Terminolgy <ul><li>802.3:: </li></ul><ul><li> Ethernet CSMA/CD </li></ul><ul><li>802.11a/b/g:: </li></ul><ul><li>WiFi CSMA/CA </li></ul><ul><li>802.15.4:: </li></ul><ul><li>ZigBee 802.11-based </li></ul><ul><ul><li>lower data rates, lower power </li></ul></ul><ul><li>Bluetooth:: </li></ul><ul><li>TDMA </li></ul><ul><li> - wireless Personal Area Networks (PANs) that provide secure, globally unlicensed short-range radio communication. </li></ul><ul><ul><li>Clusters with max of 8: cluster head + 7 nodes </li></ul></ul>WSNs
  7. 7. Wireless LANS <ul><li>Infrastructure with AP (Access Point) </li></ul><ul><li>Mobile Ad Hoc Networks (MANETs) </li></ul><ul><li>Wireless Sensor Networks (WSNs) </li></ul><ul><li>Interacting AP Topologies </li></ul>
  8. 8. Infrastructure Access Point client client client client Internet
  9. 9. Mobile Ad Hoc Network (MANET) BS Wireless Sensor Network (WSN)
  10. 10. Wireless LAN Protocols <ul><li>(a) A sending an RTS to B. </li></ul><ul><li>(b) B responding with a CTS to A. </li></ul>Tanenbaum slide node D is possible hidden terminal
  11. 11. Virtual Channel Sensing in CSMA/CA <ul><li>C (in range of A) receives the RTS and based on information in RTS creates a virtual channel busy NAV(Network Allocation Vector). </li></ul><ul><li>D (in range of B) receives the CTS and creates a shorter NAV. </li></ul>Tanenbaum slide
  12. 12. Basic CSMA/CA [N. Kim] possible collision !!
  13. 13. One-to-One Configuration {Ad Hoc} Access Point client
  14. 14. One-to-One Configuration {Ad Hoc} Access Point client
  15. 15. One-to-One Configuration {Ad Hoc} Access Point client Distance Impacts: attenuation fading interference
  16. 16. #1 Dynamic Rate Adaptation
  17. 17. 802.11 Physical Layer [N. Kim] ‘ Adjust transmission rate on the fly’
  18. 18. BER vs SNR [Pavon]
  19. 19. Throughput vs SNR [Pavon]
  20. 20. Rate Adaptation versus Distance [J. Kim]
  21. 21. Single AP multiple clients (homogeneous) Access Point client client client client Node Contention: Produces collisions
  22. 22. Node Contention [N. Kim] without RTS/CTS
  23. 23. Single AP multiple clients (heterogeneous) Access Point client client client client Multiple Node Effects Collisions AP queue overflow link capture hidden terminal performance anomaly different NIC cards (Rate Adaptation NOT Standardized!!)
  24. 24. Unfairness [Choi]
  25. 25. Multiple APs multiple clients (heterogeneous) Access Point client client client client Access Point client client client client
  26. 26. Hidden Terminals [Wong] Without a hidden terminal, loss ratio ~5.5%. One hidden AP with mild sending rate (0.379 Mbps) yields:
  27. 27. RTS/CTS Summary <ul><li>RTS/CTS can reduce collisions. </li></ul><ul><li>RTS/CTS can guard against and reduce hidden terminals. </li></ul><ul><li>RTS/CTS adds overhead that reduces throughput. </li></ul><ul><li>Normally, RTS/CTS is turned off! </li></ul>
  28. 28. Rate Adaptation Algorithms <ul><li>AARF ARF AMRR </li></ul><ul><li>CARA CROAR DOFRA </li></ul><ul><li>Fast-LA HRC LA </li></ul><ul><li>LD-ARF MiSer MultiRateRetry </li></ul><ul><li>MPDU OAR ONOE </li></ul><ul><li>PER RBAR RFT </li></ul><ul><li>RRAA SampleRate SwissRA </li></ul>
  29. 29. Rate Adaptation Algorithms <ul><li>1997 ARF </li></ul><ul><li>1998 </li></ul><ul><li>1999 </li></ul><ul><li>2000 </li></ul><ul><li>2001 RBAR </li></ul><ul><li>2002 MPDU OAR PER </li></ul><ul><li>2003 LA MiSer SwissRA </li></ul><ul><li>2004 AARF AMRR HRC MultiRateRetry </li></ul><ul><li>2005 Fast-LA LD-ARF RFT SampleRate </li></ul><ul><li>2006 CARA CROAR DOFRA RRAA </li></ul><ul><li>2007 </li></ul>
  30. 30. Rate Adaptation Algorithms <ul><li>Uses recent history and probes: ARF, AARF, SampleRate </li></ul><ul><li>Long interval smoothing: ONOE, SampleRate </li></ul><ul><li>Multiple rates: MultiRateRetry, AMRR, RRAA </li></ul><ul><li>Uses RTS/CTS: RBAR, OAR, CROAR, CARA </li></ul><ul><li>Uses RSSI to approximate SNR, each node maintains 12 dynamic RSS thresholds: LA </li></ul><ul><li>Puts checksum on header and use NACK to signal link loss error: LD-ARF </li></ul><ul><li>Table lookup with thresholds: HRC,MPDU(len,rSNR,count) </li></ul><ul><li>Fragmentation: DOFRA, RFT </li></ul><ul><li>Miscellaneous: PER, MiSer, SwissRA, Fast-LA </li></ul>
  31. 31. Auto Rate Fallback (ARF) <ul><li>If two consecutive ACK frames are not received correctly, the second retry and subsequent transmissions are done at a lower rate and a timer is started. </li></ul><ul><li>When the number of successfully received ACKs reaches 10 or the timer goes off, a probe frame is sent at the next higher rate. However, if an ACK is NOT received for this frame, the rate is lowered back and the timer is restarted. </li></ul>
  32. 32. ARF and AARF
  33. 33. Receiver Based Auto Rate (RBAR) <ul><li>Receivers control sender’s transmission rate. </li></ul><ul><li>RTS and CTS are modified to contain info on size and rate {not 802.11 compatible}. </li></ul><ul><li>Uses analysis of RTS reception (RSSI) to estimate SNR and send choice back to sender in CTS. </li></ul><ul><li>Receiver picks rate based on apriori SNR thresholds in a lookup table. </li></ul>
  34. 34. Collision Aware Rate Adaptation (CARA) <ul><li>CARA uses two methods for identifying collisions: </li></ul><ul><ul><li>RTS probing </li></ul></ul><ul><ul><li>Clear Channel Assessment (CCA) detection </li></ul></ul><ul><li>RTS Probing </li></ul><ul><li>{Idea: Assume all RTS/CTS transmission failures after a successful RTS/CTS exchange must be due to channel errors. </li></ul><ul><li>(Note – this assumes hidden terminals are not possible) } </li></ul>
  35. 35. RTS Probing <ul><li> CARA-1 </li></ul><ul><li>Data frame transmitted without RTS/CTS. </li></ul><ul><li>If the transmission fails, RTS/CTS exchange is activated for the next retransmission. If this retransmission fails {assume channel quality problem} , then the rate is lowered. </li></ul><ul><li>If retransmission with RTS/CTS is successful {assume collision occurred}, stay at same rate and send next frame without RTS/CTS. </li></ul>
  36. 36. Clear Channel Assessment (CCA) [J. Kim] ACK
  37. 37. CCA Option <ul><li>Case 2: It is a collision. </li></ul><ul><ul><li>Transmit without increasing failure count and lowering the transmission rate. No RTS/CTS probe is needed. </li></ul></ul><ul><li>Case 1 and Case 3: </li></ul><ul><ul><li>Initiate RTS/CTS probe scheme. </li></ul></ul>
  38. 38. CARA-1 (with RTS Probing) [J. Kim]
  39. 39. CARA-2 (with CCA) [J. Kim]
  40. 40. 802.11 MAC Fragmentation [Zhu]
  41. 41. Rate-based Fragmentation Thresholding (RFT) <ul><li>Fragmenting a frame can increase the probability of the fragment being received successfully. </li></ul><ul><li>Propose a dynamic fragmentation scheme with different fragmentation thresholds based on different channel conditions. </li></ul><ul><li>Namely, fragment sizes vary with the chosen adaptation rate. </li></ul>
  42. 42. CARAF (CARA with Fragmentation) <ul><li>Dan Courcey’s MS thesis: </li></ul><ul><li>Combine CARA with Fragmentation. </li></ul><ul><li>Top Level Scheme: </li></ul><ul><li>Upon CCA determination of collision, use fragmentation. </li></ul><ul><li>If CCA shows idle, initiate RTS/CTS probe. </li></ul><ul><li>If probe fails lower transmission rate. </li></ul><ul><li>{Investigate how to vary fragment size to maximize throughput and increase the likelihood of CCA case 2} </li></ul>
  43. 43. Wireless Sensor Networks (WSNs)
  44. 44. Wireless Sensor Networks <ul><li>A distributed connection of nodes that coordinate to perform a common task. </li></ul><ul><li>In many applications, the nodes are battery powered and it is often very difficult to recharge or change the batteries. </li></ul><ul><li>Prolonging network lifetime is a critical issue. </li></ul><ul><li>Sensors often have long period between transmissions (e.g., in seconds). </li></ul><ul><li>Thus, a good WSN MAC protocol needs to be energy efficient. </li></ul>
  45. 45. Wireless Sensor Networks <ul><li>Another attribute is scalability to change in network size, node density and topology. </li></ul><ul><ul><li>In general, nodes can die, join later or be mobile. </li></ul></ul><ul><li>Often high bandwidth is not important . </li></ul><ul><li>Nodes can take advantage of short-range, mulit-hop communication to conserve energy. </li></ul>
  46. 46. Wireless Sensor Networks <ul><li>Sources of energy waste: </li></ul><ul><ul><li>Idle listening, collisions, overhearing and control overhead. </li></ul></ul><ul><ul><li>Idle listening dominates (measurements show idle listening consumes between 50-100% of the energy required for receiving.) </li></ul></ul><ul><li>Idle listening:: listen to receive possible traffic that is not sent. </li></ul>
  47. 47. Power Measurements
  48. 48. Wireless Sensor Networks <ul><li>Duty cycle:: ratio between listen time and the full listen-sleep cycle. </li></ul><ul><li>central approach – lower the duty cycle by turning the radio off part of the time. </li></ul><ul><li>Three techniques to reduce the duty cycle: </li></ul><ul><ul><li>TDMA </li></ul></ul><ul><ul><li>Schedule contention periods </li></ul></ul><ul><ul><li>LPL (Low Power Listening) </li></ul></ul>
  49. 49. Techniques to Reduce Idle Listening <ul><li>TDMA requires cluster-based or centralized control. </li></ul><ul><li>Scheduling – ensures short listen period when transmitters and listeners can rendezvous and other periods where nodes sleep (turn off their radios). </li></ul><ul><li>LPL – nodes wake up briefly to check for channel activity without receiving data. </li></ul><ul><ul><li>If channel is idle, node goes back to sleep. </li></ul></ul><ul><ul><li>If channel is busy, node stays awake to receive data. </li></ul></ul><ul><ul><li>A long preamble (longer than poll period) is used to assure than preamble intersects with polls. </li></ul></ul>
  50. 50. #2 Dynamic Cluster Formation
  51. 51. Choosing Cluster Heads/ Forming Clusters <ul><li>Two-tier scheme: </li></ul><ul><li>A fixed number of cluster heads that communicate with BS (base station). </li></ul><ul><li>Nodes in cluster communicate with head (normally TDMA). </li></ul><ul><li>TDMA allows fixed schedule of slots for </li></ul><ul><li>sensor to send to cluster head and receive head transmissions. </li></ul>BS
  52. 52. BS Choosing Cluster Heads/ Forming Clusters <ul><li>Periodically select new cluster heads to minimize power consumption and maximize WSN lifetime. </li></ul><ul><li>More complex problem when size of cluster changes dynamically. </li></ul><ul><li>As time goes by, some sensor nodes die! </li></ul><ul><li>Not worried about coverage issues! </li></ul>X X X X X
  53. 53. Dynamic Cluster Formation <ul><li>TDMA cluster algorithms: </li></ul><ul><ul><li>LEACH, Bluetooth, … </li></ul></ul><ul><li>Rick Skowyra’s MS thesis: </li></ul><ul><li>‘ Energy Efficient Dynamic Reclustering Strategy for WSNs’ </li></ul><ul><ul><li>‘ Leach-like’ with a fitness function and periodic reclustering. </li></ul></ul><ul><ul><li>He hopes to design a distributed genetic algorithm to speed the recluster time. </li></ul></ul>
  54. 54. #3 Power-Aware MAC Protocols
  55. 55. Tiered WSN Architectures [ Stathopoulos]
  56. 56. Power Aware MAC Protocols <ul><li>1997 </li></ul><ul><li>1998 PAMAS </li></ul><ul><li>1999 </li></ul><ul><li>2000 </li></ul><ul><li>2001 SMAC </li></ul><ul><li>2002 LPL NPSM </li></ul><ul><li>2003 TMAC TRAMA TinyOS-MAC EMACs </li></ul><ul><li>2004 BMAC DMAC LMAC WiseMAC </li></ul><ul><li>2005 PMAC ZMAC SP </li></ul><ul><li>2006 SCP-MAC </li></ul><ul><li>2007 Crankshaft </li></ul>
  57. 57. Power Aware MAC Protocols <ul><li>Three approaches to saving power: </li></ul><ul><li>1. TDMA : TRAMA, EMACs, LMAC </li></ul><ul><li>Crankshaft </li></ul><ul><li>2. Schedule : PAMAS, SMAC , TMAC , DMAC, PMAC, SCP-MAC </li></ul><ul><li>3. Low Power Listening : LPL, BMAC, WiseMAC </li></ul><ul><li>Cross-Layering : SP, BSD </li></ul>
  58. 58. SMAC <ul><li>All nodes periodically listen, sleep and wakeup. Nodes listen and send during the active period and turn off their radios during the sleep period. </li></ul><ul><li>The beginning of the active period is a SYNC period used to accomplish periodic synchronization and remedy clock drift. </li></ul><ul><li>Following the SYNC period, data may be transferred for the remainder of the active period using RTS/CTS for unicast transmissions. </li></ul><ul><li>Long frames are fragmented and transmitted as a burst. </li></ul><ul><li>SMAC controls the duty cycle to tradeoff energy for delay. </li></ul><ul><li>However, as density of WSN grows, SMAC incurs additional overhead in maintaining neighbors’ schedules. </li></ul>
  59. 59. SMAC
  60. 60. TMAC <ul><li>TMAC employs an adaptive duty cycle by using a very short listening window at the beginning of each active period. </li></ul><ul><li>After the SYNC portion of the active period, RTS/CTS is used in listening window. If no activity occurs, the node goes to sleep. </li></ul><ul><li>TMAC saves power at the cost of reduced throughput and additional delay. </li></ul>
  61. 61. TMAC
  62. 62. WiseMAC <ul><li>Algorithm focused on downlink protocol </li></ul><ul><li>for infrastructure WSNs: </li></ul><ul><ul><li>Access Point (AP) is assumed to have wired link to Internet and not battery-powered. </li></ul></ul><ul><li>Based on preamble sampling. </li></ul><ul><li>WiseMac regularly samples (via listening) for a short duration during preamble. </li></ul><ul><ul><li>All sensor nodes sample with same constant period T W . </li></ul></ul>
  63. 63. WiseMAC <ul><li>Normally, wake-up preamble needs to be of size T W . </li></ul><ul><ul><li>This implies low power use when WSN is idle. </li></ul></ul><ul><ul><li>But this yields large power consumption overhead for reception. </li></ul></ul><ul><li>WiseMAC AP learns and keeps the sampling schedule of all sensors in a up-to-date table. </li></ul><ul><li>Sensors’ ACKs provide info for the table. </li></ul><ul><li>WiseMAC then minimizes the preamble duration, T P . </li></ul><ul><li>Needs to deal with clock drift to get this right. </li></ul>
  64. 64. WiseMAC
  65. 65. Zigbee MAC <ul><li>802.11 MAC PSM (Power Save Mode) uses beacon frames to coordinate and periodic wake-up by sensor nodes. </li></ul><ul><li>Mike Putnam’s thesis: </li></ul><ul><li>‘A Beaconless Protocol for Improving Energy Efficiency in Wireless Sensor Networks’ </li></ul>
  66. 66. WiseMAC
  67. 67. Scheduled Channel Polling (SCP-MAC) <ul><li>With channel polling (LPL scheme), receiver efficiency is gained through cost to sender. </li></ul><ul><li>LPLs are very sensitive to tuning for neighborhood size and traffic rate. </li></ul><ul><li>By synchronizing channel polling times of all neighbors, long preambles are eliminated and ultra-low duty cycles (below the LPL 1-2% limits) are possible. </li></ul>
  68. 68. Scheduled Channel Polling (SCP-MAC) <ul><li>The issue is knowing my neighbors’ schedule information. </li></ul><ul><li>SCP piggybacks schedule info on data packets when possible or a node broadcasts its schedule in a SYNC packet in synch period (as in SMAC) </li></ul><ul><li>Knowing schedules  short wakeup tone. </li></ul><ul><li>Optimal synchronization reduces overhearing. </li></ul>
  69. 69. SCP-MAC
  70. 70. Crankshaft <ul><li>Designed specifically for dense WSNs. </li></ul><ul><li>Employs channel polling mechanism similar to SCP-MAC. </li></ul><ul><li>Employs synchronization, framing and slotting mechanisms similar to TDMA-based LMAC. </li></ul><ul><li>Unlike LMAC, Crankshaft schedules receivers rather than senders. </li></ul><ul><li>Basic principle: nodes are only awake to receive messages at fixed offsets from the start of a frame. </li></ul><ul><li>The hope is to develop modified version of Crankshaft. </li></ul><ul><li>I need students interested in exploring this idea!! </li></ul>
  71. 71. Crankshaft
  72. 72. Crankshaft Simulations <ul><li>Focused on two traffic types that are common in WSNs: </li></ul><ul><li>Convergecast – monitoring traffic </li></ul><ul><ul><li>All sensor nodes periodically send data to a sink node (either AP or sensor cluster head) </li></ul></ul><ul><li>Broadcast floods – packets sent in the other direction to either send routing update or to distribute queries over the WSN. </li></ul>
  73. 73. Energy Conservation Results
  74. 74. Latency
  75. 75. Questions? <ul><li>Thank You! </li></ul>Go Tribe!!!