Media Access Control Chapter 10

Transcription

1 Raing Media Access Conrol Chaper 10 Area mauriy Firs seps Tex book Pracical imporance No apps Mission criical Theoreical imporance No really Mus have Ad Hoc and Sensor Neworks Roger Waenhofer 10/1 Ad Hoc and Sensor Neworks Roger Waenhofer 10/2 Overview Moivaion Moivaion Classificaion Case sudy: Oher MAC layer echniques The broadcas problem Can we apply media access mehods from fixed neworks? Example CSMA/CD Carrier Sense Muliple Access wih Collision Deecion send as soon as he medium is free, lisen ino he medium if a collision occurs (original mehod in IEEE 802.3) Problems in wireless neworks signal srengh decreases quickly wih disance senders apply CS and CD, bu he collisions happen a receivers Energy efficiency: having he radio urned on coss almos as much energy as ransmiing, so o seriously save energy one needs o urn radio off! Ad Hoc and Sensor Neworks Roger Waenhofer 10/3 Ad Hoc and Sensor Neworks Roger Waenhofer 10/4

2 Moivaion Hidden erminal problem Moivaion Exposed erminal problem A sends o B, C canno receive A B sends o A, C wans o send o D C has o wai, CS signals a medium in use collision a B, A canno receive he collision (CD fails) since A is ouside he radio range of C waiing is no necessary A B C A B C D Ad Hoc and Sensor Neworks Roger Waenhofer 10/5 Ad Hoc and Sensor Neworks Roger Waenhofer 10/6 Moivaion - near and far erminals MAC Alphabe Soup Terminals A and B send, C receives C canno receive A A B C This is also a severe problem for CDMA neworks precise power conrol -MAC Aloha AI-LMAC B-MAC BiMAC BMA CMAC Crankshaf CSMA-MPS CSMA/ARC DMAC E2-MAC EMACs f-mac FLAMA Funneling-MAC G-MAC HMAC LMAC LPL MMAC nanomac O-MAC PACT PCM PEDAMACS PicoRadio PMAC PMAC Preamble sampling Q-MAC Q- QMAC RATE EST RL-MAC RMAC S-MAC S-MAC/AL SMACS SCP-MAC SEESAW Sif SS-TDMA STEM T-MAC TA-MAC TRAMA U-MAC WiseMAC X-MAC Z-MAC [TU Delf] Ad Hoc and Sensor Neworks Roger Waenhofer 10/7 Ad Hoc and Sensor Neworks Roger Waenhofer 10/8

3 Tradiional MAC proocol classificaion Conenion Proocols Transmi when you feel like ransmiing Rery if collision, ry o minimize collisions, addiional reservaion modes Problem: Receiver mus be awake as well random slos STEM Preamble sampling LPL S-MAC WiseMAC T-MAC RATE EST CSMA-MPS B-MAC DMAC SCP-MAC X-MAC Scheduling Proocols - Disribued, adapive soluions are difficul frames PEDAMACS TRAMA LMAC AI-LMAC FLAMA Oher proocols Hybrid soluions, e.g. conenion wih reservaion scheduling - hybrid Crankshaf Z-MAC PMAC Ad Hoc and Sensor Neworks Roger Waenhofer 10/9 Ad Hoc and Sensor Neworks Roger Waenhofer 10/10 Access mehods SDMA/FDMA/TDMA Comparison SDMA/TDMA/FDMA/CDMA SDMA (Space Division Muliple Access) segmen space ino secors, use direced anennas Use cells o reuse frequencies FDMA (Frequency Division Muliple Access) assign a cerain frequency o a ransmission channel permanen (radio broadcas), slow hopping (GSM), fas hopping (FHSS, Frequency Hopping Spread Specrum) TDMA (Time Division Muliple Access) assign a fixed sending frequency for a cerain amoun of ime CDMA (Code Division Muliple Access) Combinaions! Ad Hoc and Sensor Neworks Roger Waenhofer 10/11 Approach SDMA TDMA FDMA CDMA Idea segmen space ino spread he specrum cells/secors using orhogonal codes Terminals Signal separaion Advanages Disadvanages Commen only one erminal can be acive in one cell/one secor cell srucure, direced anennas very simple, increases capaciy per km² inflexible, anennas ypically fixed only in combinaion wih TDMA, FDMA or CDMA useful segmen sending ime ino disjoin ime-slos, demand driven or fixed paerns all erminals are acive for shor periods of ime on he same frequency synchronizaion in he ime domain esablished, fully digial, flexible guard space needed (mulipah propagaion), synchronizaion difficul sandard in fixed neworks, ogeher wih FDMA/SDMA used in many mobile neworks segmen he frequency band ino disjoin sub-bands every erminal has is own frequency, uninerruped filering in he frequency domain simple, esablished, robus inflexible, frequencies are a scarce resource ypically combined wih TDMA (frequency hopping paerns) and SDMA (frequency reuse) all erminals can be acive a he same place a he same momen, uninerruped code plus special receivers flexible, less frequency planning needed, sof handover complex receivers, needs more complicaed power conrol for senders sill faces some problems, higher complexiy, lowered expecaions; will be inegraed wih TDMA/FDMA [J.Schiller]

4 FDD/FDMA - general scheme, example 900Mhz TDD/TDMA - general scheme, example DECT 960 MHz f MHz 915 MHz MHz 200 khz 417 µs downlink uplink MHz 1 Ad Hoc and Sensor Neworks Roger Waenhofer 10/13 Ad Hoc and Sensor Neworks Roger Waenhofer 10/14 TDMA Moivaion TDMA Sloed Aloha We have a sysem wih n n1) and one shared channel The channel is a perfec broadcas channel, ha is, if any single saion ransmis alone, he ransmission can be received by every oher saion. There is no hidden or exposed erminal problem. If wo or more ransmi a he same ime, he ransmission is garbled. Round robin algorihm: saion k sends afer saion k1 (mod n) There is a maximum message size m ha can be ransmied How efficien is round robin? Wha if a saion breaks or leaves? All deerminisic TDMA proocols have hese (or worse) problems We assume ha he saions are perfecly synchronous In each ime slo each saion ransmis wih probabiliy p. P1 Pr[Saion 1 succeeds] p(1 p) P Pr[any Saion succeeds] np1 dp dp 1 n 1 1 hen, P (1 ) n e In sloed aloha, a saion can ransmi successfully wih probabiliy a leas 1/e. How quickly can an applicaion send packes o he radio ransmission uni? This quesion is sudied in queuing heory. n 1! maximize P: n(1 n 2 p) (1 pn) 0 pn 1 Ad Hoc and Sensor Neworks Roger Waenhofer 10/15

5 Queuing Theory he basic basics in a nushell Sloed Aloha vs. Round Robin Simples M/M/1 queuing model (M=Markov): Poisson arrival rae, exponenial service ime wih mean 1/ Sloed aloha uses no every slo of he channel; he round robin proocol is beer. + Wha happens in round robin when a new saion joins? Wha abou more han one new saion? Sloed aloha is more flexible. In our ime slo model, his means ha he probabiliy ha a new packe is received by he buffer is ; he probabiliy ha sending succeeds is, for any ime slo. To keep he queue bounded we need = / < 1. In he equilibrium, he expeced number of packes in he sysem is N = /(1 ), he average ime in he sysem is T = N/. Example: If he acual number of saions is wice as high as expeced, here is sill a successful ransmission wih probabiliy 30%. If i is only half, 27% of he slos are used successfully. Adapive sloed aloha Adapive sloed aloha 2 Idea: Change he access probabiliy wih he number of saions How can we esimae he curren number of saions in he sysem? Assume ha saions can disinguish wheher 0, 1, or more han 1 saions send in a ime slo. Idea: If you see ha nobody sends, increase p. If you see ha more han one sends, decrease p. Model: Number of saions ha wan o ransmi: n. Esimae of n: n Transmission probabiliy: p = 1/ n Arrival rae (new saions ha wan o ransmi): ; noe ha < 1/e. n n P 1 1 We have o show ha he sysem sabilizes. Skech: P 2 P1 P0 P0 P2 n n 1, if success or idle n n 1, if collision e 2 n Ad Hoc and Sensor Neworks Roger Waenhofer 10/19 Ad Hoc and Sensor Neworks Roger Waenhofer 10/20

6 Adapive sloed aloha Q&A Backoff Proocols Q: Wha if we do no know, or is changing? A: Use = 1/e, and he algorihm sill works Q: How do newly arriving saions know n? A: We send n wih each ransmission; new saions do no send before successfully receiving he firs ransmission. Q: Wha if saions are no synchronized? A: Aloha (non-sloed) is wice as bad Q: Can saions really lisen o all ime slos (save energy by urning off)? Can saions really disinguish beween 0, 1, and more han 1 sender? A: Maybe. One can use sysems ha only rely on Ad Hoc and Sensor Neworks Roger Waenhofer 10/21 Backoff proocols rely on acknowledgemens only. Binary exponenial backoff, for example, works as follows: If a packe has collided k imes, we se p = 2 -k Or alernaively: wai from random number of slos in [1..2 k ] I has been shown ha binary exponenial backoff is no sable for any > 0 (if here are infiniely many poenial saions) [Proof skech: wih very small bu posiive probabiliy you go o a bad siuaion wih many waiing saions, and from here you ge even worse wih a poenial funcion argumen sadly he proof is oo inricae o be shown in his course ] Ineresingly when here are only finie saions, binary exponenial backoff becomes unsable wih > 0.568; Polynomial backoff however, remains sable for any < 1. Ad Hoc and Sensor Neworks Roger Waenhofer 10/22 Demand Assigned Muliple Access (DAMA) DAMA: Explici Reservaion Channel efficiency only 36% for Sloed Aloha, and even worse for Aloha or backoff proocols. Pracical sysems herefore use reservaion whenever possible. Bu: Every scalable sysem needs an Aloha syle componen. Reservaion: a sender reserves a fuure ime-slo sending wihin his reserved ime-slo is possible wihou collision reservaion also causes higher delays ypical scheme for saellie sysems Examples for reservaion algorihms: Explici Reservaion (Reservaion-ALOHA) Implici Reservaion (PRMA) Reservaion-TDMA Muliple Access wih Collision Avoidance (MACA) Aloha mode for reservaion: compeiion for small reservaion slos, collisions possible reserved mode for daa ransmission wihin successful reserved slos (no collisions possible) i is imporan for all saions o keep he reservaion lis consisen a any poin in ime and, herefore, all saions have o synchronize from ime o ime collisions Aloha Aloha Aloha Aloha reserved reserved reserved reserved Ad Hoc and Sensor Neworks Roger Waenhofer 10/23 Ad Hoc and Sensor Neworks Roger Waenhofer 10/24

7 DAMA: Packe Reservaion MA (PRMA) DAMA: Reservaion TDMA a cerain number of slos form a frame, frames are repeaed saions compee for empy slos according o he sloed aloha principle once a saion reserves a slo successfully, his slo is auomaically assigned o his saion in all following frames as long as he saion has daa o send compeiion for his slos sars again as soon as he slo was empy in he las frame reservaion ime-slo ACDABA-F frame 1 A C D A B A F ACDABA-F frame 2 A C A B A AC-ABAFframe collision a 3 A B A F A---BAFD reservaion frame 4 A B A F D aemps ACEEBAFD frame 5 A C E E B A F D every frame consiss of n mini-slos and x daa-slos every saion has is own mini-slo and can reserve up o k daaslos using his mini-slo (i.e. x = nk). oher saions can send daa in unused daa-slos according o a round-robin sending scheme (bes-effor raffic) N mini-slos reservaions for daa-slos Nk daa-slos n=6, k=2 oher saions can use free daa-slos based on a round-robin scheme Ad Hoc and Sensor Neworks Roger Waenhofer 10/25 Ad Hoc and Sensor Neworks Roger Waenhofer 10/26 Muliple Access wih Collision Avoidance (MACA) MACA examples Use shor signaling packes for collision avoidance Reques (or ready) o send RTS: a sender requess he righ o send from a receiver wih a shor RTS packe before i sends a daa packe Clear o send CTS: he receiver grans he righ o send as soon as i is ready o receive Signaling packes conain sender address receiver address packe size Example: Wireless LAN (802.11) as DFWMAC MACA avoids he problem of hidden erminals A and C wan o send o B A sends RTS firs RTS C wais afer receiving CTS from B CTS CTS A B C MACA avoids he problem of exposed erminals B wans o send o A, and C o D now C does no have RTS o wai as C canno receive CTS from A RTS CTS A B C D Ad Hoc and Sensor Neworks Roger Waenhofer 10/27 Ad Hoc and Sensor Neworks Roger Waenhofer 10/28

8 MACA varian: DFWMAC in IEEE Polling mechanisms ACK sender idle RxBusy ime-ou or NAK RTS wai for ACK RTS wai for he righ o send CTS daa ime-ou RTS ime-ou or corrup daa NAK receiver idle daa ACK wai for daa RTS CTS (a.k.a. base saion) can poll all oher erminals according o a cerain scheme Use a scheme known from fixed neworks The base saion chooses one address for polling from he lis of all saions The base saion acknowledges correc packes and coninues polling he nex erminal The cycle sars again afer polling all erminals of he lis An aloha-syle componen is needed o allow new saions join ACK: posiive acknowledgemen NAK: negaive acknowledgemen RxBusy: receiver busy RTS RxBusy Ad Hoc and Sensor Neworks Roger Waenhofer 10/29 Ad Hoc and Sensor Neworks Roger Waenhofer 10/30 Inhibi Sense Muliple Access (ISMA) Design goals Global, seamless operaion he base saion signals on he downlink (base saion o erminals) wheher he medium is free erminals mus no send if he medium is busy erminals can access he medium as soon as he busy one sops he base saion signals collisions and successful ransmissions via he busy one and acknowledgemens, respecively (media access is no coordinaed wihin his approach) Example: for CDPD (USA, inegraed ino AMPS) Low power consumpion for baery use No special permissions or licenses required Robus ransmission echnology Simplified sponaneous cooperaion a meeings Easy o use for everyone, simple managemen Ineroperable wih wired neworks Securiy (no one should be able o read my daa), privacy (no one should be able o collec user profiles), safey (low radiaion) Transparency concerning applicaions and higher layer proocols, bu also locaion awareness if necessary Ad Hoc and Sensor Neworks Roger Waenhofer 10/31 Ad Hoc and Sensor Neworks Roger Waenhofer 10/32

9 Characerisics Infrasrucure vs. ad hoc mode + Very flexible (economical o scale) + Ad-hoc neworks wihou planning possible + (Almos) no wiring difficulies (e.g. hisoric buildings, firewalls) + More robus agains disasers or users pulling a plug Infrasrucure nework AP AP wired nework AP: Access Poin AP Low bandwidh compared o wired neworks (10 vs. 100[0] Mbi/s) Many proprieary soluions, especially for higher bi-raes, sandards ake heir ime Producs have o follow many naional resricions if working wireless, i akes a long ime o esablish global soluions (IMT-2000) Securiy Economy Ad-hoc nework Ad Hoc and Sensor Neworks Roger Waenhofer 10/ Proocol archiecure The lower layers in deail mobile erminal applicaion TCP server access poin fixed erminal infrasrucure nework applicaion TCP PMD (Physical Medium Dependen) modulaion, coding PLCP (Physical Layer Convergence Proocol) clear channel assessmen signal (carrier sense) PHY Managemen channel selecion, PHY-MIB Saion Managemen coordinaion of all managemen funcions MAC access mechanisms fragmenaion encrypion MAC Managemen Synchronizaion roaming power managemen MIB (managemen informaion base) IP LLC MAC PHY LLC MAC MAC PHY PHY IP LLC MAC PHY PHY DLC LLC MAC PLCP PMD MAC Managemen PHY Managemen Saion Managemen Ad Hoc and Sensor Neworks Roger Waenhofer 10/35 Ad Hoc and Sensor Neworks Roger Waenhofer 10/36

10 MAC layer: DFWMAC MAC layer Traffic services Asynchronous Daa Service (mandaory) - suppor of broadcas and mulicas Time-Bounded Service (opional) implemened using PCF (Poin Coordinaion Funcion) Access mehods DFWMAC-DCF CSMA/CA (mandaory) collision avoidance via binary exponenial back-off mechanism minimum disance beween consecuive packes ACK packe for acknowledgemens (no used for broadcass) DFWMAC-DCF w/ RTS/CTS (opional) avoids hidden erminal problem DFWMAC-PCF (opional) access poin polls erminals according o a lis defined hrough differen iner frame spaces no guaraneed, hard prioriies (Shor Iner Frame Spacing) highes prioriy, for ACK, CTS, polling response PIFS (PCF IFS) medium prioriy, for ime-bounded service using PCF (DCF, Disribued Coordinaion Funcion IFS) lowes prioriy, for asynchronous daa service medium busy PIFS direc access if medium is free conenion nex frame Ad Hoc and Sensor Neworks Roger Waenhofer 10/37 Ad Hoc and Sensor Neworks Roger Waenhofer 10/38 CSMA/CA Compeing saions - simple example conenion window (randomized back-off mechanism) saion 1 bo e bo r bo e bo r bo e busy medium busy direc access if medium is free slo ime nex frame saion 2 saion 3 busy bo e busy saion ready o send sars sensing he medium (Carrier Sense based on CCA, Clear Channel Assessmen) if he medium is free for he duraion of an Iner-Frame Space (IFS), he saion can sar sending (IFS depends on service ype) if he medium is busy, he saion has o wai for a free IFS, hen he saion mus addiionally wai a random back-off ime (collision avoidance, muliple of slo-ime) if anoher saion occupies he medium during he back-off ime of he saion, he back-off imer sops (fairness) saion 4 saion 5 busy bo e busy bo e bo r bo e bo r bo e busy bo e bo r medium no idle (frame, ack ec.) bo e elapsed backoff ime packe arrival a MAC bo r residual backoff ime backoff Ad Hoc and Sensor Neworks Roger Waenhofer 10/39 Ad Hoc and Sensor Neworks Roger Waenhofer 10/40

11 CSMA/CA 2 Sending unicas packes saion has o wai for before sending daa receivers acknowledge a once (afer waiing for ) if he packe was received correcly (CRC) auomaic reransmission of daa packes in case of ransmission errors sender receiver oher saions daa waiing ime ACK conenion daa DFWMAC saion can send RTS wih reservaion parameer afer waiing for (reservaion deermines amoun of ime he daa packe needs he medium) acknowledgemen via CTS afer by receiver (if ready o receive) sender can now send daa a once, acknowledgemen via ACK oher saions sore medium reservaions disribued via RTS and CTS RTS daa sender CTS ACK receiver oher saions NAV (RTS) NAV (CTS) defer access conenion daa Ad Hoc and Sensor Neworks Roger Waenhofer 10/41 Ad Hoc and Sensor Neworks Roger Waenhofer 10/42 Fragmenaion Fragmenaion: Wha fragmen size is opimal? sender receiver oher saions If packe ges oo long ransmission error probabiliy grows A simple back of he envelope calculaion deermines he opimal fragmen size RTS CTS frag 1 NAV (RTS) NAV (CTS) ACK 1 frag 2 ACK2 NAV (frag 1 ) NAV (ACK 1 ) conenion daa Toal daa size: D bis Overhead per packe (header): h bis We wan f fragmens, hen each fragmen has k = D/f + h daa + header bis Channel has bi error probabiliy q = 1-p Probabiliy o ransmi a packe of k bis correcly: P := p k Expeced number of ransmissions unil packe is success: 1/P Expeced oal cos for all D bis: f (k/p+a) Goal: Find a k > h ha minimizes he expeced cos Ad Hoc and Sensor Neworks Roger Waenhofer 10/43 Ad Hoc and Sensor Neworks Roger Waenhofer 10/44

12 Fragmenaion: Wha fragmen size is opimal? DFWMAC-PCF For he sake of a simplified analysis we assume a = O(h) An access poin can poll saions If we furher assume ha a header can be ransmied wih consan probabiliy c, ha is, p h = c. 0 1 SuperFrame We choose k = 2h; Then clearly D = f h, and herefore expeced cos medium busy poin coordinaor PIFS D 1 D 2 wireless saions U 1 U 2 If already a header canno be ransmied wih high enough probabiliy, hen you migh keep he message very small, for example k = h + 1/q NAV NAV Ad Hoc and Sensor Neworks Roger Waenhofer 10/45 Ad Hoc and Sensor Neworks Roger Waenhofer 10/46 DFWMAC-PCF 2 Frame forma Duraion ID Address 1 Address 2 Address 3 Sequence Conrol Frame Conrol 2 4 Address Daa byes CRC poin coordinaor wireless saions NAV D 3 PIFS D 4 NAV conenion free period U 4 CFend conenion period Bye 1: version, ype, subype Bye 2: wo DS-bis, fragm., rery, power man., more daa, WEP, order Type conrol frame, managemen frame, daa frame Sequence conrol imporan agains duplicaed frames due o los ACKs Addresses receiver, ransmier (physical), BSS idenifier, sender (logical) Miscellaneous sending ime, checksum, frame conrol, daa Ad Hoc and Sensor Neworks Roger Waenhofer 10/47 Ad Hoc and Sensor Neworks Roger Waenhofer 10/48

13 MAC address forma Special Frames: ACK, RTS, CTS scenario o DS from address 1 address 2 address 3 address 4 DS ad-hoc nework 0 0 DA SA BSSID - infrasrucure 0 1 DA BSSID SA - nework, from AP infrasrucure 1 0 BSSID SA DA - nework, o AP infrasrucure nework, wihin DS 1 1 RA TA DA SA DS: Disribuion Sysem AP: Access Poin DA: Desinaion Address SA: Source Address BSSID: Basic Service Se Idenifier RA: Receiver Address TA: Transmier Address Acknowledgemen Reques To Send Clear To Send ACK RTS CTS byes Frame Duraion Receiver CRC Conrol Address byes Frame Duraion Receiver Transmier CRC Conrol Address Address byes Frame Duraion Receiver CRC Conrol Address Ad Hoc and Sensor Neworks Roger Waenhofer 10/49 Ad Hoc and Sensor Neworks Roger Waenhofer 10/50 MAC managemen Synchronizaion Synchronizaion ry o find a LAN, ry o say wihin a LAN imer ec. Power managemen sleep-mode wihou missing a message periodic sleep, frame buffering, raffic measuremens Associaion/Reassociaion inegraion ino a LAN roaming, i.e. change neworks by changing access poins scanning, i.e. acive search for a nework MIB - Managemen Informaion Base managing, read, wrie In an infrasrucure nework, he access poin can send a beacon access poin medium beacon inerval B busy B B B busy busy busy value of imesamp B beacon frame Ad Hoc and Sensor Neworks Roger Waenhofer 10/51 Ad Hoc and Sensor Neworks Roger Waenhofer 10/52

14 Synchronizaion Power managemen In an ad-hoc nework, he beacon has o be sen by any saion beacon inerval saion 1 B 1 B 1 saion 2 B 2 B 2 medium busy busy busy busy value of he imesamp B beacon frame backoff delay Idea: if no needed urn off he ransceiver Saes of a saion: sleep and awake Timing Synchronizaion Funcion (TSF) saions wake up a he same ime Infrasrucure Traffic Indicaion Map (TIM) lis of unicas receivers ransmied by AP Delivery Traffic Indicaion Map (DTIM) lis of broadcas/mulicas receivers ransmied by AP Ad-hoc Ad-hoc Traffic Indicaion Map (ATIM) announcemen of receivers by saions buffering frames more complicaed - no cenral AP collision of ATIMs possible (scalabiliy?) Ad Hoc and Sensor Neworks Roger Waenhofer 10/53 Ad Hoc and Sensor Neworks Roger Waenhofer 10/54 Power saving wih wake-up paerns (infrasrucure) Power saving wih wake-up paerns (ad-hoc) TIM inerval DTIM inerval ATIM window beacon inerval access poin medium D B busy T T d D busy busy busy B B saion 1 A D B 1 1 B saion 2 B 2 a d 2 saion T TIM D DTIM p awake d B beacon frame random delay A ransmi ATIM D ransmi daa B broadcas/mulicas p PS poll d daa ransmission o/from he saion awake a acknowledge ATIM d acknowledge daa Ad Hoc and Sensor Neworks Roger Waenhofer 10/55 Ad Hoc and Sensor Neworks Roger Waenhofer 10/56

15 WLAN: IEEE b WLAN: IEEE b Daa rae 1, 2, 5.5, 11 Mbi/s, depending on SNR User daa rae max. approx. 6 Mbi/s Transmission range 300m oudoor, 30m indoor Max. daa rae <10m indoor Frequency Free 2.4 GHz ISM-band Securiy Limied, WEP insecure, SSID Cos Low Availabiliy Declining Connecion se-up ime Connecionless/always on Qualiy of Service Typically bes effor, no guaranees unless polling is used, limied suppor in producs Manageabiliy Limied (no auomaed key disribuion, sym. encrypion) + Advanages: many insalled sysems, lo of experience, available worldwide, free ISM-band, many vendors, inegraed in lapops, simple sysem Disadvanages: heavy inerference on ISM-band, no service guaranees, slow relaive speed only Ad Hoc and Sensor Neworks Roger Waenhofer 10/57 Ad Hoc and Sensor Neworks Roger Waenhofer 10/58 IEEE b PHY frame formas Channel selecion (non-overlapping) Long PLCP PPDU forma variable bis synchronizaion SFD signal service lengh HEC payload Europe (ETSI) channel 1 channel 7 channel 13 PLCP preamble PLCP header 192 µs a 1 Mbi/s DBPSK 1, 2, 5.5 or 11 Mbi/s Shor PLCP PPDU forma (opional) variable bis shor synch. SFD signal service lengh HEC payload MHz [MHz] US (FCC)/Canada (IC) channel 1 channel 6 channel 11 PLCP preamble (1 Mbi/s, DBPSK) PLCP header (2 Mbi/s, DQPSK) MHz [MHz] 96 µs 2, 5.5 or 11 Mbi/s Ad Hoc and Sensor Neworks Roger Waenhofer 10/60

16 WLAN: IEEE a WLAN: IEEE a Daa rae 6, 9, 12, 18, 24, 36, 48, 54 Mbi/s, depending on SNR User hroughpu (1500 bye packes): 5.3 (6), 18 (24), 24 (36), 32 (54) 6, 12, 24 Mbi/s mandaory Transmission range 100m oudoor, 10m indoor: e.g., 54 Mbi/s up o 5 m, 48 up o 12 m, 36 up o 25 m, 24 up o 30m, 18 up o 40 m, 12 up o 60 m Frequency Free , , GHz ISM-band Securiy Limied, WEP insecure, SSID Cos $50 adaper, $100 base saion, dropping Availabiliy Some producs, some vendors No really deployed in Europe (regulaions!) Connecion se-up ime Connecionless/always on Qualiy of Service Typically bes effor, no guaranees (same as all producs) Manageabiliy Limied (no auomaed key disribuion, sym. Encrypion) + Advanages: fis ino 802.x sandards, free ISM-band, available, simple sysem, uses less crowded 5 GHz band Disadvanages: sronger shading due o higher frequency, no QoS Ad Hoc and Sensor Neworks Roger Waenhofer 10/61 Ad Hoc and Sensor Neworks Roger Waenhofer 10/62 Quiz: Which sandard? Pimp my MAC proocol Some general echniques o improve MAC proocols. In he following we presen a few ideas, solen from a few known proocols such as S-MAC T-MAC B-MAC Dozer WiseMAC RFID Many of he hundreds of MAC proocols ha were proposed have Ad Hoc and Sensor Neworks Roger Waenhofer 10/64

17 Energy vs. Delay (e.g. S-MAC) Adapive periods (e.g. T-MAC) Compue a conneced dominaing se (CDS) Nodes in he CDS choose and announce an awake schedule, and synchronize o an awake schedule of heir neighbor CDS nodes. The oher nodes synchronize o he awake schedule of heir dominaor (if hey have more han one dominaor, an arbirary dominaor can be chosen) Then use acive periods o iniiae communicaion (hrough RTS/CTS), and poenially communicae during sleep period More raffic higher duy cycles Conrol problems: Assume linked lis nework A B C. Assume ha AB and BC have very low duy cycle. Now A needs o send daa o C, hus increasing duy cycle of AB. Then A migh send B a lo of daa before B has a chance o increase duy cycle of BC. This is even worse when nework is more complicaed, as several Problems: Large overhead because of connecing domains, may poenially T-MAC proposal: When receiving he nex RTS of A, node B immediaely answers wih an RTS iself o signal A ha is buffer needs o be empied firs. Ad Hoc and Sensor Neworks Roger Waenhofer 10/66 Long preambles (e.g. B-MAC) Synchronize o receiver (e.g. Dozer) As idle lisening coss abou as much energy as ransmiing, we migh o ry o reduce idle lisening. Nodes sill have heir sleeping cycles as before. Maybe sender knows wake-up paern of receiver. Then i can simply sar sending a he righ ime, almos wihou preamble If sender wans o ransmi message, i aaches a preamble of he size of a sleep period o make sure ha he receiver wakes up during preamble. Problem: Receiver needs o wai for whole preamble o finish, even if i wakes up early in he preamble. Soluion 1: Send wake-up packes insead of preamble, wake-up packes ell when daa is saring so ha receiver can go back o sleep as soon as i received one wake-up packe. Soluion 2: Jus send daa several imes such ha receiver can une in a any ime and ge ail of daa firs, hen head. Problem: How o know he wake-up paern? Dozer soluion: Inegrae i wih higher-layer proocol, coninuously exchange informaion, resric number of neighbors (or align many of hem o reduce informaion) Oher soluions, e.g. WiseMAC: Firs send long preamble; receiver hen ACKs packe, and encodes is wake-up schedule in ACK for fuure use. Ad Hoc and Sensor Neworks Roger Waenhofer 10/68

18 Two radios The bes MAC proocol?!? Nodes have wo radios, a regular (high-power) radio o exchange daa, and a low-power radio o sense ransmissions. Energy-efficiency vs. hroughpu vs. delay Wors-case guaranees vs. bes-effor Uopia: Maybe i is even possible o send a high-power pulse over some disance which can wake up receiver (e.g. RFID) Problem: Sender mus be excepionally high-power; may lead o very asymmeric design such as in RFID where he reader is orders of magniudes larger han a passive RFID chip. This may no be feasible in ad hoc or sensor neworks. Cenralized/offline vs. disribued/online Random opology vs. wors-case graph vs. wors- Communicaion paern Nework layer: local broadcas vs. all-o-all vs. broadcas/echo So, clearly, here canno be a bes MAC proocol! Ad Hoc and Sensor Neworks Roger Waenhofer 10/69 Ad Hoc and Sensor Neworks Roger Waenhofer 10/70 Model Deerminisic algorihms (anonymous) Nework is an undireced graph Nodes do no know opology of graph Synchronous rounds Nodes can eiher ransmi or receive (no boh, no sleep) Message is received if exacly one neighbor ransmis No collision deecion: Tha is, a node canno disinguish wheher 0 or 2 or more neighbors ransmi If nodes are anonymous (hey have no node IDs), hen one canno solve he broadcas problem For he graph on he righ nodes 1 and 2 always have he same inpu, and hence always do he same hing, and hence node 3 can never receive he message We sudy broadcasing problem sor of MAC layer, no quie Iniially only source has message finally every node has message So he nodes need IDs. How long does his ake?!? Ad Hoc and Sensor Neworks Roger Waenhofer 10/71 Ad Hoc and Sensor Neworks Roger Waenhofer 10/72

19 Deerminisic algorihms (no anonymous) How o choose golden nodes? Consider he following nework family: n+2 nodes, 3 layers Firs layer: source node (green) Las layer: final node (red) Middle layer: all oher nodes (n) Source conneced o all nodes in middle layer Middle layer consiss of golden and blue nodes Golden nodes connec o red node, Clearly, in one single sep all middle nodes Task: Given deerminisic algorihm, e.g. n-1 ses M i of nodes Choose golden and blue nodes, such ha no se M i conains a single golden node. Consrucion of golden se We sar wih golden se S being all middle nodes While M i such ha M i \ {M i Any deerminisic algorihm needs a leas n rounds In every ieraion a golden node inersecing wih M i is removed from S; se M i does no have o be considered again aferwards. Thus afer n-1 rounds we sill have one golden node lef and all ses M i do no conain exacly one golden node. Ad Hoc and Sensor Neworks Roger Waenhofer 10/73 Improvemen hrough randomizaion? Randomized proocol for arbirary graphs If in each sep a random node is chosen ha would no help much, because a single golden node sill is only found afer abou n/2 Randomly selec n i/k nodes, for ik-1 also chosen randomly. Assume ha here are abou n s/k golden nodes. Then he chance o randomly selec a single golden node is abou Pr(success) = n i=k n s=k 1 (1 n s=k 1 ) ni=k 1 Posiions for golden node Probabiliy for golden node All ohers are no golden If we are lucky and k = i+s his simplifies o µ Pr(success) ¼ n i=k ¼ 1=e n i=k If we choose k = log n and do he compuaion correcly, we have polylogarihmic rials o find a single golden node. O(D log n + log 2 n) N: upper bound on node number : upper bound on max degree ²: Failure probabiliy, hink ² = 1/n N,,² are globally known D: diameer of graph Algorihm runs in synchronous phases, nodes always ransmi slo number in every message Ad Hoc and Sensor Neworks Roger Waenhofer 10/75 Ad Hoc and Sensor Neworks Roger Waenhofer 10/76

20 Proof overview Proof of he firs sep During one execuion of Decay a node can successfully receive a message wih probabiliy p 1/(2e) During one execuion of Decay a node can successfully receive a message wih probabiliy p 1/(2e): Ieraing Decay c log n imes we ge a very high success probabiliy of p 1/n c A he sar of Decay d nodes ry o reach our arge node. Abou half of hem fail each sep. More formally, afer sep i, s.. 2 i-1 < d 2 i 1 2d <Pr(node ransmis in sep i-1) = i d Since a single execuion of Decay akes log n seps, all nodes of he nex level receive he message afer c log 2 n seps (again, wih very high probabiliy). And hence Pr(exacly 1 node ransmis in sep i-1) d 1 2d (1 1 d )d e (Sep i does exis since k = 2 log.) Having D layers a oal of O(D log 2 n) rounds is sufficien (wih high probabiliy). Ad Hoc and Sensor Neworks Roger Waenhofer 10/78 Fases algorihm Open Problem Known lower bound (D log(n/d) + log 2 n) Fases algorihm maches lower bound. Skech of one case: = loglog n Alhough he MAC alphabe soup is consanly growing, he radeoffs delay, hroughpu, energy-efficiency, localiy, dynamics, problems are abou lower bounds: Node ha received message from source We are looking for a non-rivial lower bound using some of he ingrediens above, e.g. local communicaion model realisic model wih inerference, e.g. wo-radii some kind of edge dynamics/churn and sill guaranees for delay/hroughpu/ec. Ad Hoc and Sensor Neworks Roger Waenhofer 10/79 Ad Hoc and Sensor Neworks Roger Waenhofer 10/80