Communication Networks. MAP-TELE 2011/12 José Ruela

Transcription

1 Communication Networks MAP-TELE 2011/12 José Ruela

2 Network basic mechanisms

3 Multiple Access

4 Multiple Access In early data communication systems, one solution to provide access of remote terminals to a central computer consisted in sharing a multipoint link in a multidrop configuration, under control of the computer the link was shared on a time basis using centralized polling, for communications between each terminal and the computer With the advent of LANs (wired or wireless) and of wireless cellular networks, multiple access protocols have been developed with the goals of providing general connectivity among a large number of stations and, at the same time, sharing the medium in an efficient way A multiple access protocol is necessary to coordinate and arbitrate the access of the stations to a shared medium, free of conflicts Access protocols may be centralized or distributed and are usually classified in a small set of categories Random access (contention based) protocols Controlled access protocols Channelization protocols

5 Centralized vs. distributed schemes In centralized schemes one station acts as a master, while the other stations (slaves) only transmit when they are allowed by the master, which decides the transmission schedule One simple example is centralized polling A centralized scheme may be a natural choice in some cases, such as in wireless LANs (stations communicate through an access point) and cellular networks (mobile stations communicate through a base station, which is connected to a wired network) Centralized schemes are simple and provide a single coordination point, but at the same time the master is a single point of failure (and thus redundancy may be necessary) and a possible bottleneck In distributed schemes stations cooperate on equal terms to arbitrate access to the medium without the intervention of a master Although more complex, distributed schemes are robust and reliable and usually provide lower access delay and higher network utilization than centralized ones (due to less overhead)

6 Medium Access Control From an architectural point of view, multiple access protocols belong to the Data Link layer, which has to be enhanced with functions not required in point-to-point links To account for such functions, a reference model for shared medium LANs was developed by IEEE IEEE 802 Reference Model In this model the Data Link layer is split into two sub-layers LLC(Logical Link Control) MAC(Medium Access Control) LLC provides a common service interface to higher layers, as well as optional error and flow control MAC provides medium access functions, framing, addressing (MAC addresses are assigned to physical interfaces) and error detection The IEEE 802 MAC service is connectionless Most of the existing LAN standards (that specify MAC and Physical layer functions) have been developed under IEEE 802 or adopt the IEEE 802 model

7 IEEE 802 Reference Model

8 IEEE 802 standards examples IEEE IEEE 802.1D IEEE 802.1Q IEEE IEEE IEEE IEEE IEEE IEEE IEEE IEEE IEEE IEEE LAN/MAN architecture; internetworking among LANs, MANs and WANs; link security; network management; protocol layers above MAC / LLC Media Access Control (MAC) Bridges Virtual Bridged Local Area Networks Logical Link Control CSMA/CD (Ethernet) Token Bus Token Ring Distributed Queue Dual Bus (DQDB) Wireless LAN Wireless Personal Area Network Broadband Wireless Access Resilient Packet Ring (RPR) Mobile Broadband Wireless Access (MBWA)

9 Topologies The basic physical topologies used in shared LANs are bus, tree, ring and star A physical star whose central node is a hub (multiport repeater) is logically equivalent to a bus (and will be implicitly considered as such when discussing bus access protocols) A ring can also be based on a physical star configuration in this case the central system is simply a wiring concentrator HUB Bus Star Ring

10 Random access protocols Random access is a class of multiple access protocols in which stations share a communications channel in a way that can lead to conflicts, usually called collisions Stations compete (contend) for the use of the channel The protocols in this class differ in a number of ways How stations cooperate to reduce the risk of collisions How stations detect collisions How stations handle collisions (and subsequent retransmissions) In the simplest protocols, stations do not sense the medium before transmitting and thus there is a high risk of collisions (even at moderate loads) this is the case of the ALOHA family of protocols In Carrier Sense Multiple Access (CSMA) protocols the risk of collisions is reduced (but is not null) since the stations sense (listen to) the medium before transmitting (carrier sense) Random access protocols are mainly used in pure broadcast networks buses and hub-based networks

11 ALOHA The ALOHA protocol was designed for a packet radio network developed to interconnect remote low speed terminals to a central computer in the University of Hawaii In ALOHA, a station transmits a frame whenever it has data to send Simultaneous transmissions result in collisions (even if frame overlap is partial) Under light network traffic, the delay to access the medium is low A collision appears to the receiver as an erroneous frame and therefore collisions are handled as part of the error control procedures The receiver sends a positive ACK to any frame received without errors The sender waits for an ACK during a round trip time (time-out interval) If an ACK is received, a new frame can be transmitted as soon as it is available If an ACK is not received (the frame was lost either due to a transmission error or a collision, or the ACK was lost), the station must retransmit the frame If stations retransmit immediately after a time-out, a new collision will occur if the real cause for the missing ACK was a previous collision To minimize the risk of new collisions, a station waits for a random time (back-off interval) before retransmitting the frame After a predefined number of unsuccessful attempts, the frame is discarded

12 ALOHA

13 ALOHA performance model S Useful load (in packets per packet time) rx useful packet rate (successful transmissions) S rx T frame G Total offered load (in packets per packet time) total packet rate (successful and unsuccessful transmissions) G T frame Traffic model (assumptions) Frames are of fixed length (transmission time equal to T frame ) Total offered traffic (G) is a Poisson process with infinite population S = G P 0 P 0 is the probability that no other packet is generated during the vulnerable interval of a station that is transmitting a packet S is always less than 1, but G may be larger than 1 (since it includes successful transmissions as well as all retransmission attempts)

14 ALOHA vulnerable period of a frame In the analysis we assume that frames have fixed length and thus a constant transmission time (T frame ) If a station starts transmitting a frame at t = 0, this reference frame will suffer a collision if another transmission starts in the interval ] T frame, + T frame [ The vulnerable period of a frame has a duration twice its transmission time, that is, 2 * T frame Collides with the start of reference frame Collides with the end of reference frame reference frame - T frame 0 + T frame + 2 * T frame Time vulnerable period

15 2 1 0 ) 2 (1 0 max 2 G G e dg ds G S G 18,4% 2 1 max e S! ) (2! ) 2 ( ] _ 2 _ [ 2 2 k e G k e T T in offered packets k P P G k T k frame frame k frame G G e e G P ! ) 2 ( G Ge GP S 2 0 ALOHA efficiency

16 Slotted ALOHA Stations synchronize their transmissions to the start of time slots It is necessary to distribute to the stations a synchronization signal for the start of time slots (it may be broadcast by a central station) When a station has a frame ready to transmit, it waits for the start of the next time slot and transmits There are no partial collisions either there is no collision or there is a full collision, that is, the vulnerable period is now equal to T frame (the duration of a time slot)

17 ! ) (! ) ( ] [ k e G k e T T in offered packets k P P G k T k frame frame k frame G G e e G P 0! ) ( 0 0 G Ge GP S max G dg ds G S 36,8% 1 max e S Slotted ALOHA efficiency

18 Aloha and Slotted ALOHA efficiency

19 Carrier Sense Multiple Access (CSMA) Carrier Sense Multiple Access (CSMA) protocols are based on sensing the medium before transmission A station does not start a transmission (defers) if it senses that the medium is busy, thus avoiding a collision that would occur with certainty Sensing that the medium is free does not avoid collisions Handling collisions in CSMA protocols can be performed in different ways but all CSMA variants have in common the duration of the vulnerable period In the worst conditions, the vulnerable period for a station after starting a transmission is the maximum round trip delay on the medium (2 ) If a collision does not occur within the vulnerable period, the station has acquired exclusive use of the medium (its presence is known by the other stations) and will complete transmission with success, which is a great improvement over ALOHA The use of CSMA protocols is recommended when the vulnerable period is much less than the transmission time of a frame, that is, when a = / T frame << 1 (a common situation in short distance, low speed LANs), but performance degrades when a increases and may become unacceptable for values close to (and higher than) one

20 CSMA vulnerable period of a frame If A starts a transmission at t 0, this will be known by B at t 0 +, and thus B will not start a transmission after t 0 + If station B starts a transmission before t 0, this will be known by A before t 0 and thus A will not start a transmission at t 0 The vulnerable period for A corresponds to B starting a transmission in the interval ] t 0, t 0 + [ and in this case a collision will occur at A after t 0 and before t (an interval of 2 at the start of transmission) A frame t B t 0 - t 0 t 0 + t

21 Carrier Sense Multiple Access (CSMA) basic protocol In the basic CSMA protocol, a station starts a transmission if it senses that the medium is free, completes the transmission and waits for an ACK during a time equal to the maximum round trip delay on the medium In case no ACK is received, the frame will be retransmitted after a random time (up to a maximum number of retransmissions) When the network is lightly or moderately loaded and a << 1 the probability of two or more stations starting transmission during a vulnerable period is low, but the risk increases when there is a transmission in course and there are stations waiting for the medium to become free A collision is certain to occur if two or more stations with frames ready to transmit persistently wait until the medium becomes free and then start transmitting The basic CSMA protocol offers various alternatives to handle this situation

22 CSMA alternatives Persistent If a station senses that the medium is free, it transmits If it senses that the medium is busy, it waits until it becomes free and transmits Non-persistent If a station senses that the medium is free, it transmits If it senses that the medium is busy, it waits a random time and repeats the algorithm p-persistent A slot time equal to the maximum round trip delay is used as a time unit to determine the delay for retransmission attempts If a station senses that the medium is free, it transmits with probability p and defers for one slot time with probability 1-p, and then repeats the algorithm If after deferring for a number of consecutive time slots a station finds that the medium is now busy, it waits a random time and applies the algorithm from the beginning If it senses that the medium is busy, it waits until it becomes free and proceeds as when sensing that the medium is free (i.e., applies the algorithm from the beginning) In all cases, if a station does not receive an ACK after a round trip time, it waits a random time and repeats the algorithm from the beginning

23 CSMA efficiency Persistent CSMA has a low delay under light and moderate loads, but is unstable at high loads Non-persistent CSMA is stable with high loads at the price of higher delays S = G / (1 + G), if a = 0 (ideal)

24 CSMA with collision detection (CSMA/CD) In CSMA, once a transmission starts, it will continue until the end, even if a collision would occur (a collision is not directly detected by the station, since it only senses the medium before transmitting) A major improvement can be achieved if a station continues to sense the medium after starting transmission This will allow a station to detect whether a collision occurs and, if this is the case, to abort transmission since the frame will be garbled and discarded by the receiver, thus not wasting transmission resources In fact, it is not really necessary to sense the medium during the whole transmission period, but just during the initial vulnerable period This method is called Carrier Sense Multiple Access with Collision Detection (CSMA/CD) and was designed and used for the first time in the experimental Ethernet and was later adopted as the basis for the IEEE standard Since then the Ethernet technology has evolved in different ways (type of media, physical topology, speed) and from shared to switched medium (switched Ethernet)

25 Ethernet sketch

26 CSMA/CD access protocol The protocol is based on the possibility of detecting collisions during the vulnerable period at the start of a transmission The vulnerable period (2) is taken as the unit of time (slot time) to synchronize the retransmission attempts after a collision occurs A station senses the medium before attempting transmission (carrier sense) If free, it transmits If busy, it waits until it becomes free and transmits (persistent) The drawbacks of persistent CSMA are attenuated with collision detection The station continues to sense the medium during the contention slot If no collision is detected, the station will complete transmission without the risk of a collision (under normal operation of stations) If a collision is detected, it is reinforced (jamming), the station aborts transmission and schedules frame retransmission according to a binary exponential back-off algorithm (the same frame may suffer successive collisions) For the first transmission attempt the algorithm is persistent (p = 1) but, after each collision, p is halved (p =1/2 n ) and the station selects with probability p one of the following 2 n contention slots for a retransmission attempt (n is the number of collisions already experienced by a frame and thus n = 0 for the first attempt) To guarantee that a collision is detected by a station during frame transmission, the condition a < 0.5 must hold

27 CSMA/CD efficiency Efficiency n A tx T T frame slot N P 1 1 S (1 n T tx frame 2 T P) N 1 n E prop tx n cont 1 2a NP (1 P) N 1 T 2 slot T prop P Probability of a station transmitting in one contention slot A Probability of exactly one station transmitting in one slot and acquiring the medium E n cont i(1 A) 1 A A S 1/ 2a i A i1 1/ 2a (1 A) / A 1 2a(1 A) / A 1 P=1/N A MAX 1 1 N N 1 lim 1 N 1 N N 1 1 e 1 lim S N a

28 CSMA with Collision Avoidance (CSMA/CA) In wireless LANs (WLANs) it is not possible to adopt CSMA/CD It is difficult to detect collisions on a radio interface, due to significant differences between transmitted and received power levels the benefits of detecting collisions and immediately aborting transmission of corrupted frames are lost Sensing the medium while transmitting would increase the complexity of the system (and thus its cost) It is necessary to use ACKs to deal with collisions (as in CSMA) and frames must be retransmitted if no ACK is received Retransmissions seriously degrade the performance and thus schemes that try to reduce the probability of collisions are essential in WLANs These reasons led to the adoption of a scheme called Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) in WLANs In IEEE , CSMA/CA is used by a Distributed Coordination Function (DCF), which supports asynchronous data transfer on a best-effort basis as the basic access method An optional Point Coordination Function (PCF) can provide contention free access by means of polling (this will be discussed under polling)

29 CSMA/CA protocol CSMA/CA (used by DCF in IEEE ) is based on sensing the medium before transmitting and on a binary exponential back-off deference If the medium is free, a station waits during an Interframe Space (IFS) period and starts transmission if the medium is still free If the medium is busy (or became busy during this IFS period) The station waits until it becomes idle Then it waits during IFS and sets a contention timer to a value that is a function of a contention window randomly chosen in the range [1, CW], where CW = 2 k - 1 is a predefined value of the window (a minimum value of k is defined for the first transmission attempt of a frame) When the contention timer expires, the station sends the frame and waits for an ACK If another transmission starts while counting down, the contention timer is disabled until the end of that transmission and then is enabled again If no ACK is received, the station assumes the frame was lost and tries again, after increasing k by one (until a maximum value is reached)

30 CSMA/CA back-off procedure Example of back-off procedure Example of exponential increase of the contention window CW The back-off time is equal to the current value of the contention window multiplied by the value of a parameter called aslottime that depends on the type of physical medium technology

31 CSMA/CA Interframe Space (IFS) Interframe Space (IFS) SIFS (Short IFS) used by high priority frames (e.g., ACK, answer to polling, etc.) PIFS (PCF IFS) used for polling by the master DIFS (DCF IFS) used for asynchronous access (contention)

32 Controlled access protocols In controlled access protocols, access to the medium is arbitrated by giving explicit authorizations to stations either in a centralized or distributed manner In centralized schemes the order and frequency of accesses is decided by a master station In distributed schemes the stations cooperate to determine transmission schedule Since a station is granted exclusive use of the medium when its turn arises, accesses will take place in an orderly manner and free of collisions Protocols of this class are usually classified as Reservation Polling Token passing (or Control Token)

33 Reservation schemes Reservation schemes are used when contention based schemes fail (e.g., when a is large) or when the overhead with polling is prohibitive It is common in satellite links and a reservation scheme has been specified for EPONs (Ethernet Passive Optical Networks) in IEEE (section 5) A station must make a reservation before sending In centralized schemes, a master station coordinates the reservations and grants the access rights, but distributed schemes are also possible (stations listen to each other reservations and know where in the cycle to transmit their frames) Transmissions are organized in cycles of variable length A cycle includes a reservation field and a data field that carries the frames scheduled for transmission on that cycle to transmit frames of variable length the reservation messages must include the frame lengths The reservation field is made up of minislots used for reservation Each station may own a dedicated minislot Stations may contend for reservation on minislots (using, for example, slotted Aloha) contention is only for reservations and thus the overhead is reduced As an example, GPRS uses slotted Aloha in the reservation channel in order to provide a data service over GSM

34 Polling schemes In polling schemes, stations take turns accessing the medium and at a given time only one station has the right to transmit a mechanism is required to pass the right to transmit between stations The basic polling mechanism is centralized a master station asks each station in turn whether it has data to send (roll-call polling) The order and frequency of polling is the responsibility of the master (round robin is the easiest) Polling is inefficient when the overhead of polling messages is high, when only a few stations are active (polling inactive stations wastes time) or when there are many stations (polling cycles are too long, especially when propagation delays are high) In distributed polling there is no central controller (stations agree on a polling order among themselves) but this is usually considered under the category of token passing schemes

35 Polling in IEEE In IEEE , the optional Point Coordination Function (PCF) supports a priority, polling-based contention-free access service Polling is centralized, with the Access Point (AP) as a Point Coordinator (PC) Channel Access time is divided into periodic intervals composed of a contention period (CP) and a contention-free period (CFP) During a CPF period, stations do not compete for the medium, but transmit only when polled by the AP (synchronous service) Polling is based on a single round-robin scheduling algorithm (not adequate to handle QoS requirements of different classes of traffic) IEEE e is a QoS enhancement that specifies a Hybrid Coordination Function (HCF) that supports a HCF Controlled Channel Access (HCCA) QoS stations send reservation requests that carry a TSPEC The Hybrid Coordinator (HC) is QoS-aware and allocates (by means of polling) transmission opportunities (TXOPs) to QoS stations, taking into account their specific flow requirements when scheduling the polling list

36 Token passing Token passing protocols may be viewed as a kind of distributed polling The authorization for a station to transmit is given by a special control frame called a token The control frame carries a free token in a simple case this may be indicated by a bit on a control field on its header (e.g., T = 0) The protocol is based on passing a free token among stations, so that stations take turns at accessing the medium on an orderly way Conflicts are avoided since there will be at most one free token in the network In order to transmit, a station has to wait for a free token, capture it to get exclusive right to access the medium, and transmit a data frame in its place This is equivalent to changing the state of the token to busy (T = 1, in a simple case), thus momentarily preventing other stations from accessing the network (no free token is available, since a data frame has replaced the token) Several problems have to be solved in token passing networks How to pass the free token among stations When and how to launch a new free token (change its state from busy to free) after it has been captured and used

37 Token passing in rings and buses A ring is the natural topology for implementing a token passing scheme (token ring) When there is no traffic, a free token circulates on the ring (the token is simply retransmitted by the repeaters that are part of the ring) A free token does not need to be explicitly addressed, since it visits all stations during its travel around the ring and a station not willing to transmit simply lets the free token untouched as it passes by In token rings another problem needs to be solved when and how to remove a frame from the ring (by not repeating it) and this is related with the launching of a new free token A token passing scheme can be implemented in bus (or broadcast) networks (token bus), by creating a logical ring among stations (each station has a logical predecessor and a logical successor) A free token must be explicitly addressed, that is, the station holding the token must send it with the physical (MAC) address of its successor The station holding the token must issue a new free token at once if it has no traffic to send or, otherwise, at the end of a frame transmission

38 Token ring A frame may be addressed to a single station, a group of stations or all stations (unicast, multicast or broadcast addresses, respectively) Each station must copy (receive) a frame addressed to it in the case of multicast or broadcast addresses, a frame must make a complete turn to the ring so that it can be copied by all addressed stations For frames addressed to a single station, there are two alternatives The frame is removed by the destination station, which must insert a new free token in the ring (care must be taken so that the token does not catch the frame still being transmitted, if this is too long) The frame is removed by the source station, which will issue a new free token when appropriate (two alternatives, Single Token and Multiple Token, will be discussed) Removing a frame by the source station has been the choice in two token ring standards (IEEE and FDDI) This has the advantage that stations have equal access opportunities, since the access is round robin (determined by the physical location) On the other hand, resources are wasted when a frame travels from the destination back to the source

39 Single Token and Multiple Token Single Token (IEEE at 4 Mbit/s) A free token is launched by the source after completing transmission of a frame and having started its removal as well If a < 1, the first condition implies the second Single Token simply means that there is at most one token in the ring (either free or busy) a free token is launched only after the busy token (on the frame header) is removed Multiple Token (FDDI) / Early Token Release (IEEE at 16 Mbit/s) A free token is launched by the source after completing transmission of a frame Multiple Token means that it is possible to have various tokens on the ring at the same time, but at most one free token since a busy token is carried by a frame, it is possible to have multiple frames in transit, provided that the ring latency allows it (a > 1) Single Packet A free token is launched by the source after having completely removed a transmitted frame (no fragments of different frames in transit simultaneously) It is highly inefficient and thus it has not been used in real networks Multiple Token has a better performance than Single Token when a > 1, and therefore should be used in such cases Single Token and Multiple Token have similar performance when a < 1, and thus Single Token is preferred since it is less complex to implement

40 Single Token example of operation

41 Token Ring IEEE (Single Token) When there is no frame transmission in course a free token circulates on the ring A station with a frame ready for transmission Waits for the free token Changes the state of the token to busy in the access control field Appends the rest of the frame to the header fields When the frame completes a turn to the ring (the header is detected), starts removing the frame (by not retransmitting it) Launches a new free token when has both completed the transmission of the frame and removed its header (that contains the busy token) The second condition allows to implement a priority mechanism, which is based on reservations Stations access the medium in a round robin manner and the duration of a cycle depends on the number of stations that transmit on that cycle and the duration of each transmission, which is bounded by a Token Holding Time (THT)

42 Token Ring IEEE (Single Token)

43 Single Token and Multiple Token efficiency Consider a ring with N attached stations, with a latency Lat (propagation plus station delays) and the time to transmit a frame T f < THT (a = Lat / T f ) Assume that during an access cycle N a < N stations transmit a frame each In Single Token with a < 1 or in Multiple Token the duration of a cycle (token rotation time) is given by N a * T f + Lat and then the efficiency is S N a N a * T * T f f Lat T f T f Lat N a 1 a 1 N a In Single Token with a > 1, the token rotation time is N a * Lat + Lat and then S N a N a * T f * Lat Lat Lat T f Lat N a a 1 a N a In all cases, the maximum efficiency is achieved when N a = N

44 Token Bus It is necessary to create a logical ring among stations, independent of their physical location (A B C D A in the example) The station that holds the token addresses it explicitly to its logical successor, even if this has no frames to transmit IEEE is a Token Bus standard for industrial applications A B D C

45 Slotted rings The ring is divided into an integer number of fixed size slots that continuously circulate on the ring (the number of slots n is equal to a > 1) Each slot may carry a packet (or fragment) The state of each slot (empty / occupied) is indicated by a bit on the header Slots are initially created empty One station ready to transmit waits for an empty (idle) slot, changes its state to occupied (busy) and inserts a packet on that slot Releasing a slot (changing its state to empty) may be performed by the source station (Cambridge ring) or the destination station (Orwell ring) Releasing a slot by the source station allows round robin access to the slot Releasing a slot by the destination station allows a better utilization of the ring, but requires additional measures to avoid unfair access among stations Since access to slots is independent, simultaneous accesses by stations are possible, in case there are multiple slots circulating on the ring A slot may be seen as carrying its own token (free or busy) and thus the operation of a slot is equivalent to a single token with a > 1 (but the ring carries n = a slots) The access protocol is efficient, but this advantage is lost in low latency rings, since the size of the slots may be so small that the relative overhead due to the header (control, addresses) is very high (such as in the Cambridge ring)

46 Channelization Channelization techniques are used to divide a shared medium into multiple channels for allocation to different stations, thus isolating their traffic The available bandwidth may be shared in frequency, time and code, to which correspond three basic mechanisms Frequency division multiple access (FDMA) Time division multiple access (TDMA) Code division multiple access (CDMA) The first two are generalizations of FDM and TDM used in point to point links, while the third one is based on coding techniques The basic techniques can be combined in different ways, as exemplified with various cellular networks

47 FDMA The bandwidth is divided into channels and each station transmits on its own frequency band (channel) Frequency bands are separated by guard bands Receivers tune to the right frequency to listen to a given transmitter FDMA is best suited for analogue links and for constant rate traffic The channel allocated to a station that is not transmitting cannot be reused by other stations The number of frequencies is limited due to spectrum limitations It is not feasible to allocate a separate frequency to hundreds or thousands of devices in a listening area The solution is to partition an area into cells, reduce transmitter power and reuse frequencies in non adjacent cells A consequence is the need for hand-off when a mobile station crosses a cell boundary, which requires synchronization between the mobile station and two base stations

48 TDMA In TDMA the whole bandwidth is considered as a single channel that is shared by stations on a time basis (time slots are allocated to stations) All stations transmit data on the same frequency, but at different times For the same frequency band, FDMA and TDMA can handle approximately the same number of stations, but TDMA has some advantages over FDMA It has a better performance concerning the mean delay to transmit a packet It is possible to dynamically allocate different bandwidth shares, by controlling the amount of slots given to each station Mobile stations require less power, since they can switch off the transmitters except during their own time slots Mobile stations can use the slots when they are not transmitting to measure the received power from multiple base stations and select the best one (for a possible hand-off) Time synchronization is required among stations, since they are located at different distances from a base station, and thus an extra overhead is incurred while, in the direction from the base station, only TDM is involved TDMA has greater problems than FDMA with multipath interference on wireless links since transmissions have a shorter duration (higher bit rate)

49 CDMA In CDMA, transmissions from different stations occupy the same frequency band at the same time The bit time is divided into n short intervals called chips and each station is assigned a unique n bit codeword (chip sequence) The chip sequence replaces a 1, while a 0 is replaced by the one s complement of the chip sequence When a station is idle, it sends no signal Symbolically: data bit 1 +1 data bit 0-1 idle 0 CDMA is a form of spread spectrum communications, since the bandwidth of the transmitted signal is increased by a factor of n (spreading factor) If the codewords (chip sequences) of the stations are orthogonal (i.e., the inner product of any two sequences is zero), the bit sequence of one station may be recovered by a decorrelator that applies the station s chip sequence to the received signal

50 CDMA coder and decoder +

51 Frequency and Time Division Duplex Frequency Division Duplex (FDD) and Time Division Duplex (TDD) are two ways of converting a wireless medium into a duplex channel In FDD, uplink and downlink traffic use different frequencies FDD requires that the frequency bands are allocated in advance and thus cannot adapt to changes in the ratio of uplink to downlink traffic In TDD, uplink and downlink traffic use different time slots, which are used in turn by the stations and the base station TDD requires a coordination of transmissions in both directions, but easily adapts to changes in the ratio of uplink to downlink traffic and is flexible and efficient in handling asymmetric traffic FDD and TDD can be combined with FDMA, TDMA and CDMA TDD/FDMA is used in second-generation cordless phones FDD/TDMA/FDMA is used in digital cellular phones FDD/CDMA and TDD/CDMA are used in UMTS

52 Annex Bridging and LAN switching

53 Evolution of Ethernet / IEEE networks The span of Ethernet / IEEE networks based on shared media may be increased using repeaters to interconnect coaxial cable segments or hubs (multiport repeaters) in a tree-like structure In both cases, such interconnected physical segments constitute a collision domain Repeaters are physical layer devices and therefore do not process frames or packets they simple regenerate and broadcast signals on output ports Logical segmentation of the network allows reducing the size of such collision domains, thus improving performance Bridges, which operate at the MAC layer, perform an intelligent filtering of traffic (based on MAC addresses carried on frame headers), thus isolating collision domains However, a bridged LAN is still a logical broadcast domain (at MAC layer) Routers, which operate at the Network layer, isolate broadcast domains, that is, they are barriers to broadcast traffic (at physical and MAC layers)

54 Collision and broadcast domains with bridges Collision domain 1 Collision domain 2 Collision domain 3 Broadcast domain 1 Broadcast domain 2 coaxial cable repeater hub bridge router Collision domain Collision domain

55 Bridges and bridging Although IEEE 802 has defined two bridging mechanisms, source routing and transparent bridging, the former was tightly coupled to IEEE token ring networks Today, Ethernet / IEEE networks are dominant in LAN environments and thus, for practical purposes, transparent bridging is the only widely used solution Transparent bridging requires that the logical topology of the network is open, thus eliminating loops that exist on mesh topologies, which may be adopted in order to improve the robustness of the network The logical topology of a bridged LAN must be a spanning tree (in order to cover all LAN segments) The spanning tree is built, maintained and reconfigured (if necessary) by the bridges, based on a Spanning Tree Protocol (SPT) Bridges are transparent to the stations, which are unaware of their presence Transparent bridging is defined in IEEE 802.1D

56 Bridges frame forwarding In order to decide whether a frame should be forwarded and to which output port(s), a bridge has to build a forwarding or switching table (called filtering database in IEEE 802.1D) this is not based on a routing protocol but rather on a learning process Learning is based on inspecting the source MAC address (SA) of any received MAC frame and associating that MAC address to the corresponding incoming port creating a new entry (if the address is not yet in the table) or updating an existing one Entries are temporary and are removed after an ageing time, if not refreshed as a result of traffic activity Frame forwarding is based on the destination MAC address (DA) of received frames either unicast, multicast or broadcast If DA is a unicast address not present on the forwarding table, a multicast or a broadcast address, the frame is copied and forwarded to all ports in the forwarding state, except the incoming port If DA is a unicast address present on the forwarding table, the frame is forwarded to the associated port, except if this is the incoming port, in which case it is discarded this is how intelligent filtering is performed

57 IEEE 802.1D bridges Learning Bridge architecture Forwarding

58 LAN switches LAN switches were developed as an evolution of the hub technology but perform similar functions as conventional bridges With LAN switches a collision domain collapses when a single station is connected to a switch port (private LAN) in this case full-duplex operation becomes possible and CSMA/CD is no longer necessary Stations may be connected to a switch port through a hub, in which case half-duplex operation is mandatory (CSMA/CD enabled) Full-duplex operation applies in the point-to-point links between switches LAN switches may be connected in a mesh topology, which requires configuring the logical topology as a spanning tree LAN switches allow a logical segmentation of the network independently of the physical location of stations (not possible with conventional bridges) This is the basis for the concept of Virtual LAN (VLAN) stations on a VLAN form a logical group and belong to the same broadcast domain (at the MAC layer), independently of their physical location A switch may support multiple VLANs, a single VLAN may span multiple switches and a station may belong to multiple VLANs (e.g., a router or a server) VLANs need to be connected by means of routers, but a new family of routing devices (usually called router switches) was developed they route packets between VLANs but switch frames on the same VLAN

59 Collision and broadcast domains with LAN switches 1 i n A collision domain is associated with each switch port A single station may be connected to a switch port (private LAN) A number of stations may be connected to a switch port through a hub (shared LAN) Collision domain 1 Private LAN Collision domain i Shared LAN Collision domain n Private LAN hub Broadcast domain 1 Broadcast domain 2 Broadcast domain 3 switch Broadcast domains are created by configuration

60 Virtual LANs concept VLAN 1 VLAN 2 VLAN 3 S3 S6 S9 Floor n + 1 S2 S5 S8 Physical partition Floor n Bridge or 7 8 S1 S4 S7 switch 9 Floor n 1 Logical partition IEEE 802.1Q extends IEEE 802.1D so that it applies to VLANs (virtual bridged LANs)