High performance communication networkss
- 1. NETWORK ELEMENTS A communication network is a collection of network elements interconnected and managed to support the transfer of information from a user at one network location or node to a user at another node. There are two principal network elements, and we must know how the properties of these elements affect the characteristics of the services that they implement.
- 2. NETWORK ELEMENTS -Principal Network Elements-links and switches. A link transfers a stream of bits from one end to the other at a certain rate with a given bit error rate and a fixed propagation time. Links are unidirectional. The most important links are optical fiber, copper coaxial cable, and microwave or radio "wireless" links. Optical fiber and copper links are usually point-to- point links, whereas radio links are usually broadcast links. Several incoming and outgoing links terminate at a switch, which is a device that transfers bits from its incoming links to its outgoing links. Whenever the rate of incoming bits exceeds that of outgoing bits, the excess bits are buffered at the switch.
- 3. NETWORK ELEMENTS -Principal Network Elements-links and switches. When we view a network as an interconnection of network elements, we may represent the network as in the graph of Figure 2.1. In the graph, edges denote links, large circles denote switches (including buffers), and small circles denote user nodes where bits are generated or consumed. We will use the names user, source, destination, station, and node as near synonyms.
- 4. NETWORK ELEMENTS -Principal Network Elements
- 5. NETWORK ELEMENTS -Principal Network Elements Another very important view of an interconnection of network elements is provided by the queuing network model. Consider a switch with η incoming links and m outgoing links, as in the left of Figure 2.2. The diagram on the right is the queuing model. Each incoming link's receiver writes into its input buffer, and each outgoing link's transmitter reads from its output buffer. The switch transfers bits or packets from input buffers to the appropriate output buffers. This queuing network model of switches and links is used to describe and evaluate network performance. For example, the queuing delay encountered by a packet in an output buffer is proportional to the number of packets in the buffer in front of it. If packets arrive into a full buffer, there will be packet loss. It is difficult to calculate queuing delay and packet loss.
- 6. NETWORK ELEMENTS - Network Elements and Service Characteristics A packet generated by a source travels over one link, gets buffered at a switch, is then routed to another link, and so on, until it arrives at its destination. The delay packets experience through a network depends on the elements that constitute the network, the traffic that goes through these elements, and the way the network is operated. The detailed analysis of the delay through a network is rather involved. For now, note that we can decompose the total delay into four components: total delay = TRANS + PROP + QD + PROC TRANS is the time required to transmit a packet, so TRANS = (packet size)/(transmission speed) For example, for a 10,000-bit packet and a transmission speed of 1 Mbps, TRANS is 10 ms. PROP is the signal propagation time, so PROPP = (distance from source to destination) / (speed of electrical or optical signal).
- 7. NETWORK ELEMENTS - Network Elements and Service Characteristics Propagation time for an electrical or optical signal is between 3.3 and 5 mi- croseconds per kilometer (µs/km). QD is the queuing delay in the switch. It occurs whenever the bit rate of the traffic coming into the switch exceeds the capacity of the outgoing link. The excess bits are queued in the switch buffers. In contrast with the other two sources of delay, queuing delay is significantly affected by the network control policy. Finally, PROC is the processing time required by the network switches. We will assume that this processing time is negligible
- 8. NETWORK ELEMENTS - Network Elements and Service Characteristics Suppose as a rough rule of thumb that the network is controlled so that each packet coming into the switch has to wait on average for four previous packets to be transmitted. Then the average queuing delay is four times the transmission delay, so total delay = 5 χ TRANS + PROP. Delay depends on the transmission rate and length of the links and on the length or size of messages.
- 9. Network Mechanisms A network's bearer services comprise the end-to-end transport of bit streams, in specific formats, over a set of routes. These services are differentiated by quality: speed, delay, errors. They are produced using five basic mechanisms: multiplexing, switching, error control, flow control, congestion control, and resource allocation
- 10. Network Mechanisms Multiplexing combines data streams of many such users into one large band width stream for long durations. Users thereby can share in the scale economies of transmission. However, users are not concentrated in a few locations; they are geographically dispersed. Switching allows us to bring together the data streams of these dispersed users. (In essence, the communication network in- dustry makes profits by installing large transmission capacity and by "renting" this capacity to users in smaller amounts using multiplexing and switching.)
- 11. Network Mechanisms In the left side of Figure 2.4 we see a network with no multiplexing or switching. Each pair of phones is connected by a dedicated link of capacity 64 Kbps needed to carry one voice conversation. In this network, the average number of links per phone, and hence the average cost, increases with the number of nodes. Since a telephone is engaged in at most one conversation at any time, link utilization, that is, the fraction of time the link is busy, decreases as the number of nodes increases. Thus the cost per phone grows with the size of the network. This network wastes resources. The network on the right employs multiplexing and switching. Each phone is connected by a dedicated access link to one of two local switches, which are connected by a single link called a trunk. (In telephone networks, a switch is located in a central office; a link between two switches is called a trunk; a link between a subscriber telephone and a switch is called an access line or subscriber loop.) The access line capacity is 64 Kbps, and the trunk capacity is a multiple of 64 Kbps. (In this example it may be 2 χ 64 Kbps.)
- 12. Network Mechanisms Depending on its capacity, a trunk can carry several voice conversations simultaneously by multiplexing. A conversation between two phones will occupy only two access lines—if both phones are connected to the same local switch—or, in addition, it will occupy a fraction of the trunk capacity. It is the task of the switch to determine whether a call originating from a local phone is destined for another local phone or for a phone connected to the remote switch. In this network, there is one access line per phone, but the switch and trunk capacities grow much less rapidly than the number of phones. As a result, the average cost of the network per user decreases with the number of phones. This decreasing cost structure of communication networks is made possible by multiplexing and switching. Another important mechanism is error control. All transmission links oc- casionally corrupt the messages they transmit. Although carefully designed and maintained links have a very small bit error rate (e.g., 10-12), even the rare errors in the transmissions may not be acceptable. Moreover, in addition to transmission errors, a message may fail to reach its destination because it arrived at a switch whose buffer was full and so the message was discarded. It is therefore important for the network to control such errors.
- 13. Network Mechanisms A source must prevent overwhelming the receiver by sending messages faster than the receiver can store or process them. Flow control is a mechanism that enables the receiver to pace the transmissions of the source. The end-to-end delay is the sum of a fixed propagation delay and a variable queuing delay. In order to keep this delay within an acceptable range, the rate at which packets enter the network must be controlled. Congestion control is the generic name for a set of mechanisms designed to limit the rate or number of packets introduced into the network by a source or a switch. If the congestion control mechanism does not function properly, an excessive number of packets may accumulate in the switch buffers causing unacceptable delay or loss.
- 14. Network Mechanisms We have seen that some applications require the network to provide a minimum bandwidth to ensure acceptable performance. For example, a voice conversation requires 64 Kbps and MPEG1 requires 1.5 Mbps. A variable bit rate application will require a guaranteed combination of minimum bandwidth and buffers. Because network resources—link bandwidth and switch buffers—are shared by many applications at the same time, resource allocation mechanisms must be designed to ensure that each application receives the necessary resources to maintain its quality of service. Multiplexing is carried out in switches or specialized hardware. Switching and resource allocation are implemented in switches. Error control and flow control are implemented in switches and host computers.
- 15. Multiplexing Three important techniques for multiplexing Ν incoming channels onto one outgoing channel. These are 1. time-division multiplexing, 2. statistical multiplexing, and 3. frequency-division multiplexing. For our present purposes a channel is a communication link of fixed capacity measured, say, in bits per second (bps). That is, each second of time on the channel is divided into a number of intervals called bit times, the number being equal to the channel capacity. For example, in a 1-Mbps channel, a bit time is 1 μs long. Each bit time may be occupied by an information or data bit, or it may be empty. We also say, accordingly, that the channel is busy or idle at that time. The average data rate divided by channel capacity, or the fraction of time the channel is busy, is the channel utilization. Utilization is equal to 100% only if the instantaneous data rate is always equal to the channel capacity.
- 16. Multiplexing-TDM Multiplexing is the process by which information bits from Ν incoming channels are transferred into bit times on one outgoing channel. Demultiplexing is the reverse process: the information bits on one incoming multiplexed channel are separated and transferred onto Ν outgoing channels. The multiplexed outgoing channel contains some extra bits (in addition to the incoming channels' data bits) that the demultiplexer uses to determine which data bits belong to which incoming channel. Time-division multiplexing or TDM is illustrated in Figure 2.5. In TDM, the capacity of the outgoing channel is divided into Ν logical channels, and data in each of Ν incoming channels is placed in a designated outgoing logical channel. This is achieved as follows. Time on the outgoing channel is divided into fixed- length intervals called frames. Frames are delimited by a special bit sequence called a framing pattern, not shown in the figure. Time in each frame is further subdivided into Ν fixed-length intervals called slots (this name is not always used). Thus each frame consists of a sequence of slots: slot 1, slot 2, . . . , slot N.
- 17. Multiplexing-TDM (A slot is usually 1 bit or 1 byte wide.) A logical channel occupies every Nth slot. There are thus Ν logical channels. The first logical channel occupies slots 1, Ν + 1, 2N + 1,...; the second occupies slots 2, Ν + 2, 2N + 2,...; and so on. The multiplexer operates as follows. The data bits in each incoming channel are read into a separate FIFO (first in, first out) buffer. The multiplexer reads this buffer in sequence for an amount of time equal to the corresponding slot time: buffer 1 is read into slot 1, buffer 2 is read into slot 2. (If there are not enough bits in a buffer, the corresponding slot remains partially empty.)
- 18. Multiplexing-TDM The bit stream of the outgoing channel is easily demultiplexed: the demultiplexer detects the framing pattern from which it determines the beginning of each frame, and then each slot. TDM is easy to implement. The overhead due to extra framing bits is small. However, the utilization of the outgoing channel may vary a great deal depending on the burstiness of the incoming data streams. To see this, observe that the capacity of the outgoing channel must be as large as the sum of the capacities of the Ν incoming channels. As a result, the utilization of the outgoing channel will be low or high accordingly as the utilization of the incoming channels is low or high. TDM leads to high utilization if the incoming data is not bursty. Thus TDM is ideal for constant bit rate traffic
- 19. Multiplexing-SM Statistical multiplexing or SM is most effective in the case of bursty input data. As in TDM, the data bits in each incoming channel are read into separate FIFOs. The multiplexer reads each buffer in turn until the buffer empties. (It is customary to call the data read in one turn a data packet.) In TDM each FIFO is read for a fixed amount of time—one slot—and so each incoming channel is allocated a fixed fraction of the outgoing channel capacity, independent of the data rate on that channel. By contrast, in SM, the capacity allocated to each incoming channel varies with time, depending on its instantaneous data rate: the higher the rate, the larger the capacity allocated to it at that time. As a result, the capacity of the outgoing channel needs to be only as large as the sum of the average data rates of the incoming channel, which, for bursty traffic, may be much smaller than the sum of the peak data rates. Hence the capacity of the outgoing channel maybe smaller than the sum of the incoming channel capacities. We call the ratio of the total incoming capacity to the total outgoing capacity as the multiplexing gain. This gain is unity for TDM but can be much larger for SM. Figure illustrates the possible multiplexing gains for SM relative to TDM and FDM for various applications.
- 20. Multiplexing-SM
- 21. Multiplexing gain is unity for TDM but can be much larger for SM. Figure illustrates the possible multiplexing gains for SM relative to TDM and FDM for various applications
- 22. Multiplexing-SM In SM the size of packets read from each FIFO can vary across channels and over time within each channel. Therefore, the demultiplexer cannot sort the packets belonging to different channels merely from their position within a frame. Consequently, additional bits, which delimit each packet and identify the corresponding incoming channel or source, must be added to each packet. The resulting overhead is significantly larger than under TDM. It also becomes more difficult to implement the multiplexer (which must now add the packet delimiter and channel or source identifier) and the demultiplexer (which must locate and decode those bit patterns). These in creases in complexity and overhead must be balanced against high utilization in the face of bursty data to determine whether SM or TDM is more efficient. In general, constant bit rate traffic, such as voice, fixed-rate video, and control and sensor signals, are better handled by TDM, whereas message traffic such as database transactions and variable bit rate video are better handled by SM. Not surprisingly, telephone networks use TDM, whereas computer communication networks use SM
- 23. Multiplexing- Frequency-division multiplexing or FDM Frequency-division multiplexing or FDM is illustrated in Figure. The frequency band of the outgoing channel is divided into distinct fixed bands, one for each incoming channel. The signal in each incoming channel is modulated to fit into its assigned band. The signal on the outgoing channel is simply the sum of these modulated signals. Thus the bandwidth of the outgoing channel must be greater than or equal to the sum of the bandwidths of the incoming channel. (In this sense, FDM is similar to TDM.) For demultiplexing, the FDM signal is passed through an appropriate filter. By demodulating the filter output, we obtain the appropriate input signal. FDM is used in AM and FM radio and TV broadcast as well as in CATV distribution. cellular radio. FDM is more flexible than the other two multiplexing schemes. Indeed, one incoming channel may contain an analog signal, and another may be digital.
- 24. Multiplexing- Frequency-division multiplexing or FDM FDM has two disadvantages. First, it is wasteful of bandwidth since the frequency bands assigned to an incoming channel must be separated by a "guard band" from the other channels. Second, if the transmission link exhibits significant nonlinearities, there will be "cross-talk" among the different signals, leading to errors. We have used bandwidth (or size of a link's frequency band in Hz) and speed or bit rate of a link (in bps) interchangeably, but that is not strictly correct. A (digital) bit stream is converted into an analog signal by a modulation scheme like phase-shift keying (PSK). The frequency spectrum of the resulting analog signal must lie within the link's frequency band. As a result the bit rate is proportional to the bandwidth.
- 25. Multiplexing- Frequency-division multiplexing or FDM The ratio bit rate/bandwidth measured in bps/Hz is called the spectral efficiency, which typically ranges between 0.5 and 8.0. The maximum value of the achievable spectral efficiency is given by Shannon's theorem as a function of the noise in the communication channel. When bandwidth is scarce, as in wireless radio and satellite channels, communication engineers spend a lot of effort designing modulation schemes to increase spectral efficiency.
- 26. Multiplexing- Code division multiplexing CDMA Code-division multiplexing is yet another method for multiplexing different bit streams on a common link. Some cellular networks use this multiplexing method. This modulation method is used for wireless systems
- 27. Switching-Circuit Switching and Packet Switching Circuit Switching Network
- 28. Circuit Switching Network When a network is to transfer a stream of data from a source to a destination (e.g., from A to D or C to Β in the following figures), it must assign to the stream a route, that is, a sequence of links or channels connecting the source to the destination, and then allocate to the stream a portion of the capacity or bandwidth in each channel along the route to be used to transfer the stream. Those decisions are implemented in switches. Route selection and bandwidth allocation are examples of resource allocation. The name switch is used in telephony; in computer communications, the device that performs routing is also called a router. We shall use switch and router interchangeably.
- 29. Circuit Switching Network In circuit switching, illustrated in Figure, the route and bandwidth allocated to the stream remain constant over the lifetime of the stream. Further, the capacity of each channel is divided into a number of fixed-rate logical channels, called circuits. The division is usually accomplished by TDM. Thus circuit switching involves assigning to the bit stream a route and one circuit in each link along the route. This assignment is made before any bits are transferred in a phase known as call or connection setup. At the end of this phase, data transfer is carried out. When the transfer is complete, the route and the circuits are deallocated. That phase is called connection teardown.
- 30. Circuit Switching Network Circuit switching thus involves three phases: (1) the source makes a connection or call request to the network, the network assigns a route and one idle circuit from each link along the route, and the call is then said to be admitted (if the network is unable to make this assignment, the call is rejected); (2) data transfer now occurs—the duration of the transfer is called the call holding time; (3) the call is then torn down. The switch computers maintain information about which circuits are busy (i.e., currently allocated to calls in progress) and the routing tables. The computers also execute algorithms implementing call admission policies and routing strategies. They also record call duration and other statistics needed for purposes of administration, billing, and maintenance.
- 31. Circuit Switching Network Circuit switching is easy to implement relative to other schemes, but because a stream is assigned a fixed-rate circuit, capacity utilization may be low if the data stream is bursty. It is therefore used in voice networks but not in networks designed for data transfer. Since circuit switching assigns a fixed bandwidth to a call, there is no queuing delay, and we can guarantee ahead of time the maximum end-to-end delay from source to destination experienced by the data stream. This guaranteed-delay feature may be essential in control, videoconferencing, and other real-time applications. From this delay equals the transmission plus propagation times. DisAdv: LOWER LINK UTILISATION
- 33. Packet Switching Network The data stream originating at the source is divided into packets of fixed or variable size. The time interval between consecutive packets may vary, depending on the burstiness of the stream. As the bits in a packet arrive at a switch or router, they are read into a buffer. When the entire packet is stored, the switch routes the packet over one of its outgoing links. The packet remains queued in its buffer until the outgoing link becomes idle. This store-and-forward technique thereby introduces a random queuing delay at each link; the delay depends on the other traffic sharing the same link. (Packets from different sources sharing the same link are statistically multiplexed.) The total delay at a switch is the queuing delay plus the time taken to transmit the entire packet.
- 34. Packet Switching Network The routing decision is determined in one of two ways: datagram and virtual circuit. In datagram packet networks, each packet within a stream is independently routed, as illustrated in Figure 2.11. A routing table stored in the router (switch) specifies the outgoing link for each destination. The table may be static, or it maybe periodically updated. In the latter case, the outgoing link assigned to a destination depends on the route path to the destination. Since the estimate may change with time, consecutive packets may be routed over different links. (Observe in the figure that packets from A to D travel over two routes.) Therefore each packet must contain bits denoting the address of the source and destination. This may be a significant overhead (since the addresses are often quite long) if the average packet size is small; the overhead may be negligible if the packet size is long, but then the packet transmission time and the queuing delay are also long.
- 35. Packet Switching Network-datagram network
- 36. Packet Switching Network-Virtual circuit packet network
- 37. Packet Switching Network-virtual circuit packet network In virtual circuit packet networks, a fixed route is selected before any data is transmitted in a call setup phase similar to circuit-switched networks. (See Figure 2.12.) However, there is no notion of a fixed-rate circuit or logical channel. All packets belonging to the same data stream follow this fixed route, called a virtual circuit. Packets must now contain a virtual circuit identifier; this bit string is usually shorter than the source and destination address identifiers needed for datagrams. However, the call setup phase takes time and creates a delay not present in datagram packet networks.
- 38. Datagram Switching 1.higher link utilization especially when traffic is bursty 2. used in data networks 3. the end-to-end delay may be so large or so random as to preclude applications that demand guaranteed delay. 4. the overhead due to source and destination identifiers and bits needed to delimit packets may waste a significant fraction of the transmission capacity if the packets are very short. 5. Since consecutive packets may travel over different routes, packets may not arrive at the destination in the same sequence in which they were sent. 6.The host at the destination may then have to buffer several packets and resequence them, adding to the end-to-end delay. 7. Lastly, a datagram switch does not have the state information to recognize if a packet belongs to a particular application. Hence the switch cannot allocate resources (bandwidth and buffers) that the application may require. Thus it is difficult to implement sophisticated resource allocation schemes in a datagram network.
- 39. Virtual circuit switching Since packets arrive at the destination in order of transmission, no resequencing is needed Statistical multiplexing of packets at the router or switch can achieve better utilization than in circuit switching. But the utilization maybe lower than under datagram switching, since routing decisions in the latter may be changed at the time scale of individual packet durations. those decisions can be changed only at the time scale of call durations. Lastly, since packets contain their virtual circuit identifiers or VCIs, the switch can allocate resources depending on the VCI. During the connection setup phase the switches may be notified that a particular VCI should be given extra resources. compromise between datagram and circuit switching
- 40. Circuit switching Vs Packet Switching In circuit switching , bit streams are allocated fixed-capacity circuits, and incoming b streams are time-division multiplexed into outgoing streams. Only a tiny amount of buffering is needed at the circuit switch—enough to compensat for short-term fluctuations in the frequency and phase of the incoming and outgoing b streams. In packet switching, the rate of the incoming stream may exceed the capacity of th outgoing link for significant time intervals. Thus packet switches must be able to buf the excess incoming traffic. Consequently, buffer management is another task performed at the switch in addition to routing and multiplexing.
- 41. Error Control To control errors, a transmission link can use two methods: 1. error detection and 2. error correction. A simple error-detection method is the parity bit used, for example, in the RS-232-C serial line link ,the transmitter appends a 1 or a 0 to the packet it transmits so that the resulting string (packet + parity bit) contains a number of Is that has an agreed parity, say odd. For instance, if the original packet is 0100100, then the transmitter appends a 1 and transmits the resulting string 01001001, which contains an odd number (three) of Is. If transmission errors modify the string into one whose number of Is is even, then the receiver detects an error. For instance, if the receiver gets the string 01101001, then it knows that the transmission modified at least one bit. This parity bit method does not detect errors that modify an even number of bits.
- 42. Error Control Thus, this method is not very reliable for long packets: it is not unlikely that transmission errors modify 2 or 4 bits in long packets. Communication engineers developed error-detection methods that detect all but very unlikely errors by adding more than one error-detection bit to the packets. Two methods: cyclic redundancy check (CRC) and the checksum code (CKS)
- 43. Error detection and Error Correction
- 44. CRC method The CRC method works as follows. A binary word G forms the basis of the CRC calculation. Standards specify the word G that should be used for different networks. Denote by r + 1 the number of bits of G. Assume that we want to transmit a packet that consists of the finite string of bits P. We denote by P2r the binary word obtained by appending r 0s to the right of the packet P. By performing a long division, an electronic circuit in the transmitter calculates the remainder R of the division of P2r by G. This long division is calculated by performing the additions of the binary words bit by bit modulo 2 and without carry. The remainder R has r bits and is such that P2r =A.G + R. The binary word R is the CRC code that the transmitter appends to the packet. That is, the transmitter sends the binary word P2r + R.
- 45. CRC method Note that, because of the rules used to perform the operations, P2r + R = A.G + R + R = A.G. Thus, every transmitted packet, together with its CRC bits, is an exact multiple of G. If the receiver gets a packet that is not a multiple of G, then it knows that transmission errors corrupted the packet. We can represent transmission errors by a binary word Ε whose 1’s indicate which bits the transmission corrupts. With this representation the packet that the receiver gets is A.G + E. This received packet is a multiple of G, and the receiver does not detect the errors if and only if Ε is a multiple of G.
- 46. Error Correction method The middle panel of Figure 2.13 illustrates a simple error-correction method that transmits every packet three times. By performing a majority vote on every bit, the receiver can correct transmission errors as long as they affect each packet bit in at most one of the three transmissions. Thus, if the third bit in the packet is a 0 and if one of the transmissions modifies it into a 1, then the receiver can decide that the transmitter sent a 0 and thereby correct the error. This replication method is wasteful and is not very robust.
- 47. Typical implementation of a protocol
- 48. Typical implementation of a protocol The transmitter and receiver follow a specific set of rules, a protocol, for making sure that the receiver eventually gets every packet correctly and that it discards corrupted packets and duplicated correct copies. These protocols use timers and acknowledgments. Before sending a packet, the transmitter makes a copy that it keeps. The transmitter then sets a count-down timer to a specific value and sends the packet. We say that the packet times out when the count-down timer reaches the value 0. If the receiver gets the intact packet, then it sends back an acknowledgment to the transmitter. If the transmitter gets the acknowledgment of the packet before the packet times out, then the transmitter knows that the receiver got the packet. In that case, the transmitter discards its copy of the packet and repeats the procedure with the next packet.
- 49. Transmission corrupted by errors
- 50. Transmission corrupted by errors If the receiver gets a corrupted packet (see Figure 2.15), then it does not send an acknowledgment. When the packet times out, the transmitter assumes that something went wrong during the packet transmission, and it repeats the procedure with a copy of the packet. The packet also times out when its acknowledgment is corrupted during its transmission. If there is no permanent hardware problem and if the initial value of the timer is large enough, the transmitter eventually gets an acknowledgment before it times out and every packet eventually reaches the receiver. We explain the retransmission protocols that networks use below. For real time applications and storage, the network should use error correction instead of error detection.
- 51. Alternating Bit Protocol -Simplest retransmission protocol The sender numbers the packets alternately 0 and 1. The receiver acknowledges every correct packet that it receives with the same number as the received packet. The sender waits for the receiver's acknowledgment. Figure 2.16 shows the sequence of events when there is no transmission error.
- 52. Alternating Bit Protocol If the acknowledgment does not reach the sender before the packet times out, then the sender retransmits the packet (with the same number), as illustrated in Figure 2.17. The receiver gets the packets in order and needs only to keep track of the number (0 or 1) of the last correctly received packet to identify duplicates that may arrive when acknowledgments are lost or late. Figure 2.18 shows that the transmitter must number packets. In the situation of the figure, if the packets were not numbered, the receiver could not tell whether the second packet it receives is a copy of the first or a new packet.
- 53. Alternating Bit Protocol -Transmission Errors that corrupts the packets
- 54. Alternating Bit Protocol – Why the receiver must number the acknowledgments?
- 55. Alternating Bit Protocol - Why the receiver must number the acknowledgments? The figure assumes that the acknowledgments are not numbered and shows a sequence of events that lead the protocol to fail to retransmit a packet. The figure shows that the transmitter can be led to confuse the acknowledgment of a packet 0 with the acknowledgment of a packet 1. To prevent such confusion, it is necessary for the receiver to number the acknowledgments.
- 57. ABP Protocol The ABP protocol is inefficient if the propagation time is long compared with the transmission time, because the sender waits for the acknowledgment before sending the next packet. The efficiency of the Alternating Bit Protocol is defined as the fraction of time that the transmitter transmits new packets. When the bit error rate is negligible, the Efficiency =TRANS/(TRANS + 2PROP + ACK). TRANS denotes the time to transmit a packet, ACK the time to transmit an acknowledgment, and PROP the time taken for the signal to travel from the sender to the receiver and from the receiver to the sender. (We assume these two times to be equal.)
- 59. Go Back Ν -Protocol More efficient than ABP for long propagation times The network designer or user selects a window size N. Typically, Ν is just large enough so that the pipe is full: the sender gets the acknowledgment of the first packet when it finishes transmitting packet number N. The sender numbers the packets sequentially modulo Ν + 1, that is, 1, 2,..., Ν, 1, 2,..., N,.... The receiver acknowledges every correct packet that it receives with an acknowledgment that it numbers as the largest numbered packet that it has received so far. For instance, if the receiver gets the packets 1, 2, 4, 5, then it sends the acknowledgments .
- 60. Go Back Ν N=4
- 61. Go Back Ν Whenever the sender sends a packet, it starts a timer with a timeout Τ selected by the network designer. When the sender fails to receive the acknowledgment of a packet before the packet times out, the sender retransmits that packet and all the packets that it has transmitted since it last transmitted that packet. Therefore, the receiver need not store any packet since it eventually gets them correctly in sequence. The sender must be able to store up to Ν packets. Note that with this protocol the sender may have to retransmit packets that were correctly received by the receiver. TRANS denotes the time to transmit a packet, ACK the time to transmit an acknowledgment, and PROP the time taken for the signal to travel from sender to the receiver and from the receiver to the sender. Go Back Ν with Ν = 1 is exactly ABP.
- 62. Go back N
- 63. Selective Repeat Protocol SRP is similar to Go Back Ν but retransmits only packets that were not correctly received. A window size 2N is agreed on, and the packets and acknowledgments are numbered modulo 2N. Once again, the sender waits for the acknowledgment of packet n before sending packet n + Ν (modulo 2N). The sender retransmits packets that have not been acknowledged within a timeout after their transmission, and the receiver acknowledges all the correct packets. The sender and the receiver must both be able to store up to Ν packets. Figure 2.23 shows typical sequences of events when the nodes use SRP (with Ν = 4).
- 65. Flow Control Consider a source that is sending packets to a receiver. The receiver sets aside some memory space to store the incoming packets until it can process them, possibly by displaying them or storing them on a hard disk. When the memory gets full, the receiver should notify the source to stop transmitting. When the source uses a window mechanism, the receiver can indicate, in the acknowledgments, the maximum window size that it can store. The source then uses that value as a limit to its window size. As the receiver buffer becomes full, this maximum value decreases, which slows down the source and prevents overflowing the receiver memory.
- 66. Congestion Control The bit streams of many sources get multiplexed and read into a switch buffer. If the sources are unregulated, occasionally the buffer will get filled up, causing long delays and, possibly, buffer overflow and resulting packet loss. Congestion-control mechanisms can cause the sources or switches to reduce the number of packets they inject into the network when conditions of congestion are detected. If the sources use a window mechanism as in Go Back Ν or SRP, then this mechanism can be adapted to serve the purposes of congestion control. Notice that the sources can detect congestion or packet loss from the fact that their packets time out. Thus as soon as several timeouts occur in quick succession, the source should reduce the window size. This reduction automatically reduces the number of outstanding packets, and hence network congestion. There is a conflict between keeping the window size small to reduce congestion and keeping it large enough to "fill the pipe" to maintain high efficiency.
- 67. Congestion Control the delay-bandwidth product is large and the window size is large. Then most of the outstanding packets are propagating through the links, rather than waiting in the buffers. As a result, once congestion is detected and the window size is reduced, the many packets already in transit in the links are unaffected by the reduced window size. Those packets continue to arrive into the buffers, increasing congestion and packet loss. Thus window flow control is not effective when individual connections have a large bandwidth-delay product. In that case, "open loop" congestion control in the form of so-called rate control is more effective.
- 68. Resource Allocation-Bandwidth, buffers A circuit-switched network dedicates a fixed bandwidth to each connection, so it can guarantee minimum delay to an application whose peak rate is less than the fixed bandwidth. A datagram network is connectionless, and the network switches do not have the state information needed to distinguish packets from different appli cations. As a result, the network cannot dedicate resources that are specific to individual applications. Some crude resource allocation is possible, however. In virtual circuit-switched networks, each packet carries its virtual circuit identifier or VCI. This allows switches to treat packets differently based on their VCI. Thus, at the time its virtual circuit is set up, an application can negotiate with the network for certain service quality. The network switches can then reserve bandwidth and buffers for that virtual circuit so as to provide that quality. Guaranteed service quality is thus a possibility.
- 71. THANK YOU