!"#$%"&'()(*"+,-."/(0,.1$+./$2(3(4*4(56)7(89++7(:6))( 1.

Suppose that a certain communications protocol stack involves an overhead on the application layer PDU of 100 bytes for headers and framing. We send 1 million bytes of data using this application and this protocol stack; however, exactly one data byte is corrupted during transmission, and the packet containing it is thus lost and must be retransmitted. (a) What is the total number of overhead + loss bytes retransmitted, for packet data sizes (not including overhead) of 100, 500, 1000, and 2000 bytes? (b) What packet size would be optimal for this system, i.e require least number of bytes to be transmitted overall? 1. (a) The actual 1 million bytes to be transferred has to be transmitted once for any packet size, so that does not depend on p. The byte overhead dependent on p is as follows. For any packet size p, there are 106 / p packets to transmit, thus a total of these many headers, each 100 bytes. There is also the p + 100 bytes of the single retransmitted packet. Thus the total number of (overhead + retransmission) bytes transferred during the communication are: For a packet size of 100, 1000000 + 200 = 1000200 bytes For a packet size of 500, 200000 + 600 = 200600 bytes For a packet size of 1000, 100000 + 1100 = 101100 bytes For a packet size of 2000, 50000 + 2100 = 52100 bytes (b) Among the given packet sizes, 2000 bytes is clearly the most beneficial one; so far, the larger the better. But this trend would reverse at some point, when the overhead to retransmit one packet (very large size) overcomes all the header overhead on all the other packets. If packet sizes may be arbitrary, we can simply express the overhead as 108/p + p + 100. You can differentiate this expression w.r.t. p, and find the optimal p to be 104 . (Alternatively, since the terms are inversely and directly proportional to p, we can apply a well-known result and conclude that the minimum of the sum occurs when they are equal. Thus the size of 10000 is the best among all possible sizes, for the conditions given.) 2. How long (in meters) was a bit in the original 802.3 standard? Use a transmission speed of 10 Mbps and assume the propagation speed in coax is 2/3 the speed of light in vacuum. 2. The speed of light in coax is about 200,000 km/sec, which is 200 meters/sec. At 10 Mbps, it takes 0.1 sec to transmit a bit. Thus, the bit lasts 0.1 sec in time, during which it propagates 20 meters. Thus, a bit is 20 meters long here. 3. Consider a slotted TDM hierarchical network in which there are 8 computers sharing a 10 Mbps channel to a router R1, sending packets to R1 which must all be forwarded by R1 over an outgoing link (called the MAN link), as shown in Figure 1. Computers A-H always have packets waiting to be sent, so no timeslot on the shared LAN is idle. Ignore the propagation delay on the LAN. The MAN link outgoing from R1 is a 100 Mbps link.

The network layer protocol in use has a PDU with exactly 40 bytes of header and exactly 200 bytes of payload (SDU). The DLC layer protocol in use on the LAN has a PDU with exactly 40 bytes of header and 20 bytes of trailer. The DLC layer protocol in use on the outgoing link from R1 has a PDU with 40 bytes of header and no trailer. The physical layer does not impose any bit overhead in either channel. Use 1 Mbps = 1000000 bps.

C D E

A R1 H G

R2

Figure 1

(a) What is the transmission delay of a bit for Computer A? (b) What is the transmission delay of a DLC PDU for Computer A? (c) The TDM slotting of Computers A-H is at the individual PDU level, i.e. a PDU transmission from Computer A on the LAN is followed by one from Computer B, and so on. In the LAN, each network layer PDU is encapsulated in one DLC PDU. If the outgoing link from R1 follows the same discipline, what is the width of a DLC PDU on the MAN link? (The transmission time of a bit or group of bits is sometimes referred to as their width on the line.) (d) R forwards each packet as soon as it is completely received. Ignore processing and queueing time at R1, if any. What is the delay between the instant when a packet first starts transmission at Computer A and the instant when the same packet starts transmission at R1? (e) What is the amount of time between two successive PDU transmissions are seen on the MAN link? (f) What is the bitrate perceived by the network layer of Computer A? Answer by filling up a table like the following. You may include additional reasoning etc., but they must be separate from the table providing the final answer. Answers should be in natural units (bits, Mb, bps, Mbps, seconds). If your answer is not a whole number, answer in decimal form with enough decimal places to preserve significance, but not less than two places. 3. Part (a) is obtained simply by considering the definition of bitrate on the LAN; 10000000 bits/sec is the same as 10-7 or 1e-7 seconds per bit. For Part (b), combine this answer to the size of the PDU as obtained from the overhead information. The PDU transmitted on the LAN link is 300 bytes = 2400 bits, for a transmission time of 2.4e-4 seconds. If a single LAN PDU becomes a single MAN PDU, the MAN PDU is 280 bytes = 2240 bits, and with the higher transmission speed of the MAN, the transmission delay of the MAN PDU is 2.24e-5 seconds. For Part (d), since both the propagation delay over

the LAN and the processing delay at R1 is assumed to be zero, the delay between a packet starting transmission at A and the same packet starting the forwarding transmission at R1 is simply the transmission delay of A, and the answer is the same as Part (b). Part (d) points up the fact that the MAN link is idle most of the time to forward a PDU, R1 needs only about one-tenth the time it takes to receive it, and the rest of the time there is no traffic which can utilize the link. (In reality, R1 would probably serve several 10 Mbps LANs and not just one as in this example, so the larger bitrate would be useful.) The successive transmissions on the MAN are thus exactly as far apart as the successive transmissions on the LAN, i.e. the width of a LAN PDU, i.e. same as Part (b) again. Finally, the network layer at computer A transmits 240 byte PDUs, and gets to transmit one of them once every eight slots of 2.4e-4 seconds, for a perceived bitrate of 1 Mbps. (a) 1e-7 sec (b) 2.4e-4 sec (c) 2.24e-5 sec (d) 2.4e-4 sec (e) 2.4e-4 sec (f) 1 Mbps

4. (a) (Tanenbaum, 1.1) Imagine you have trained your St. Bernard, Bernie, to carry a box of three 8mm tapes instead of a flask of brandy. (When your disk fills up, you consider that an emergency.) These tapes each contain 7 gigabytes. The dog can travel to your side, wherever you may be, at 18 km/hour. For what range of distances does Bernie have a higher data rate than a transmission line whose data rate (excluding overhead) is 150 Mbps? (b) (Ungraded) In the lecture we distinguished between two types of delay: Propagation Delay: The time from when the first bit of the message is sent to when the first bit of the message is received. Often called simply delay. Transmission Delay: The time from when the first bit is sent (received) to when the last bit is sent (received). This can be considered as the ratio of the time taken to the number of bits in the message. Often the reciprocal of this ratio is considered very important, and is called bit rate, or throughput. In the problem in part (a) of this question, which of the above two quantities are you being asked to compare, or what combination of them? 4. (a) Suppose that the maximal distance is x km, then the dog will take x/18 hours to reach me. Then we have: 3 * 7 * 109(bytes) * 8(bits/byte) ---------------------------------------150 * 106(bits/sec) i.e., 0 " x " 5.6 km (b) In part (a) I am being asked to compare the Propagation Delay for the dog, the time that Bernie takes to reach me (right hand side of Equation (1)), to the Transmission Delay for the EM transmission, the time from when the first bit is sent to when the last bit is sent (left hand side of Equation(1)). The propagation delay for the EM transmission is

! 3600 * x/18(sec) (1)

not given, so we cannot make a complete comparison. If we try to apply the concepts of transmission delay and propagation delay separately to the dog, then we are likely to conclude that the entire delay calculated by the RHS of Equation (1) is the propagation delay, and the transmission delay is zero (because all bits start off at the same time). Yet this is not quite true, because Bernie can carry only so many tapes at the same time. There is also the matter of putting them on Bernies collar and taking them off. In other words, this analogy, like all others, turns false when pursued too far. 5. A protocol uses hierarchical TDM techniques for various levels of networks. For the purpose of this question, assume header/trailer overhead to be zero, use the different DLC frame sizes as given. In this question we consider only DLC frames. At the level of Local Area Networks (LANs) containing each individual computer, the links operate at 10 Mbps and the frames (L-frames) are each 1500 bytes long. The LANs link to Metropolitan Area Networks (MANs) which use 15000 byte M-frames that are made by multiplexing 10 L-frames, one each from 10 different LANs. Every eight of these MANs feed into a Regional Area Network (RAN) that uses 1 Gbps links and uses a R-frame, and eight RANs feed into a Wide Area Network (WAN) using a W-frame, which is 16 times the size of an R-frame. (a) The transmission time of a bit or group of bits is sometimes referred to as their width on the line. What is the width of an L-frame on a LAN link? (b) How wide is an L-frame on a RAN link? (c) What is the ratio (e.g. 2:1) of TDM of L-frames into R-frames? (d) If the MAN links are exactly fast enough to allow M-frames to be sent at a rate that does not require any of the LANs to slow down (or cause a buffer to build up at the intermediate node connecting the LANs to the MAN) what is the bandwidth of the MAN links? (e) If the RAN transmits one R-frame in the same amount of time that a LAN transmits one L-frame, what fraction, if any, of the R-frame bits is wasted? (f) If a WAN link is 8 times faster than a RAN link, what is the TDM ratio of R-frames into W-frames? (g) What fraction, if any, of the W-frame bits is wasted? 5. (a) The width of the L-frame on the LAN link is 1500 bytes/10Mbps = 1.2 ms. (b) As before, 1500 bytes/1 Gbps = 12 #s. (c) Since 10 LANs feed into 1 MAN, the consolidation ratio there is 10:1. Then 8 MANs feed into the RAN, for a total consolidation ratio of 80:1. Note: If you assume that the TDM ratio refers to how many L-frame can go into the Wframe (not how many do go in), that is a perfectly good answer as well, as long as your answer makes that clear. (d) In the time it takes to transmit one L-frame on a LAN link, the MAN must transmit 1 M-frame (i.e., 10 L-frames) on the MAN link, if it is to keep up with the LANs. Thus the bit rate must be 10 times as much, or 100 Mbps. (e) At 1 Gbps, a RAN link has 100 times the bit rate of a LAN link. If a R-frame is as wide on the RAN link as an L-frame is on a LAN link, it must have 100 times as many bits. In other words, a R-frame can hold 100 L-frames. But we already know that only

80 L-frames are actually being fed into a R-frame. Thus 20% of the bits of the R-frame are not being used. (f) During the width of an R-frame, 8 R-frames arrive at the WAN. Since the speed of the WAN link is 8 times that of the RAN link, there is no slack in the WAN link, i.e. at all times the WAN link has to be transmitting R-frames (encapsulated in W-frames), otherwise it will fall behind. But since the W-frame is 16 times the length of one Rframe, the intermediate node connecting the RANs to a WAN takes twice as long to transmit one W-frame as to receive one R-frame from each of the contributing RANs, thus 16 R-frames will have built up in the buffer (2 from each contributing RAN) while one W-frame is transmitted. So the TDM ratio is 16:1. (g) Now this becomes a matter of interpretation. Looked at one way, the W-frame is nothing but a combination of 16 R-frames, each of which has 20% wastage, so the answer is the same 20% for the W-frame. But looked at another way, the W-frame is solidly packed with 16 R-frames, so no bits are being wasted. In the second point of view, the interpretation is that the W-frame is being fully utilized to carry encapsulated R-frames; what is inside the R-frame is not relevant to judging the utilization of the WAN. 6. End stations A-H are served by a router R1 as shown in Figure 1. Assume the link from R1 to R2 has 80 Mbps capacity. Stations A-H each get equal time slots of 1 ms each, Time slot 0 for A, 1 for B, etc., repeating after 8 timeslots. The link scheduling by R1 is as follows - after packet is completely received, it is sent at next slot for that station, if there is room in that slot. Assume a packet can be sent in a slot even if it arrives at a time partway through the slot, if enough time remains in the slot. No pre-emption is allowed if a packet would not finish transmission in the current slot completely, it is not transmitted and is held for the next slot. Packets come to R1 from the end stations, requiring transmission on the link to R2, as shown in the table below. The Time column lists instants at which the packet reception at R1 completes. The link(s) from Stations A H to R1 are not relevant to this question they can be assumed to have sufficient throughput capacity to allow the stations to send the packets at the times stated. Assume that a timeslot assigned to A begins exactly at 0 ms. 1 Mbps = 1000000 bps, 1 KB = 1000 bytes. Time (ms) 0 0.5 1.5 2.1 3.5 4.2 4.9 5.0 6.1 6.9 Station Packet size (KB) A B D C A A E F F G 5 5 10 5 5 5 5 10 10 5

(a) State, for each packet, when it is forwarded by R1, and what delay it experienced at R1. Answer by stating the time when the forwarding transmission starts for the packet. The delay should be counted from the time the packet arrived at R1 (completely received) till it starts transmission by R1. (b) Repeat the same exercise, but this time assume that there are no slot reservations, instead the link is statistically multiplexed by R1. In this case, there are no slot boundaries that must be respected. 6. Below, the timeline of arrivals and transmissions is shown. From this it should be trivial to work out the delays experienced. The first diagram represents the timeline for slotted multiplexing.
B C F F

0 1 2 3 4 5 6 7 8 9 10 from E and G, and the second one from F, have to wait until the next set of The packets
slots. Next, the statistical multiplexing case:
B C F F

0 1 2 3 4 5 6 7 8 9 10 The statistical multiplexing gain is quite clear, but some unlucky packets (F in the above)
may suffer a higher delay. 7. The performance of a client-server system is strongly influenced by two major network characteristics: the bitrate capacity of the network, and the latency. Given an

example of a network that exhibits high bitrate but also high latency. Then given and example of one that has both low bitrate and low latency. 7. A transcontinental fiber link might have many gigabits/sec of bandwidth, but the latency will also be high due to the speed of light propagation over thousands of kilometers. In contrast, a 56-kbps modem calling a computer in the same building has low bandwidth and low latency; less extremely, a WiFi link has 11 Mbps capacity but very low latency. 8. Consider the following network to carry data from Computer A to Computer B. A N1 N2 B

The propagation delay on each of the three links is 1 ms, and each is a 100 Mbps link. A transmits 100 byte network layer PDUs to B. The DLC in use at all nodes adds a 25 byte overhead. The communication between DLC and network layers is via a shared memory which can be read from and written to at upto 1 Gbps. The DLC software must completely read a packet from shared memory before transmitting it on the bitpipe, similarly the network layer software must completely read a packet from shared memory before forwarding it, and each layer must wait until a packet is completely available in shared memory before starting to read it. Draw a timing diagram to show the progress of a single PDU originating at the network layer of A and destined for the network layer of B, under the following cases. In each case, a template for the timing diagram is shown below your solution should be in the same form. (Remember, the submission must be an electronic document, and it is strongly preferable that you use a word processor or other application to directly generate your diagram in the document, rather than scanning in a hand-drawn diagram). You need not maintain scale exactly, but clearly show the distinction between significantly different time periods. During the period under consideration, the system is idle other than the single PDU we consider. (a) N1 and N2 are routers, and forward packets at layer 3. The store-and-forward process at the network layer software takes 5 ms, for each of N1 and N2. Other characteristics are all same as A and B. (b) N1 and N2 are simply electrical junction wires. 8. The first thing that happens to the packet is being copied to the L2/L3 buffer by L3. This takes 100 bytes / 1 Gbps = 800 / 1e9 = 0.8e-6 seconds. We can reasonably assume that the time taken for a bit to travel from the L3 memory to the L2/L3 buffer is negligible; i.e. the propagation delay between the two memories is zero. Similarly it takes 0.8e-6 sec for the DLC to copy it from the intermediate buffer. Once that copy is complete, the DLC starts transmitting the DLC PDU, which is 125 bytes. Since nothing is mentioned about physical layer overhead, we reasonably assume that it does not add

any overhead. At 100 Mbps, the transmission delay for the PDU is 125 bytes / 100 Mbps = 1e-5 seconds. The propagation delay is given to be 1 ms. These intervals appear again at B. In addition, for case A, the same delays appear at N1 and N2, and there is a processing delay of 5 ms at the network layer of N1 and N2. In the diagrams below, it is impossible to show the microseconds and milliseconds to scale and still have the diagram be meaningful as indicated in the question, a general indication of magnitude has been attempted.

(a) The timing diagram is: L3, A L2/L3 mem, A L2, A

L2-1, N1 L2/L3 mem, N1 L3, N1 L2/L3 mem, N1 L2-2, N1

L2-1, N2 L2/L3 mem, N2 L3, N2 L2/L3 mem, N2 L2-2, N2

L2, B L2/L3 mem, B L3, B

time (microsecond, millisecond)

0.8 0.8

10 0.8 0.8 5

0.8 0.8

10

0.8 0.8 5

0.8 0.8

10 0.8 0.8

(b) Now the three links between A and B are joined into one large physical link that the bits propagate over (we have assumed that the propagation delay through the wire junctions N1 and N2 can be neglected compared to the other delays.

L3, A L2/L3 mem, A L2, A

N1

N2

L2, B L2/L3 mem, B L3, B

time (microsecond, millisecond)

0.8 0.8

10

0.8 0.8

9. (Ungraded) Consider the networks that enable any computer on the NCSU Campus to communicate with any other computer. By visiting the websites of NCSU Communication Technologies, and any other sources, gather as much information you can about the architecture and topology of all the networks concerned in such communication. (Try not to bother too many actual people.) Pay some attention to

physical layout details as well as topology: for example, if there are some Ethernet switches in EB-I, where are they located? 9. Some idea of the topology and the physical layout can be obtained from ComTechs presentations, e.g. that are available through the URL http://comtech.ncsu.edu/networking/presentations/reliable_networks.php (click through to the PDF), and other accompanying pages. Briefly, each building is served by a hierarchical system of Ethernet switches, the lowest level being in individual offices or labs, and the highest level of switches being in a communication room (closed to the public), perhaps in the basement. Cables connect these switches in star formation by point-to-point Ethernets to the campus backbone which is composed of a few routers. Typically the number of routers between any two computers on campus is likely to be one, or maybe two. Note: Interesting problems to look at out of Chapter 1 of Tanenbaum: 4, 8, 9, 15, 20.