These solutions are just provided as a guide and do not cover all possible correct answers.
Experiment 3A,B Discussion Questions:
1. Explain briefly why TCP woud be expected to perform more poorly than UDP over
wireless connections.
TCP interprets packet losses as congestion losses and reacts by reducing the TCP window. Since packet losses are common in wireless connections, this means the TCP window is smaller than optimal. For a multi-hop route this means a lower data rate since the TCP sender will wait for acks to arrive before advancing the sliding window. For a single-hop route (like the one in the experiment) TCP window size doesn't really have a great effect on data rate (since round trip times are very small) and so TCP just gets a data rate comparable to UDP but reduced by a constant overhead due to extra TCP ACK messages and the TCP packet overhead.
2. Discuss the pros and cons of using TCP in a wireless environment.
The main disadvantage is discussed above: increased overhead and inaccurate window size estimations due to misinterpretation of the cause of packet loss.
The main advantage is that TCP provides a reliable end-to-end transport layer regardless of the media over which the packets are flowing. Since a great many internet applications already use TCP at the transport layer and are not concerned with details of the link and physical layer, it is more convenient to keep the TCP semantics unchanged and just let it run over the wireless link than to try to optimize it for wireless.
3. Name one functionality at the MAC layer that reduces the impact of wireless on TCP.
Link layer retransmissions reduce the impact of wireless losses on TCP. In fact anything that improves the reliability of the wireless link will make it less confusing to TCP, so forward error correction and lower rate selection would also help reduce the impact of wireless on TCP (at the cost of increasing redundancy and thus reducing the data rate, so be aware that the tradeoff is not always beneficial).
Discussion Questions:
1. Comment on the data rate of A with and without RTS/CTS. When is A more sensitive
to its ability to hear the hidden terminal C? Why?
A gets a better data rate with RTS/CTS than without. A is more sensitive to its ability to hear C when RTS/CTS is turned off. The reason is because, when RTS/CTS is off, A must rely on its carrier sense to detect if C is sending a packet. If the loss between A and C is low, then A can easily avoid collisions because it can tell if C is sending. As the loss from C to A increases, A can no longer easily tell when C is transmitting and so its rate of collisions will increase. On the other hand if RTS/CTS is turned on, when A receives a CTS, it knows it can send without fear of collision with C (assuming C also received the CTS). Hence A no longer needs to rely on listening for a packet from C to avoid collisions. So, the rate of A with RTS/CTS turned on is relatively independent of the loss between C and A.
2. Consider A’s bitrate in the “half-hidden” terminal scenario when A could hear C clearly
but C could not hear A. Since A can hear C, it should already be able to avoid most
collisions easily using carrier sense. Why then does RTS/CTS improve A’s data rate in
this situation? What would be a better solution for this functionality?
RTS/CTS improves A's data rate in the half-hidden scenario because the CTS causes C to suppress its transmission. Without RTS/CTS, C will always send at close to its maximum rate because C cannot hear the transmissions from A (C can only hear the much smaller ACK messages from B to A for the few infrequent messages that A actually gets through). In essence RTS/CTS is thus acting as a fairness mechanism to improve A's data rate by expicitly instructing C to reduce its rate. A better solution for this functionality is to use the 802.11 point coordination function (PCF) at B. Essentially, B becomes an AP which then polls A and C respectively to request traffic in the contention-free period. This would ensure a much fairer allocation of traffic.
1. What is the approximate difference in power between signals such that the stronger
signal experiences little to no packet loss?
A difference of approximately 10Db suffices to ensure that the stronger signal almost completely gets through.
2. Comment on the implications of capture on spatial re-use and the network topology of
dense networks with high data rates. How would this phenomenon change power level
settings in dense networks? What about the MAC Clear Channel Assessment thresholds?
In a dense network, nodes that are close together can still receive transmissions from each other regardless of simultaneous communications from more distant nodes. This implies that a network topology where each node only communicates with neighbors within a small range is preferable since it maximizes spatial re-use. Hence, in terms of transmission power management, this means that a uniformly lower transmission power is preferred since messages only need to traverse short distances (a high transmission power setting for all nodes just increases power consumption with no clear benefit since both the useful signal and the overall noise will increase). Clear channel assessment thresholds should be set high enough that nodes will attempt transmissions to their close neighbors even though other simultaneous transmissions may be occurring.