EECS 600 Advanced Computer Networks [Data] Mar 14, 2005 Monday [Notes] Dan Liu (dxl77@cwru.edu) [Paper] G. Holland and N. Vaidya. Analysis of TCP performance over mobile ad hoc networks. In Proc. of ACM MobiCom, 1999. 1.Problems of TCP performance in wireless networks & modile Ad Hoc networks In Wireless networks: signal weakness, interference problems In Modile Ad Hoc networks: packet losses and corruption caused by wireless induced errors, and link failures due to mobility. 2.The network framework of the simulation in this paper Transport Layer: TCP Reno(Transmission Control Protocol Reno) Network Layer: DSR(Dynamic Source Routing) Link Layer: ARP(address resolution protocol)(Map IP addresses to MAC addresses) 3.Figure 1 shows the TCP Reno throughput as a function of the number of hops. (a)Is it similar to that relationship between TCP throughput and RTT? (b)Why dose the throughput decrease when the number of the hops increase? Since when the number of hops increases, there will be more chance of transmission collision or interference between neighbours (c)So maybe Reno is not a good choice for mobile Ad hoc networks? Better to use some pipeline pattern of TCP? 4. When introduction tha reasonable the expected throughput, why not use other topology such as random distributed structure but just use chained pattern which may never achieved in real application? It is in order to get an upper bound values for the expected throughput. However, is there any other proper which can get a tighter upperbound? 5.In Figure 2(b), the pattern number as the x label dose not mean anything. These values have been arranged in increasing order of the throughput. 6. Explain the phenominon in Figure 4. Shown in Figure 4(b)and 4(d), when the speed of the mobile nodes increase, the throughput drops to 0 for increasing time. This is because even when a new route is found after the detecting of the old route stale caused by the moving of the node, it may still become stale before the new packets arrive at the destination node due to the quick movement of that node. However just like in 4(c) there may be chance for the new detected route being usful before the node moved again. 7. How about turning off the replying from the caches? See Figure 6, for a single TCP connection in a network with no other data traffic, it has a startling improvement in performance. However, in a network with multiple data source, the additional routing traffic introduced when replies from caches are not used could degrade performance. 8. Who send the ELFN?