Trabalho Prático de Redes de Computadores

ECE 6110 Lab Assignment 2 Congestion Control In this lab you will create a topology as specified below and compare and contrast the performances of the TCP CUBIC and NewReno congestion control algorithms This lab builds on the discussions in class of the TCP CUBIC paper which can be found at httpsbloughecegatechedu6110tcpcubicpaperpdf You can find example code for most of what is needed for this assignment in the following program NSDIRexamplestcptcpvariantscomparisoncc where NSDIR stands for the toplevel directory in which you installed ns3 The fifthcc program from the ns3 tutorial also introduces the concept of congestion window tracing although it uses a different sending application from this assignment which has a different method for tracing the congestion window changes Therefore I recommend the tcpvariantscomparisoncc program as a starting point However to produce results matching the instructors code you should remove the following lines from tcpvariantscomparisoncc ConfigSetDefault ns3TcpSocketRcvBufSize UintegerValue 1 21 ConfigSetDefault ns3TcpSocketSndBufSize UintegerValue 1 21 ConfigSetDefault ns3TcpSocketBaseSack BooleanValue sack ConfigSetDefault ns3TcpL4ProtocolRecoveryType TypeIdValue TypeIdLookupByName recovery and use the error model in fifthcc PtrRateErrorModel em CreateObjectRateErrorModel emSetAttribute ErrorRate DoubleValue errorRate instead of the one in tcpvariantscc The model above defaults to bytelevel errors while the one in tcp variants sets the error unit to packets Turnin Instructions Create a separate folder for each part of the assignment inside a main folder named Lab2yourlastnameyourfirstname Then create a single tarball of the main folder and all of its subfolders and submit it in Canvas Important programming instructions All code must be hand typed in your editor of choice No auto formatting is allowed to change the text that you type in manually Also make sure to use the exact program name and commandline parameter names specified in each part You will get points deducted if you do not follow these instructions even if your code is functionally correct Part 1 Topology with a Bottleneck Link and Homogeneous Flows Using what you learned from Lab 1 create the depicted topology with 4 nodes and 3 pointtopoint links Set the data rates and delays on the link from the source and the link to the destination as shown Make the data rate delay and error rate on the bottleneck link commandline parameters in your program Also include commandline parameters for the number of flows that will be sent from source to destination maximum of 20 and the protocol to use TcpCubic or TcpNewReno Name your program lab2 Bottleneck link 100 Mbps 001 ms 100 Mbps 001 ms source dest part1cc and name the command line parameters dataRate delay errorRate nFlows and transportprot Using the BulkSend application from tcpvariantscomparisoncc set up nFlows TCP flows from the source node to the destination node set the parameter datambytes to 0 for each flow to make the flow send data as fast as possible for the duration of the simulation The bottleneck link in this topology represents the Internet Set the default parameters for the bottleneck link to be 1 Mbps data rate 20 msec delay and an error rate of 000001 Your program should output congestion window changes during the simulation and the goodput achieved by each flow at the end of the simulation1 Set the simulation duration to 20 seconds as is done in fifthcc start the sink applications at time 0 and start all the flows at time 1 second Part1a For a bottleneck link data rate of 10 Mbps delay of 100 msec and error rate of 000001 and with one flow plot the congestion window size of TCP with CUBIC and TCP with NewReno Compare and contrast these two plots Also comment on the shape of the CUBIC plot Do you see some behaviors that were discussed in class If so mention what those are and point them out in the plot Also report the goodputs achieved by CUBIC and NewReno in this example and compare them using what you learned from the CUBIC paper Part1b For a bottleneck link data rate of 1 Mbps and error rate of 000001 and for 1 2 and 4 flows plot the aggregate goodput across all flows vs bottleneck link delay for delays of 50 to 300 msec in steps of 50 and for both TCP CUBIC and TCP NewReno Compare and contrast the CUBIC and NewReno goodput behaviors There should be 6 line plots on your graph aggregate goodput for 1 2 and 4 flows for CUBIC and the same for NewReno Part1c For a bottleneck link data rate of 1 Mbps and delay of 1 msec and for 1 2 and 4 flows plot the aggregate goodput across all flows vs bottleneck link error rate for error rates of 000001 000005 00001 00005 and 0001 for both TCP CUBIC and TCP NewReno Compare and contrast the CUBIC and NewReno goodput behaviors There should be 6 line plots on your graph aggregate goodput for 1 2 and 4 flows for CUBIC and the same for NewReno Turnin instructions In your Part 1 subfolder include the source code of your program with the given topology all plots specified above and your discussion about each set of results as called for above Also include screen shots and text files of two sample outputs one for CUBIC and one for NewReno with 4 flows and the link data rate delay and error rate values from Part 1a Part 2 Topology with a Bottleneck Link and Heterogeneous Flows In this part the nFlows TCP flows you create will be split equally across the two destinations which have very different delays on their last links Assume the number of flows is always an even number This will result in the RTTs of the two sets of flows being quite different Name your program lab2part2cc 1 Goodput refers to the amount of data received at the packet sink application divided by the time that the flow was active 19 seconds given the start and stop times specified above Report the goodput in bits per second Bottleneck link 100 Mbps 001 ms 100 Mbps 001 ms source dest1 dest2 100 Mbps 50 ms and have the same commandline parameters as in Part 1 Run the following experiments and plot the results For nFlows 2 4 6 and 8 and for a bottleneck link data rate of 1 Mbps a bottleneck link delay of 20 msec and an error rate of 000001 calculate the average goodput of the flows to dest1 and the average goodput of flows to dest2 for both TCP CUBIC and TCP NewReno To ensure you get representative behavior run this experiment 10 times for each value of nFlows and each algorithm and average the results across all runs Here you should generate a different sequence of random numbers for each run2 Your graph should show average goodputs for dest1 and dest2 flows versus nFlows for the values of nFlows specified and for both CUBIC and NewReno 4 line plots with 4 data points each Comment on the RTT fairness of the two algorithms and how the results compare to the goodputs you observed in Part 1 All parameters not specified here should be the same as in Part 1 Turnin instructions In your Part 2 subfolder include the source code of your program with the modified topology the goodputs vs nFlows graph and your discussion about goodputs and RTT fairness Also include screen shots and text files of two sample outputs one for CUBIC and one for NewReno with 4 flows the default RNG seed 123456789 and other parameters as specified above 2 You can either do this by having a different random number seed each time you run your program eg with the following line of code at the beginning of your program RngSeedManagerSetSeedtimeNULL or you can advance the run number after each run using RngSeedManagerSetRun as described in the ns3 documentation on Random Variables