Changes between Version 28 and Version 29 of GEC12WiMaxDeploymentAndExperimentation

11/04/11 12:49:51 (8 years ago)



  • GEC12WiMaxDeploymentAndExperimentation

    v28 v29  
    269269=== 4.2)  Throughput Experiment Using iperf ===
    2702702:15pm   [[BR]]
    271 Manu Gosain (GPO) [[BR]]
    273 NOTE:  Only 2 slides!  [[BR]]
     271Manu Gosain (GPO) and Harry Mussman (GPO) [[BR]]
    275273Overview of basic throughput experiment, using bidirectional iperf, both TCP and UDP  [[BR]]
     275The basic throughput experiments we have done utilize iperf, both TCP and UDP. [[BR]]
     276  In iperf, the data is generated in the client, and flows to the server. [[BR]]
     277  We put the client in the Mobile Station, and the server in the Base Station. [[BR]]
     278  We used -d dualtest (bidirectional mode), where test is initiated at the client, data begins to flow to the server, and then a second data flow starts at the server;  at the end of the test, results are available at the client. [[BR]]
     279  Because of this, all tests can be initiated at the Mobile Station, and then results are available there. [[BR]]
     281iperf results in the TCP mode depend upon buffer sizes;  overall delay;  and lost packets.   [[BR]]
     282  Because of wireless propagation conditions, lost packets are common, and slight changes can significantly affect the measured throughput. [[BR]]
     283  Thus, TCP results are highly variable;  we took multiple measurements at each point, and identified the best and worst results. [[BR]]
     284  It would certainly be good to have a better way to evaluate available channel bandwidth. [[BR]]
     285  On the other hand, most apps use TCP and the variable results are typical of how these apps would see the channel. [[BR]]
     287iperf results in the UDP mode, counts  % packets received, for a given (fixed) transmit rate. [[BR]]
     288  If the rate is set below the available bandwidth, typically 100% of the packets are received. [[BR]]
     289  If the rate is set above the available bandwidth, typically % of the packets are received typically equals available bandwidth divided by offered bandwidth, but there is no way to understand how many packets are actually lost. [[BR]]
     290  An extended test that ramped up the offered bandwidth in multiple tests, could actually verify the available bandwidth;  this could then be repeated to see real variations in available bandwidth. [[BR]]
    276292References:  [[BR]]
    277   {  iperf tutorial][[BR]]
     293  [  iperf tutorial slides]  [[BR]]
     294  [  iperf tutorial web site][[BR]]
    279296=== 4.3)  Throughput Experiment Using Bit Torrent ===