This test is similar to the previous test, except the network congestion is relieved manually about three-fourths of the way through the transfer.
XUDP received approximately half of the parcels generated by the sending side (table 8.11, with the majority lost during the period of 13 seconds to 28 seconds. The transfer was able to recover nicely once the network congestion was relieved, evidenced by the increased slope and point density in the upper right of the graph in Figure 8.29.
|Total Data Acknowledged||4033.607 KBytes|
|Total Transmission Time||33.206 seconds|
|Network Bandwidth Utilization||121.472 KBytes/second|
|Average Round Trip Time||50.1 milliseconds|
|Average Window Size||16.3 packets|
|Total Parcels Received||265 parcels|
|Total Transmission Time||32.219 seconds|
|Total Parcels Skipped||316 parcels|
|Network Bandwidth Utilization (14.4 KBytes/parcel)||118.439 KBytes/second|
|Average Parcel Reception Frequency||8.225 parcels/second|
Figure 8.30 shows the plateau in bytes received during the period of time the network was congested.
The congestion window (figure 8.32), interestingly did not adjust itself once network congestion had been relieved. The activity from 31 seconds until 35 seconds is due to residual acknowledgments remaining in XUDP's queues (acknowledgments of XUDP command parcels, primarily).
TCP held close to a 200KBytes/second bandwidth for the duration of the transmission (table 8.12). Note that the slope of the parcel received curve (figure 8.33 rises towards the end of the connection, indicating TCP's response to the increase in available bandwidth.
|Total Parcels Received||581 parcels|
|Total Transmission Time||42.571 seconds|
|Total Parcels Skipped||0 parcels|
|Network Bandwidth Utilization (14.4 KBytes/parcel)||196.528 KBytes/second|
|Average Parcel Reception Frequency||13.648 parcels/second|
This example is by far the best in this suite at illustrating the functionality of XUDP's timed-obsolescence algorithm. XUDP's stream is finished in just over 32 seconds, while TCP's stream isn't finished until the 42 second mark - a ten second difference. A XUDP application could have presented the XUDP stream unbuffered with satisfactory results, while the TCP's unbuffered stream would, on average, get presented 25% slower!