We have 2 nodes and one 1 access point in a wireless network. Nodes 1 and 2 work in a single-hop fashion and only send data to the access point.
Each TCP connection works on an FTP application. Both connections begin with 1500 Byte packet sizes. We ran simulations where both use fixed packet sizes and others where both have a random packet size, changing each second. The simulation time is 150 seconds.
The buffer determines how many packets can remain in the queue before they are transmitted, If a buffer is full, no other packets can be added. Depending on the buffer management scheme, certain packets will be dropped, such as the last one entered in a DropTail scheme.
With a higher buffer size, fewer packets get dropped, but more packets will be waiting to be transmitted, so the delay increases. If there are fewer drops, more packets will arrive at their destination in less time. If there are buffers that are small and drops do not occur too often, there will be higher throughput due to less waiting time.
In our simulations, we noticed that buffer sizes of at least 20 experience no packet drops. As we decrease the buffer size, the throughput increases as well as the delay until around buffer size of around 16. Then both throughput and delay decrease at a linear rate until we reach a very small buffer size of 3. We believe that the higher delays around 16 buffer size occurs because of the combination of waiting time and re-transmissions. Smaller buffer sizes than 16 have more transmissions due to packet drops, but the waiting time is much less, so it would seem that the waiting time has a greater affect on the total delay than re-transmitting.
In class, Basem explained that the decrease in delay and throughput after buffer size 16 was due to the link being saturated. This make sense as the maximum throughput is reached at 16 buffer size. Then as the buffer size is increased, we see less delay since there are fewer re-transmissions, which also causes less throughput.
-
Smaller buffers give less delay, but with a higher drop ratio.
-
We wanted to use other queue management solutions, however very few would work, or give confusing results. For example, using JoBS throws an error when trying to run. Unfortunately, we could not find any helpful documentation on how to use it. We also tried using CoDel, but we kept getting an compilation error. It would seem that Debian's copy of NS2 does not have it. Some queues would only give the same results as DropTail. others would compile, but not output anything in the trace file.