-
Notifications
You must be signed in to change notification settings - Fork 1
Results and Further old
Prajval M edited this page Dec 10, 2018
·
1 revision
Comparison for Fq-PIE and PIE has been done in ns3 using a dumbbell topology with bottleneck as follows:
10Mb/s, 2ms 10Mb/s, 4ms
n0--------------| |---------------n4
| 1.5Mbps, 20ms |
n2------------------n3
10Mb/s, 3ms | QueueLimit = 100 | 10Mb/s, 5ms
n1--------------| |---------------n5
The same test case was used to simulate the Fq-PIE and PIE so as to get more truthful result and the cases were run many times to get a accurate value.
All the ClientHelpers(n0, n1) with TCP sockets installed
n0 with UDP and n1 with TCP. In this PIE failed miserably and due to not having enough size in the queue it dropped a lot of packets. Also this caused an error in PIE simulation and this case was suppressed to get the results.
-
Test Suite:
- PASS : 251
- SKIP : 3
- FAIL : 0
- CRASHED : 0
- Valgrind Errors : 0
- Only when there are TCP Packets in the network:
Drops | PIE | FQ-PIE |
---|---|---|
Random Drops | 204 | 0 |
Queue limit Drops | 0 | 0 |
- When there are both TCP and UDP packets in the network:
Drops | PIE | FQ-PIE |
---|---|---|
Random Drops | 3512 | 483 |
Queue limit Drops | 454 | 0 |
- Fq-PIE is a drastic increment over the already existing problem and handles the buffer bloat problem really well.
- As comparision of Fq-PIE with other AQMs is not done it is currently not possible to compare its performance with others.
- These results are only representative of how better FqPie is with respect to Pie.
- Comparison with FqCodel and Codel is yet to be done as a test-suite could not be prepared.
- CAKE implementation and comparison with FqPIE
- Only Basic FQ-PIE has been implemented. A enhanced version can be used for better results. Enhanced PIE