Testing between Korea and Japan
From Wiki
We conducted TCP testing between Korea and Japan.
Contents |
[edit]
ToDo
- TCP Friendliness index and link utilization graph
[edit]
Sites
- We would like to thank Leading Technology Research Department team in the Future Technology Laboratory at Korea Telecom (KT) who generously supported all listed testing machines and KOREN research backbone (1Gbps) links for this work. We would also like to Kyushu Research Center for providing an access to the server in Japan.
- The experimental results between Korea/Japan and USA will be available soon.
- Linux TCP sender - (Kernel 2.6.19), Future Technology Laboratory, Seoul, Korea
- Linux TCP receiver - (Kernel 2.4.20), Daejon, Korea (6ms)
- Linux TCP receiver - (Kernel 2.4.20), Busan, Korea (9ms)
- Linux TCP receiver - (Kernel 2.4.20), Kyushu Research Center, Kitakyushu, Japan (46ms)
- Using iperf, the Linux TCP sender sends one highspeed flow with one TCP-SACK flow to the other receivers.
[edit]
Setup
- Download the scripts for one flow testing and modify them to support multiple flows.
- You also need to download iperf ( local copy) and apply the Zopt patch.
tar vxfz iperf-1.7.0-source.tar.gz patch -p1 < iperf-1.7.0-Zopt.patch
- Run runall.sh.
[edit]
Observation
- CUBIC shows the best friendliness with standard TCP under short RTTs. The figures below show the throughput share between one CUBIC flow and one TCP-SACK flow under 6 and 9 ms RTTs, respectively.
- All other high-speed protocols except for CUBIC are aggressive even under short RTTs (STCP >= BIC > HSTCP > HTCP > CUBIC).
[edit]
Results
- Cwnd and Throughput screenshots
| RTT / Protocols | Type | BIC | CUBIC | HTCP | HSTCP | STCP |
|---|---|---|---|---|---|---|
| 6ms | CWND | | | | | |
| Throughput | | | | | | |
| 9ms | CWND | | | | | |
| Throughput | | | | | | |
| 48ms | CWND | | | | | |
| Throughput | | | | | |
