General Conclusions

Rate limit experiments have been executed with the configuration described in the "Setup" subsection of the "Rate Limit" main section. from the executed TCP bulk, TCP triangle profile, UDP bulk, and UDP triangle profile tests the following general conclusions can be drawn:

• Both in the TCP and UDP tests the configured rate limit bandwidth of 2800 Mbits/s is not even closely reached. The relatively small configured burst size of 50 KBytes could be the main reason for that.
• Given the relatively large round-trip time of a bit more than 100 ms, packet lost would lead to a considerable performance break down with TCP. A larger burst size could probably be beneficial for this point.
• With small shaped, top triangle bandwidth values of 50 and 100 Mbits/s, using one or two parallel flows, the measured TCP bandwidth profile could very well follow the theoretical triangle flow. This could also be an indication that the configured burst size of 50 KBytes was too small.
• At the UDP triangle profile tests in the rate limit direction larger server bandwidth values could be reached as in the UDP bulk tests. Also here the small burst size could be a reason for that. Caused by the relatively small grow of the shaping client bandwidth in the up-going area it takes a longer time for all buffer memory has been used.
• When more than two parallel flows are used the sum TCP and UDP client bandwidth values, taken over the parallel flows, is decreasing. Host limitations are probably the reason for that.