- The tests using Flent's rrul_be test at 100mbit symmetric are a little surprising, where Cake's average download bandwidth was around 14-15mbit, while fq_codel's was around 21mbit. I did a second trial for cake at a separate time and got the same results.
Hmm. Two possible causes - A) we have not evaluated the aqm portion at a lot of higher rates, B) we do GRO peeling and that costs cpu. How does it do on a tcp_ndown test or rtt_fair_up --swap-up_and_down?
- The results for triple-isolate look smooth, and I don't see that that option adds significant load on the router, but I don't know
rtt_fair_be (and related) lets you specify 4 -H targets. So a simple test is to aim 3 at one target, 1 at another
