| From: David Collier-Brown via talk <talk@gtalug.org> | | On 3/9/23 09:16, Lennart Sorensen via talk wrote: | > On Wed, Mar 08, 2023 at 05:04:52PM -0500, William Park via talk wrote: | >> Mine is | >> <https://www.waveform.com/tools/bufferbloat?test-id=366a5b37-ac7f-4c79-a3e3-f2a60175e726> | > Well I get an F. I then followed their instructions to add sqm to my | > openwrt setup, which made it a D, but at a cost of 75% of the throughput | > (so 6Mbit instead of 25Mbit), so I turned that off again since that was | > too high a price for a slight latency improvement. Perhaps some tweaking | > could make it better. | > | Yes: the penalty for fq_codel or CAKE should be an indicated percent or so, | and often is a benefit instead (:-)). Older software is often bad enough that | it eats bandwidth. | | Look for a openwrt release that mentions one or the other of those keywords, | preferably CAKE. There is more information about what code is best. Can you explain or point to an explanation about why the reading Lennart is seeing is misleading or not salient? Bufferbloat can occur at many places between the endpoints. Does CAKE on one endpoint try to eliminate BB on the whole path or just the points running it?