Captive Portal per user bandwidth limit and bufferbloat

[deajan]

Hello,

When using per user bandwidth limit on the captive portal of a fresh install, I get a very high ping value when transferring data, I think of bufferbloat.
Having checked with http://speedtest.dslreports.com, I get +400ms buffer bloat when downloading, and +1800ms when uploading.

Using a 100/10 Mbit cable link, CP is limiting per user to 3076/512kbit (public wifi spot usage).
Traffic shaping isn't setup in this test case because with a single CP client that's limited to 3mbits, I won't saturate the cable link.

Is there settting way to achieve tcp backoff per user on the captive portal ?
I've played with TCP buffer sizes already, without any improvement.

Client OSes tested: Fedora 24 & Windows 7.

Regards,
Ozy.

[richb-hanover]

Here's an alternate solution that might keep latency low for you. It doesn't strictly answer your question - it doesn't limit anyone's bandwidth, but instead gives all sessions/flows a fair share of the bandwidth that's there. (If only one person is using it, they get it all, if two are on, they get half, three people get a third, etc.)

Use CoDel/fq_codel. Its primary purpose is to keep latency low. It does this by pushing back (ECN or flow control) on big flows (large transfers) to limit their effect while letting small flows (VoIP, gaming, DNS queries, TCP connection packets) to go through immediately.

[deajan]

Thanks for the info, but as of my understandings today coDel is implemented but FairQueue CoDel is not, so fair bandwidth share won't happen.
Am I getting this wrong ?