Traffic Shaper difficulties - 4G wireless connection

[GeoffW]

First, I thought explaining the connection situation might let someone who knows jump straight to what I suspect might be conclusion: you cannot reliably shape this.

I operate from a remote location via a 4G wireless connection.  In simple terms:

   WiFi----\
          OPNsense---4G---ISP---Internet
   LAN-----/

The quality and speed of the 4G connection varies significantly with the weather and time of day.  Download: rarely I might see 25Mbps, most of the time I see 8Mbps..15Mbps (it really does vary that much even over a brief period), but sometimes it will drop to 3Mbps or even lower.  Upload: tends to be more consistent, anything from 7Mbps to 20Mbps; it regularly tests faster than download.

How do you define shaper pipes to encompass such variability?  I don't want to constrain the traffic unnecessarily, which would seem to exclude the possibility of dedicated pipes for certain traffic (unless I want to constrain the total traffic down to a very low value).

Without reliable pipe definitions, I have trouble seeing how the shaper can make appropriate choices, most especially as it relates to download traffic (which is where I see the worst of my problems).


What I wanted to achieve:

My mobile phone currently uses a VoIP app for a business line, and WiFi Calling for mobile calls.  (The office 4G wireless uses an antenna on an 11m mast, so it gets a much better connection than the mobile phone.)  During problem times (Internet being used as well as calls) the download side of the call in particular seems to experience lengthy transmission delays (up to 5 seconds) and sometimes whole words or sentences get dropped.  VoIP is worse than WiFi Calling.  So I was hoping to use the traffic shaper to prioritise calls over other traffic.

The VoIP traffic is easy to identify, I have a list of IPs used by the service provider.  Watching the traffic graph shows it using around 120kbps each way, and both ways are constantly active while a call is active, so utilising 240kbps even if no one speaks (which is higher than I was expecting).

WiFi Calling uses IPSec, but is not the only IPSec service being passed through this firewall, however, setting up shaper rules based the ports (4500 and 500) and the mobile IP address appears to do the job in this case.  WiFi Calling seems more efficient than VoIP, sending traffic only while speaking, but this makes it harder to get a direct measure of its bandwidth requirement.


What I have tried:

I tried setting up a dedicated pipe for VoIP but that seemed to be the worst for download delays - I think because I had oversized the lower priority pipe (so Shaper tried to give it bandwidth it assumed must exist but didn't at the time).  Using queues appeared to make some difference, but often not enough to make for reliable phone calls.

I started off using separate upload and download pipes but was still seeing problems.  When I experimented with running multiple network speed-tests it seems that upload does significantly influence download (at least in those tests‚ I could not find a test that does proper full duplex speed testing).

So then I moved to a single pipe and just prioritise queues inside that single pipe (up and down).  While it's unclear how much this helped, at least it was simpler.  It made it easier, for example, to test reducing this pipe down to lower speeds.  When I cut the pipe to 5Mbps and ran a speed test while talking, I saw very little delay and break-up in the voice call, but - obviously - the total bandwidth was reduced to under 5Mbps (the speed-test I ran while on the call showed 4.6Mbps).


Possibilities?

* Operate using a shaper pipe defined to the lowest commonly available speed.  Not really an option, but it would be possible.

* If I'm sitting at my desk when a call comes in I could bring up the Shaper and set the pipe according to current connection quality.  I'm not sure how dynamically the shaper responds, but I'm sitting at my desk enough of the day that this is probably feasible if not attractive.

* Maybe there are some advanced options in the Shaper that apply to this situation?


Suggestions welcome.