Hi There,
I am working in a highly oversubscribed LTE network (only available band is band 3 for my provider), at peak periods things slow down and it’s apparent that the ISP is implementing some type of user limiting based QOS (possibly policing) during peak periods. Unfortunately I am a high bandwidth user and I am always moving a lot of data, the result being that my service starts to crawl until I reduce the load on the service.
I am looking at a Layer 7 QOS implementation so that sites I label critical are given precedence over bulk traffic, the issue is that all seem to have a priority based on percentage interface bandwidth.
In my case this is variable, off peak I may get 50Mb/s but on peak it can be as bad as 1-2Mb/s so percentage interface bandwidth will not work.
Is there any way to implement a FIFO queuing policy based on DNS without having to specify the interface bandwidth?
Have you considered this alternative:
You could “offload” your traffic (and work) to a server of your own that is in the Internet, ie. a cheap VPS like these with 20 TB (or more) traffic per month:
https://www.hetzner.de/cloud#pricing
Maybe you can find even a cheaper provider in your own country or region.
Ie. in the above table you could take the cheapest one (CX11) that gives you 20TB/month.
Take 2 of them and now you have 40TB/month 
And: they have Gigabit interface to the Internet. I’m getting about 950 Mbps download speeds when I’m downloading files on them from my home.
And: the upload and download speed between them is more than 3 Gbps! Ie. that provider has a fast backbone among their different locations.
I’ve 3 of them at different European locations and can help if you have any questions.
Here’s an iperf test with 8 software-threads that runs 60 seconds. The client and server are at different locations (cities):
$ iperf -c $SERVER -p $PORT -P 8 -t 60
Client connecting to x.x.x.x, TCP port xxxx
TCP window size: 85.0 KByte (default)
…
[SUM] 0.0-60.0 sec 24.8 GBytes 3.55 Gbits/sec