Let them run CAKE

It sounds like CAKE does what I did using HTB and SFQ. But I suspect CAKE does it more elegantly. Even with uplink and downlink saturated, I've not seen any unevenness transmitting or receiving with my config. UL and DL BW are set to the maxima I observed. All flows are set to share 1 bps. Isochronous flows (priority 0) can use up to 64000 bps (configurable for the amount of traffic). DNS and NTP (priority 1) are allowed to use 10% or so. P2P and other bulk flows (priority 3) are allowed to use 100%, but can be starved. Everything else is priority 2 and can use 100%. Once I finally understood what is and is not said, and what is and is not true, in HTB's, mmm, imperfect documentation, I was able to produce a very stable setup.

As to actual bandwidth, that will probably always be an empirical measurement because even ISPs cannot precisely configure customers' bandwidth; 30D/5U was usually 28D/6U, and 100D/10U is usually 112D/12U. So find a fast site, start a large download, and observe the average maximum. Find/beg/borrow a site to upload to, start a large UL and,again, observe the average maximum. This works quite well for me. (OK, a DS1 or DS3 will be quite close to their natural bit rates.)

I do hope the devs use only decimal (powers of 10) multipliers (k, M,G, etc.) and *not* binary (powers of 2). Transmission speeds have always been stated and configured as powers of 10 going back 50 years and more; using powers of 2 only serves to make computations rather incomprehensible. TC and HTB interpreted and used multipliers so poorly that I stopped using them in configurations completely in favor of the plain, natural number; it's the only way I could make number come out right.

If it's done right, CAKE may be pretty close to something I suggested some years back. I look forward to giving it a go.