Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Weaver]

Say I am doing a huge download of a movie or group of tv series episodes from Amazon or Netflix. Is there any practical way of ‘lowering the priority’ of this so that everything else, especially, for example web browsing manually initiated takes priority?

I fully realise that this question is (at least) doubly ill-defined. Firstly you would have to define what counts as ‘background downloads’ somehow, in some terms, and hopefully without the matter becoming a maintenance nightmare. Secondly, would have to decide on what is meant be priority. In QoS terms it is straightforward enough, if only one can apply QoS by either marking or by hackish classification rules or whatever. Without that, it could just mean slowing some things down/throttling/speed shaping or some such, which is not good at all. Air you could try the converse and ‘boost’ things that are considered ‘foreground’ whatever on earth that might mean.

Since it is on an iPad, it would be nice if apps and iOS could manage this as they know how to classify the activity types along the lines that I am trying and failing to express. Failing that it is down to routers. My Firebrick is useless as far as QoS is concerned but very flexible with throttling/speed shaping, even though the end results are a imho inaccurate.

Question though: how on earth could one even hope to formulate rules for matching traffic to define ‘Amazon download’ or ‘Netflix download’?

I don't know if their IPv4 and IPv6 address blocks for such servers could be found somehow, apart from traffic capture and constant whack-a-mole. Also it would be really bad news to limit live streaming because I need all the speed I have got for that.

Anyway, if there is no foreground activity then it is a horrible idea to slow down background downloads.

I am not at all convinced that this issue is either sufficiently well-formed or solvable. However, if anyone has any thoughts then I would welcome them.

[banger]

I am not sure if this helps but my Asus router QoS prioritises web traffic http and https and puts downloads down to 80%. Does that help?

[Weaver]

Any idea how it categorises things?

[banger]

I would think packet tagging.

[johnson]

I know I go on about it, but I honestly believe the bufferbloat people have come up with the best solution so far. If you actively manage queues on both ingress and egress everything just works so much better. At the sacrifice of a small percentage of throughput never having anything end up waiting in a buffer means that even without specific prioritisation nothing experiences a significant enough delay to affect services. Its not perfect, but it seems better than approaches that have been taken before.

I used to run QoS that was sorted into priority express bulk etc, but with only ports and protocol used to specify it only really worked for certain things and it still had the throughput penalty. With SQM I specify only up down throughput and even when saturating the connection both ways get only 10s of ms increased latency. Also while not directly related to SQM the cake queuing discipline I use has secondary options for sharing bandwidth equally by IP, and it works well. One system maxing out the connection with a torrent is near instantly halved when loading an HD stream on a second. I believe this uses fairly recent changes in the linux kernel that allow awareness of networking streams that was previously opaque.

SQM seems to growing in availability, openwrt is the prime example, but I think the asus routers open source offering has it, also the ubiquity stuff and I believe the progress is being made with pfsense, despite resistance.

[Weaver]

No I meant how does it decide what to tag as what? (At L2 or at L3, or both or who knows.)

I agree with you completely about anti-bufferbloat technologies, that sorts so much of this stuff and nonsense out, with huge queuing delays. That is an excellent idea.

Perhaps I should ask the developer, RevK, very, very nicely if he would consider implementing the latest active queue management algorithms in the Firebrick (my router). Maybe C sources are around to look at too.

Round trip time (is ‘latency’ rtt, or one way ? ) goes up from ~40 ms to ~230 ms when a flat-out download is going on.

[jelv]

I found this from a quick search - have you tried it?
https://www.freedownloadmanager.org/

[Weaver]

Sorry jelv, don't have pc or mac as I am permanently in bed due to pain and dizziness, can't sit upright for long without feeling very ill most of the time. I am iPad-only these days as I can use one in bed.