Hacker News new | ask | show | jobs
by martin1975 3238 days ago
Has anyone put together a solid position on how QoS would operate in a world of net neutrality? QoS for particular services that might need it pretty much guarantees some packets will go ahead of others. My question or dilemma is, can QoS work while preventing abuses like abridging freedom of speech?
6 comments

My opinion:

Do QOS on the type of packets, not the source/destination. VOIP, gaming? To the front of the line. Web? Firmly in the middle. Streaming next. Downloads/BT? To the back.

Every packet is not treated equally, but prioritized according to the needs of the consumer. The source or destination of the packets shouldn't matter.

EDIT: I've implemented this very policy in an office, and despite the fact that the links in and out were regularly saturated and QOS was triggered - nobody complained, because their expectations were met. Netflix had the same priority as YouTube. Downloads of Linux ISOs or the latest GOT episode never interrupted in-progres VOIP calls.

> The source or destination of the packets shouldn't matter.

Specifically, if the source is the ISP itself. Comcast or Spectrum or AT&T shouldn't be allowed to prioritize their own streaming traffic* over Netflix's.

If they can't deliver quality streaming (theirs) to their customers, they should invest in better infrastructure.

* Yes, this includes trying to make the claim that their own VLAN / segregated TV-over-IP network is "different". They own the last mile, they need to provide RAND access.

Combination of QoS of important services (e.g. VoIP) and Net Neutrality works just fine in EU.
It's fairly simple. You let the user set QoS bits in their own packets, and discriminate based on those, with appropriate target neutral constraints to prevent abuse. (e.g. limit the bandwidth available to packets marked low latency.)

This allows people to use whatever services they like, but still get the benefits of QoS.

You can also use those same constraints without QOS bits. Small amounts of data sent bi-directionally gets lower latency (VOIP). Large amounts of data sent in one direction gets lower priority (large ISO downloads). Combine with protocol detection to set initial values.
In the very rare use cases where QoS for a specific service is needed it should be per subscriber, thus something that the subscribers router deals with. Not something the ISP needs to concern itself with at all.
That will do nothing for congestion within the core of the provider network. That's where qos is actually needed. Asking providers not to oversubscribe at all would literally increase the cost of broadband a minimum of 100x overnight.
> That will do nothing for congestion within the core of the provider network. That's where qos is actually needed.

It isn't. Congestion in the core of the network shouldn't happen.

> Asking providers not to oversubscribe at all would literally increase the cost of broadband a minimum of 100x overnight.

That's conflating two different things. It's possible to oversubscribe and still have no congestion, because having 100% of customers all use 100% of their connections at the same time literally never happens.

As long as the uplink can carry the peak load in practice, there will not be any congestion. If it can't then the network is too oversubscribed, which should not happen.

I'm not asking them to not oversubscribe. Just don't do QoS. This isn't a problem today and there is nothing that suggest it will a problem in the future either.
You can't have QoS in a pure version of net neutrality.

That's why the FCC's implementation makes exceptions for traffic management and deals only with business practice related net neutrality violations.

Of course, not all business practice related net neutrality violations are bad for the consumer, so the FCC makes lots of exceptions for those as well (see: zero rating).

If you've got a service that's that critical, shouldn't you be paying for a private leased line?