[Bloat] ultrafast broadband conference june 27-30

Toke Høiland-Jørgensen toke at toke.dk
Sat Jun 18 06:51:49 EDT 2016


Jan Ceuleers <jan.ceuleers at gmail.com> writes:

> Either you perceive the problem to be located in the link technology
> (i.e. DSL generally or only specific flavours of it). If this is the
> case what needs to be fixed is the standard so that implementations
> thereof will improve.
>
> Or else you perceive the problem to be located in the CPE that implement
> DSL, but in the layer above the DSL link layer. In this case what needs
> to be fixed is those implementations, probably starting by the reference
> firmware written by chipset vendors.
>
> I think it's the latter. If it's the former then indeed don't hold your
> breath because the standardisation is done and dusted.

It is indeed the latter. However, it is correlated with DSL technology
because the equipment tend to be using binary blobs for drivers that
themselves have huge buffers; so even if the device is running Linux,
sticking FQ-CoDel on it doesn't do much good without a software rate
limiter. And also, most devices are owned by the ISP, so the consumer
can't upgrade them anyway.

So while it is nothing inherent in the technology per se, in practice it
is a fairly safe bet to say "ah, you're on DSL? Well, you are probably
suffering from bufferbloat, then".

I know of at least one DSL vendor who supposedly has started paying
attention after pressure from a clueful ISP; not idea if they started
shipping non-bloated products yet.

-Toke



More information about the Bloat mailing list