Is the bufferbloat project also intending to improve ethernet (landline) performance?

I don't recall reading much on drivers that run ethernet networks. Even on wikipedia I could find a comparison of open-source wireless drivers but nothing on open-source wired or networking drivers.

Am I correct in my assumption that the drivers in LEDE compatible routers are fully open-source? Assuming so, is the bufferbloat project intending to improve bufferbloat on these drivers also or is the focus only on wireless drivers?

To my knowledge, it's not about wireless drivers, but about buffering in network devices in general. Since wireless adds additional latency, some open wireless drivers (e.g. ath9k) were tackled as well, but you should see improvements on regular wired routers as well.

Since you set SQM up on your WAN interface (at least, that's all I've seen recommended so far), it kind of implies wired connections are the prime target.

I don't recall reading much on drivers that run ethernet networks. Even on wikipedia I could find a comparison of open-source wireless drivers but nothing on open-source wired or networking drivers.

the bufferbloat project started on wired networks. That problem is pretty close
to solved, so they have moved on to other issues.

Unfortunantly, far too many drivers have not been updated as needed. The first
thing that needs to be done is to implement BQL (Byte Queue Lenths, instead of
allowing queues to be up to N packets of any size, limit them to M bytes
instead)

After that, using fq_codel solves the vast majority of your issues. If you have
an overloaded connection, and especially if the far end of the connection is not
managed, then sqm and cake add additional logic to provide more fairness and
pull tricks to try and limit inbound traffic.

Am I correct in my assumption that the drivers in LEDE compatible routers are
fully open-source? Assuming so, is the bufferbloat project intending to
improve bufferbloat on these drivers also or is the focus only on wireless
drivers?

The bufferbloat identified the problems and created/tested the solutions, but
it's not staffed to try and go through every driver and modify them to implement
the fixes, that's up to the maintainers of those driers (or other contributers)

The Bufferbloat folks consider simple wired networks a solved problem, they are
working on two other categories of problems

  1. providing better fairness in the face of the extreme congestion that you get
    when you go from a local gig-E network to a slow (single digit Mb or so) WAN
    network. Including trying to regulate inbound traffic

  2. addressing wifi. ath9k, ath10k and mg76(I think that's the driver) have been
    pretty much de-bloated. But it's been discovered that removing excess queue
    length isn't enough. Wifi needs the equivalent of the BQL fix, and that's
    complicated in the fact of protocol overhead that is much larger than the
    packets transmitted, and the variable transmission speed. The make-wifi-fast
    project has spun off and is working on this, with the airtime fairness that
    first hit the ath10k driver being the first results of this.

David Lang

1 Like