Bloat » Cerowrt

There was a huge thread on this in private email discussing various approaches. I don't know what was ultimately decided upon.

The ahcp package has been made manageable via the command line, so we're back in business for multiple interfaces for cerowrt 1.0. tested in the lab today. That said, coaxing the gui back in shape would be good for the next release.

This is the second of a series of emails discussing things that may be
dropped from the next release of cerowrt, and is intended to open the
discussion, only.

The goal is to find better ways to focus on fixing bufferbloat.

While I personally find babel very useful, as well as AHCP, convincing
anyone that these two bleeding edge technologies are worthwhile on
places like the homenet mailing list is an incredible struggle. The
consensus there appears to be leaning towards ospf, not that anyone on
that list has bothered to test that, either. We have had the ability
to use ospf in the system since day one as an optional package.
Routing, actually, is not needed in simple setups.

Other problems:

1) The AHCP package in openwrt keeps breaking. It broke of the late
rc7-smoketest6-ish series, and so far as I know, hasn't been fixed.

1a) The gui interface, even when the package was working, is equally
problemantic

2) Firewalling remains REALLY problematic with both babel and AHCP.
More often than not, you end up with stuff
that doesn't work.

3) Packet loss as a metric appears to not be a good metric anymore.

4) As much as I enjoy mesh networking, the few people that have tried
to set it up independently that I know of, have all failed.

And lastly,

5) As pointed out during a test in brussels, cerowrt itself has become
too complex even for someone familiar with openwrt to deal with it.

So with no real users of these subsystems in the field, these two
packages can be safely dropped.

There were some positive benefits to being able to test ad-hoc mode,
having the interfaces available at all times, etc, but deleting the
extra interfaces and firewall complexity would help.

While I believe that a pure mesh environment works a LOT better than a
pure 'as we know networking today' environment, however, the attempt
thus far,
of merging the 'best of' the two, seems to be worse, than either.

This eliminates having to fix bugs #252, #110, #112, #201, and I've
never got around to mentioning how much I was hoping for dhcp/ahcp
integration.

Fully agree. I think that CeroWRT should focus on the bufferbloat problem, and mesh networking is only tangentially related to that.

-- jch

That said, AHCP in particular needs some more time to 'bake'.

Just to clarify -- AHCP and the ahcpd implementation are pretty mature. It's the OpenWRT packaging of ahcpd that needs improved.

--jch