a sdcpe package 'feed' request

[Dave Taht]

On Wed, Mar 14, 2012 at 1:33 AM, Francis Dupont <fdupont at isc.org> wrote:
>> Things are looking brighter on the kernel side since 2.6.39 (.32? hopeless),
>> but the tools and integration of such is something an ongoing
>> nightmare. Also all the
>> main AQMs have no ipv6 support in them. Etc.
>
> => the last point is really a shame: when I worked on IPv6 in BSDs,
> AQMs had IPv6 support (it was (and still is) pretty easy to add IPv6
> support to a packet classifier), and it was more than 10 years ago!

I should rephrase this. The default underlying AQM technology (SFQ, RED, etc)
does support ipv6. Some of it even does more of the right thing when
tunneling is in use.

However, 1) the default hash algorithms are frequently stupid and hash
only on the 3rd 32 bit word of
ipv6

2) most of the userspace code (Wondershaper, adsl-shaper, openwrt's
shaper) only does classification, and hashing based on ipv4 addresses.

1) is mostly fixed by 3.3, but not entirely.

>
>> > => in fact 6rd was in the Comcast plan so the dslite-6rd got both.
>>
>> Yes, I deployed 6rd in conjunction with their testing last year. Due to
>> the brain-dead deployment of such (only a /64) I reverted to using
>> 6to4 which was at a 'just working' state at least until recently, and
>> their 6to4 gateways were excellent.
>
> => a /64 is better than a /128 or nothing. But as I have a /64 at
> home I agree it is not enough for R&D.

I adopted AHCP long ago because it did everything on /128s. And fails
over between wireless and wired.

>
>> They plan another brain-dead deployment of native ipv6, a /64 only
>> because only one of the cpe gear they have can even do that with pd.
>>
>> With wide-dhcp PD I have users in new zealand that are dealing  with /48s
>
> => wide-dhcp PD? I was in its team (so with Tomasz and dibbler we have
> most of the DHCPv6 at ISC today :-)

Applause. I've been thinking that ipv6 needs another effort on the
scale of wide,
or larger....

>> Well, I setup a chat room for this project to possibly make it easier
>> to interact in real time. Would that work?
>
> => it is hard to develop and chat at the same time. Anyway with the
> timezone difference IMHO tight interactive work can be only an
> exception.

Concur, however I CAN try to be available at a time convienent to you,
and I gotta admit, getting the first two boxes running would work best
with some handholding on your part. After that I would hope that
assembling the rest of the demo would go easier.

At the moment I'm burning the midnight oil trying to get enough pieces
assembled from the pieces assembled thus far to at least see aftr work.

The pcp dissector, tossed on top of the other pcp dissector - just worked,
so at least I have a working wireshark (from git head) on one machine now.

Tomorrow the laptops arrive.

>> I try not to do phone calls myself, being partially deaf, and when I
>> do, try to use something of reasonable quality, like skype.
>
> => skype is forbidden in France by the ANSSI (the equivalent of the NSA
> but far smaller). And I can call anyone in the USA for free with

Make a skype call, go to jail? Urgh.

> the phone handset connected to my ADSL box, so no need for skype.

-- 
Dave Täht
SKYPE: davetaht
US Tel: 1-239-829-5608
http://www.bufferbloat.net