an update on me

[Dave Taht]

On Thu, Feb 16, 2012 at 8:55 AM, Francis Dupont <fdupont at isc.org> wrote:
>> > - C server library (i.e., a PCP engine) with glue for our AFTR,
>>
>> AFTR needs to run where and on what?
>
> => a Linux box with if possible (i.e., not required but better) two
> Ethernet interfaces (BTW the two netbooks we (me and Paul) used have
> dead fans, so are not usable). A BSD box works too but I used one
> a long time ago.

My test setup that has 20,000 miles on it is two lenovo laptops,
and 1 or more wndrs. The laptops have two network cards in them.

I documented the bloatlab in a previous message.

It sounds like for your demo you need a setup similar to mine.

I note that I picked up my lenovo T400 cheap in Paris,
on 'the street of the Chinese vendors', which if I can find a
subway map

It was used (off-lease) and 260eu, which is competetive with many
netbooks, with much more power and an all-important expresscard slot.

>> > Python fake servers and test suites for DS-Lite and NAT444
>>
>> Same question.
>
> => Linux boxes, in theory but I never tried a BSD box too.
> You didn't ask but the client is supported by any POSIX box,
> included recent WIN32 (recent: VS 2010 and full IPv6 support).
> There are some differences in Python too, the client runs on
> v2 (including old versions, cf RedHat :-) and v3, fake servers only
> on recent v2.

I'm confused about the client's language now.

>
>> > - Home Gateway support (extension of miniupnpd to PCP)
>>
>> that I build already. Are the patches or a forked repo somewhere
>> I can get at them?
>
> => there is a git repository with the whole code (based on 1.6 as
> far as I can remember).

At present I do not have access to isc's internal resources, and do
not have a contract as yet, either. I've been a friend of the family
for a long time, tho...

On the other hand, I do work closely with a bunch of openwrt devs,
who are all very interested in making this year's world ipv6 day a
brighter one, so I push opening up this process at the risk of being
annoying.

>> > This is on the wiki. About the thing we don't have:
>>
>> 'The wiki'  is where?
>
> https://wiki.isc.org/twiki/bin/view/Main/PCP
> (it is possible (i.e., likely) it is not very up to date)

no access at present. need an account.

Can add you a redmine wiki on bufferbloat.net if you want...

>> UCI for me = http://wiki.openwrt.org/doc/uci
>>
>> I figure we're not talking about the same thing.
>
> => in fact we are talking about the same thing, i.e., the user config
> interface of standard OpenWRT.

good.

>> googling for mB4 was unrevealing.
>
> => IPv4 multicast support in the DS-Lite mB4 element
> (BTW this means that all multicast items in the version of the SOW I got
> are covered by available open source code).

I am discouraged by the state of multicast in linux in general.

To summarize:

openwrt ships with pim turned off entirely... by default. (cerowrt has it on)

quagga's qpim code is out of tree and is basically an elaborate igmpv3
interface, no routing, tons of limitations, and the babel code in
quagga-RE is good, but mostly untested, so...

usc pimd seems decent but thus far I haven't seen it route. (this is
what I'm using). I've seen it work on the same ethernet interface.

One of my contributors tossed a port of Xorp into the ceropackages
tree, it hold promise but is also 5MB in size, compressed.

>> My reading of the spec indicated that kernel patches
>> were required for efficient handling of these messages.
>
> => which messages? anyway kernel patches are required to fix all
> the IPv6 bugs, for instance the stupid MTU of the IPv4 over IPv6
> tunneling (I say stupid because in fact it is from extra code :-).
>
>> A couple notes: cpe usually runs dnsmasq.
>
> => I know: I asked dnsmasq team to fix the EDNS0 size (and not only
> they fixed it without delay/discussion/etc, but they read the
> corresponding RFC and applied it for other small details...)
>
>> dnsmasq has some but very little ipv6 support in it's git head.

They are looking for guidance.

> => we can contact them but they are a bit like a competitor too (:-).

Simon is great. I don't regard dnsmasq and udhcp as a competitor at
ALL - they are complementary.

They address a market that isc has not been able to address, with
tools that are 1/5 to 1/20th the size, and easy to use web interfaces.

It would be kind of cool since you have the dibbler guy to have simon
on board, too.

that would give isc a monopoly on open source dhcp, and you can sell
it for twice the nothing you already sell it for!

>
>> I can and have run isc-dhcp but it's pretty unworkable.
>
> => I know Paul made a package for it (should be easier?)

My principal kvetches with isc-dhcp are:

enormous size on disk - includes all of bind now!
huge in memory, too, for something that runs 20-30 times a day, tops
turing complete language in it
nearly impossible to configure via a web interface
needs two daemons running for ipv6 and ipv4

evanh runs it on his iscwrt box, and given that dibbler is huge, too,
I've been considering revisiting isc-dhcp, but gotta say that dnsmasq
has mindshare, tools, and an embedded focus that those two products
lack.

>> I've also been working with dibbler of late, but the client is buggy.
>
> => hum, is Tomasz in the list?

I bug reported one of my problems. I'll be looking into it harder next
week if I'm at comcast.


>> I was based in paris at the lincs.fr lab until last month
>
> => I know it (I worked for both INRIA and Institut Telecom)

Enjoyed it.

>
> Regards
>
> Francis Dupont <fdupont at isc.org>



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