[Cerowrt-devel] 3.3rc5-1 is out

[Cerowrt-devel] 3.3rc5-1 is out

[Dave Taht]

3.3rc5-1 is out at:

http://huchra.bufferbloat.net/~cero1/3.3/3.3-rc5-1/

This was the first time I've had a chance to try to actually deploy
cero as the main router in a production environment (esr's) in quite
some time. Enough problems cropped up for me to recommend against
doing that at this point, although I am writing this email through it.

I did get some good data on how FIOS works (short summary: it's *hard*
to stress out fios)

Features:

Kernel 3.3rc5 + 83 patches to get it closer to rc6
gpsd 3.4+ package (from git head, trying to fix a PPS problem)
bind9.9.0 (released version)
dibbler-client (git head, fixing a crash with multiple interfaces)

A couple bugs have crept in due to bitrot, some of which I've bug
reported. The most severe appears that I have a problem with
power-save and crypto at 2.4 ghz in this (trouble with an android
phone, and the like)

firewall: syntax has changed, rules notably re ipv6 have changed,
all kinds of problems with multiple interfaces
6to4: distributing 6to4 addresses onto interfaces didn't work
ipv6: radvd doesn't always manage to distribute an address
(multicast? client side problems?) - usually crops up on more modern
wireless interfaces
ipv6: dibbler-client eats 100% of cpu after being idle overnight
http://www.bufferbloat.net/issues/341
ipv6: dibbler-client wants to 'own' radvd
bind9: needs allow-recursion
http://www.bufferbloat.net/issues/342

I'm pretty sure that most of these bugs have existed in prior releases, with the
exception of the wireless power save one... which was kind of hard to trigger.


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

Re: [Cerowrt-devel] 3.3rc5-1 is out

[Dave Taht]

Dibbler-client is on by default in this build.

I would be very interested to know if dibbler's pd support actually
works in a real environment, (the bugfix for the interface crash *did
work for me* in testing distributing /64s and /56s)

...and if the new 100% of cpu bug (which only seems to happen the
absence of a dhcp v6 server) can be reproduced.
(I'd not seen it before because I'd always HAD a dhcpv6 server present
when playing with dhcpv6)

I note - at least at the moment - anyone trying 3.3rc5-1 without a
dhcpv6 server present, should disable dibbler entirely.

As things turn out...

Tomas updated dibbler to 0.8.2 on me yesterday, incorporating the git
patches I was testing and had proven to work, and tells me it compiles
down much smaller with -Os than it used to. So I imagine that pushing
that up to openwrt head is a good idea.

The new syntax for restricting what interfaces get ipv6 addresses from
dhcpv6-pd is:

downlink-prefix-ifaces "se00" , "eth0", "wlan0", "etc"


On Thu, Mar 1, 2012 at 3:30 PM, Dave Taht <dave.taht@gmail.com> wrote:
> 3.3rc5-1 is out at:
>
> http://huchra.bufferbloat.net/~cero1/3.3/3.3-rc5-1/
>
> This was the first time I've had a chance to try to actually deploy
> cero as the main router in a production environment (esr's) in quite
> some time. Enough problems cropped up for me to recommend against
> doing that at this point, although I am writing this email through it.
>
> I did get some good data on how FIOS works (short summary: it's *hard*
> to stress out fios)
>
> Features:
>
> Kernel 3.3rc5 + 83 patches to get it closer to rc6
> gpsd 3.4+ package (from git head, trying to fix a PPS problem)
> bind9.9.0 (released version)
> dibbler-client (git head, fixing a crash with multiple interfaces)
>
> A couple bugs have crept in due to bitrot, some of which I've bug
> reported. The most severe appears that I have a problem with
> power-save and crypto at 2.4 ghz in this (trouble with an android
> phone, and the like)
>
> firewall: syntax has changed, rules notably re ipv6 have changed,
> all kinds of problems with multiple interfaces
> 6to4: distributing 6to4 addresses onto interfaces didn't work
> ipv6: radvd doesn't always manage to distribute an address
> (multicast? client side problems?) - usually crops up on more modern
> wireless interfaces
> ipv6: dibbler-client eats 100% of cpu after being idle overnight
> http://www.bufferbloat.net/issues/341
> ipv6: dibbler-client wants to 'own' radvd
> bind9: needs allow-recursion
> http://www.bufferbloat.net/issues/342
>
> I'm pretty sure that most of these bugs have existed in prior releases, with the
> exception of the wireless power save one... which was kind of hard to trigger.
>
>
> --
> Dave Täht
> SKYPE: davetaht
> US Tel: 1-239-829-5608
> http://www.bufferbloat.net



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

Re: [Cerowrt-devel] 3.3rc5-1 is out

[Dave Taht]

I setup a test environment using 3.3rc5-1 for the crypto problem and
was unable to get it to recur.

I've put some more info on the dibbler bug on the related bug, and
confirmed there's a problem (that takes at least a few hours to occur)

(summary, if you aren't using dhcpv6, turn dibbler off. Easiest way to
do that is rename the /etc/dibbler/client.conf file to something else)

On Thu, Mar 1, 2012 at 3:48 PM, Dave Taht <dave.taht@gmail.com> wrote:
> Dibbler-client is on by default in this build.
>
> I would be very interested to know if dibbler's pd support actually
> works in a real environment, (the bugfix for the interface crash *did
> work for me* in testing distributing /64s and /56s)
>
> ...and if the new 100% of cpu bug (which only seems to happen the
> absence of a dhcp v6 server) can be reproduced.
> (I'd not seen it before because I'd always HAD a dhcpv6 server present
> when playing with dhcpv6)
>
> I note - at least at the moment - anyone trying 3.3rc5-1 without a
> dhcpv6 server present, should disable dibbler entirely.
>
> As things turn out...
>
> Tomas updated dibbler to 0.8.2 on me yesterday, incorporating the git
> patches I was testing and had proven to work, and tells me it compiles
> down much smaller with -Os than it used to. So I imagine that pushing
> that up to openwrt head is a good idea.
>
> The new syntax for restricting what interfaces get ipv6 addresses from
> dhcpv6-pd is:
>
> downlink-prefix-ifaces "se00" , "eth0", "wlan0", "etc"
>
>
> On Thu, Mar 1, 2012 at 3:30 PM, Dave Taht <dave.taht@gmail.com> wrote:
>> 3.3rc5-1 is out at:
>>
>> http://huchra.bufferbloat.net/~cero1/3.3/3.3-rc5-1/
>>
>> This was the first time I've had a chance to try to actually deploy
>> cero as the main router in a production environment (esr's) in quite
>> some time. Enough problems cropped up for me to recommend against
>> doing that at this point, although I am writing this email through it.
>>
>> I did get some good data on how FIOS works (short summary: it's *hard*
>> to stress out fios)
>>
>> Features:
>>
>> Kernel 3.3rc5 + 83 patches to get it closer to rc6
>> gpsd 3.4+ package (from git head, trying to fix a PPS problem)
>> bind9.9.0 (released version)
>> dibbler-client (git head, fixing a crash with multiple interfaces)
>>
>> A couple bugs have crept in due to bitrot, some of which I've bug
>> reported. The most severe appears that I have a problem with
>> power-save and crypto at 2.4 ghz in this (trouble with an android
>> phone, and the like)
>>
>> firewall: syntax has changed, rules notably re ipv6 have changed,
>> all kinds of problems with multiple interfaces
>> 6to4: distributing 6to4 addresses onto interfaces didn't work
>> ipv6: radvd doesn't always manage to distribute an address
>> (multicast? client side problems?) - usually crops up on more modern
>> wireless interfaces
>> ipv6: dibbler-client eats 100% of cpu after being idle overnight
>> http://www.bufferbloat.net/issues/341
>> ipv6: dibbler-client wants to 'own' radvd
>> bind9: needs allow-recursion
>> http://www.bufferbloat.net/issues/342
>>
>> I'm pretty sure that most of these bugs have existed in prior releases, with the
>> exception of the wireless power save one... which was kind of hard to trigger.
>>
>>
>> --
>> Dave Täht
>> SKYPE: davetaht
>> US Tel: 1-239-829-5608
>> http://www.bufferbloat.net
>
>
>
> --
> Dave Täht
> SKYPE: davetaht
> US Tel: 1-239-829-5608
> http://www.bufferbloat.net



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