[Bloat] an update on cerowrt-rc7

[Bloat] an update on cerowrt-rc7

[David Täht]

[-- Attachment #1: Type: text/plain, Size: 3621 bytes --]

A new smoketest of rc7 is out:

http://huchra.bufferbloat.net/~cero1/rc7-smoketests/rc7-smoketest5/

Which is basically linux-3.0.8 + openwrt head + support for the wndr3800

CeroWrt IS independently buildable now:

http://www.bufferbloat.net/projects/cerowrt/wiki/Building_Cerowrt_on_your_own_machine

It's getting easier to install:

http://www.bufferbloat.net/projects/cerowrt/wiki/OCEAN_CITY_INSTALLATION_GUIDE

and the roadmap remains overwhelming.

http://www.bufferbloat.net/projects/cerowrt/roadmap

...

I was feeling kind of discouraged about the prospect of renewing work on 
cerowrt this month.

The goal has been to get to where the OS and drivers are stable, then do 
a bunch of *repeatable* bloat-related experiments - then kick back into 
a development mode with those results.

we've been spending increasing amounts of time validating the lowest 
levels of the stack - re-analyzing the original bufferbloat data, as one 
example - and we've designed a couple experiments that should reliably 
show bufferbloat AND hopefully various mitigations against it, once they 
exist - that can be repeated by engineers (in front of management!) and 
students using the thing.

and I've been thinking really hard about what a good API to Minstrel 
would be...

All of which was more fun that getting cerowrt from rc to 1.0.

Because, on the whole, the last rc4 is working pretty good.

AND the prospect of building an entire OS AGAIN and burning it into the 
machines at bloatlab #1 was a bit daunting. Not only that, but enough 
fixes had gone into the linux-3.1 release (mostly ipv6 related) that I 
was thinking it would be best to aim for that as a final target for 1.0 
than stick with 3.0...

And THEN I was pleased to see people actively pointing to the project today.

http://ask.slashdot.org/story/11/10/25/1429235/ask-slashdot-dd-wrt-upgrade-to-80211n

darn it... so my enthusiasm is back closer to where it belongs, and I 
got that smoketest out. Darn it, bruce...

So.

My thx to Petri Rosenström for simply *ignoring* the steaming pile of 
patches for the wndr3800 that are on
"http://www.myopenrouter.com/download/list?cat=64"

and just creating the *one* tiny patch that mattered. Not only that, but 
openwrt head for the 3800 has working leds and ipv6, which the one on 
the above site's betas do not. If anyone has a 3800, please give this 
smoketest a try.

(I'm told the wndr3700*v3* is basically the same as a 3700v1 - not 
enough flash! So I hope the 3800 works out if the 3700v2 stops being 
available.)

I did one of the aforementioned experiments (I will write it up after 
linuxcon), showing how much lower latency txqueuelen of less than 40 was 
than 1000, under load...

and a talk last week at the lincs lab, then two in oslo this week... and 
I'm in Prague now for linuxcon where I am happily NOT speaking, just 
planning to kibitz on john linville's talk (hi john!)... and after that 
I'm going to buckle down for the winter and start attacking some of the 
harder, higher level stuff left on the roadmap.

I'll make a call on 3.1 after linuxcon. Thoughts welcomed... help 
welcomed...

I also have a raft of tiny patches queued up - notably what I think is a 
fix to ipv6 related 802.11e classification that I've yet to test, and a 
revamp of the openwrt qos system, and I finally got a spare box that I 
can do debloat-testing kernel work on without risking my laptop's life...

And I keep struggling to remember the name of this GREAT jazz club in 
paris... I got a lot of work done there one night while grooving on 
it... and is anybody else in prague this week?

-- 
Dave Täht


[-- Attachment #2: dave_taht.vcf --]
[-- Type: text/x-vcard, Size: 204 bytes --]

begin:vcard
fn;quoted-printable:Dave T=C3=A4ht
n;quoted-printable:T=C3=A4ht;Dave
email;internet:dave.taht@gmail.com
tel;home:1-239-829-5608
tel;cell:0638645374
x-mozilla-html:FALSE
version:2.1
end:vcard

Re: [Bloat] an update on cerowrt-rc7

[Riccardo Giuntoli]

[-- Attachment #1: Type: text/plain, Size: 4925 bytes --]

What about consider to give a version to people that want to try cerowrt on
other atheros based platforms?

Like all the http://routerboard.com ones?



On Wed, Oct 26, 2011 at 9:02 PM, David Täht <dave.taht@gmail.com> wrote:

> A new smoketest of rc7 is out:
>
> http://huchra.bufferbloat.net/**~cero1/rc7-smoketests/rc7-**smoketest5/<http://huchra.bufferbloat.net/%7Ecero1/rc7-smoketests/rc7-smoketest5/>
>
> Which is basically linux-3.0.8 + openwrt head + support for the wndr3800
>
> CeroWrt IS independently buildable now:
>
> http://www.bufferbloat.net/**projects/cerowrt/wiki/**
> Building_Cerowrt_on_your_own_**machine<http://www.bufferbloat.net/projects/cerowrt/wiki/Building_Cerowrt_on_your_own_machine>
>
> It's getting easier to install:
>
> http://www.bufferbloat.net/**projects/cerowrt/wiki/OCEAN_**
> CITY_INSTALLATION_GUIDE<http://www.bufferbloat.net/projects/cerowrt/wiki/OCEAN_CITY_INSTALLATION_GUIDE>
>
> and the roadmap remains overwhelming.
>
> http://www.bufferbloat.net/**projects/cerowrt/roadmap<http://www.bufferbloat.net/projects/cerowrt/roadmap>
>
> ...
>
> I was feeling kind of discouraged about the prospect of renewing work on
> cerowrt this month.
>
> The goal has been to get to where the OS and drivers are stable, then do a
> bunch of *repeatable* bloat-related experiments - then kick back into a
> development mode with those results.
>
> we've been spending increasing amounts of time validating the lowest levels
> of the stack - re-analyzing the original bufferbloat data, as one example -
> and we've designed a couple experiments that should reliably show
> bufferbloat AND hopefully various mitigations against it, once they exist -
> that can be repeated by engineers (in front of management!) and students
> using the thing.
>
> and I've been thinking really hard about what a good API to Minstrel would
> be...
>
> All of which was more fun that getting cerowrt from rc to 1.0.
>
> Because, on the whole, the last rc4 is working pretty good.
>
> AND the prospect of building an entire OS AGAIN and burning it into the
> machines at bloatlab #1 was a bit daunting. Not only that, but enough fixes
> had gone into the linux-3.1 release (mostly ipv6 related) that I was
> thinking it would be best to aim for that as a final target for 1.0 than
> stick with 3.0...
>
> And THEN I was pleased to see people actively pointing to the project
> today.
>
> http://ask.slashdot.org/story/**11/10/25/1429235/ask-slashdot-**
> dd-wrt-upgrade-to-80211n<http://ask.slashdot.org/story/11/10/25/1429235/ask-slashdot-dd-wrt-upgrade-to-80211n>
>
> darn it... so my enthusiasm is back closer to where it belongs, and I got
> that smoketest out. Darn it, bruce...
>
> So.
>
> My thx to Petri Rosenström for simply *ignoring* the steaming pile of
> patches for the wndr3800 that are on
> "http://www.myopenrouter.com/**download/list?cat=64<http://www.myopenrouter.com/download/list?cat=64>
> "
>
> and just creating the *one* tiny patch that mattered. Not only that, but
> openwrt head for the 3800 has working leds and ipv6, which the one on the
> above site's betas do not. If anyone has a 3800, please give this smoketest
> a try.
>
> (I'm told the wndr3700*v3* is basically the same as a 3700v1 - not enough
> flash! So I hope the 3800 works out if the 3700v2 stops being available.)
>
> I did one of the aforementioned experiments (I will write it up after
> linuxcon), showing how much lower latency txqueuelen of less than 40 was
> than 1000, under load...
>
> and a talk last week at the lincs lab, then two in oslo this week... and
> I'm in Prague now for linuxcon where I am happily NOT speaking, just
> planning to kibitz on john linville's talk (hi john!)... and after that I'm
> going to buckle down for the winter and start attacking some of the harder,
> higher level stuff left on the roadmap.
>
> I'll make a call on 3.1 after linuxcon. Thoughts welcomed... help
> welcomed...
>
> I also have a raft of tiny patches queued up - notably what I think is a
> fix to ipv6 related 802.11e classification that I've yet to test, and a
> revamp of the openwrt qos system, and I finally got a spare box that I can
> do debloat-testing kernel work on without risking my laptop's life...
>
> And I keep struggling to remember the name of this GREAT jazz club in
> paris... I got a lot of work done there one night while grooving on it...
> and is anybody else in prague this week?
>
> --
> Dave Täht
>
>
> _______________________________________________
> Bloat mailing list
> Bloat@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/bloat
>
>


-- 
Name: Riccardo Giuntoli
Email: taglio@gmail.com
Location: Canyelles, BCN, España
PGP Key: 0x67123739
PGP Fingerprint: CE75 16B5 D855 842FAB54 FB5C DDC6 4640 6712 3739
Key server: hkp://wwwkeys.eu.pgp.net

[-- Attachment #2: Type: text/html, Size: 5863 bytes --]

Re: [Bloat] an update on cerowrt-rc7

[David Täht]

[-- Attachment #1: Type: text/plain, Size: 3690 bytes --]

On 10/27/2011 02:40 PM, Riccardo Giuntoli wrote:
> What about consider to give a version to people that want to try 
> cerowrt on other atheros based platforms?
>

While it is easy to build for multiple platforms, it is impossible to 
test on more than a very few. Secondly there are currently hard coded 
differences in the filesystem(s) and device sets that make getting 
initial filesystems 'perfect' a difficult, tedious process. Fixing the 
first would be easy. The second, less so.

Merely having a difference between the 3700 and the 3700v2 has confused 
many users!

Earlier on in cerowrt we planned on doing the ubiquity nanostation M5 
(long distance wifi tests) and the dreamplug, and a kvm, and a usb stick 
based version... basically anything that met our requirements of at 
least 64MB ram and 16+MB flash.

All of which we did do at one point or another, and the net result was 
that the testing cycle took so long as to dominate the actual effort 
involved in doing anything else!

and we're trying to fix bufferbloat here, with a rapid 
compile/test/debug cycle designed to stay within the kernel development 
window. I'm happy we're staying in that window, very unhappy with the 
amount of effort required to do just that much.

But:

It should be very easy to adapt the existing cerowrt build process to 
attempt a build of your own for whatever hardware you might like -

http://www.bufferbloat.net/projects/cerowrt/wiki/Building_Cerowrt_on_your_own_machine

All you really have to do after that is a 'make menuconfig' and select 
different hardware than the default, keeping in mind that the filesystem 
is designed for a dual channel radio and non-bridged ethernet devices 
and > 8MB of flash is required...

I can EASILY do builds for all the ar7100 based hardware, but I don't 
know which of the several dozen boards supported in the build system 
meet the above requirements. I was thinking of adding routerstation pro 
support in a a future build and maybe a d-link...

And:

Patches gladly accepted as always.


> Like all the http://routerboard.com ones?
>
There are multiple issues with this idea.

Most importantly: Dave does not scale.  (see above)

Secondly, microtik ships a (very user friendly, IMHO) version of an 
integrated router distro on their own, called RouterOS.  It's high on my 
list to poke into RouterOS, (does bufferbloat truly live EVERYWHERE??) 
but I have a very, very, very long list... I would appreciate someone 
(else!) taking a look at what they do to alleviate the bloat, reporting 
back on what kernel they use, how good is there ipv6 support, what size 
buffering they use, what they are doing for QoS/AQM etc.

Or better, someone from microtik participating on the list!

... as it's my hope that talking about the issue of bufferbloat 
coherently, often, and in public is best - and cerowrt and 
debloat-testing's model of figuring out problems, testing/fixing them, 
and pushing them upstream to kernel head/openwrt head/package head means 
that everybody wins and we don't have to look all that much at other 
distros.

Speaking of that, I've been lax of late in pushing some stuff upstream...

PS I would VERY MUCH like to get out of cerowrt's current wifi 
monoculture however and get a good grasp of other wifi chip designs, 
before even thinking about an debloating api for wireless-n - I am glad 
that broadcom and iwl are now mostly open, only marvel is lagging on that...

(I do like atheros's architecture, however. the iwl cards scare me)

So at some point after cero gets more stable I'm going to go start 
hacking on debloat-testing again.

PPS cerowrt rc7-smoketest7 is out and working pretty good!

-- 
Dave Täht


[-- Attachment #2: dave_taht.vcf --]
[-- Type: text/x-vcard, Size: 204 bytes --]

begin:vcard
fn;quoted-printable:Dave T=C3=A4ht
n;quoted-printable:T=C3=A4ht;Dave
email;internet:dave.taht@gmail.com
tel;home:1-239-829-5608
tel;cell:0638645374
x-mozilla-html:FALSE
version:2.1
end:vcard