[Bloat] CeroWRT for x86?

[Dave Taht]

On Wed, Apr 11, 2012 at 7:23 PM, Christopher Sheats <yawnbox at gmail.com> wrote:
> Hi,
>
> I'm looking to process 1Gbps, be it DDoS, Tor relay, or BitTorrent
> traffic on a WAN port. Even though the WNDR3800 has a gigabit WAN
> port, I'm not certain how successful it would be, especially with
> sustained traffic and keeping the components cool.

I have run sustained ipv4 traffic in excess of 440Mbit through the
router for days at a time. That has been more (less firewall rules,
more tuning, less bugs, les aqm), and less (ipv6 traffic, hefty aqms
like qfq, etc).

I don't have baseline numbers at present, we're still tuning.

That said, I think your core applications requires more oomph. On the
high end, octeon appears to be making some 100% open source product
available that can scale easily to 10GigE. Certainly most x86
platforms today can saturate gigE with TSO/GSO off.

> OpenWRT has an x86
> version, will CeroWRT?

We have no plans for any other versions of cerowrt on any other
hardware. There's no budget, no people, and no time for that. It is a
platform for network research, and my hope is, when this release is
done, to actually get some research done on top of it.

The plan has always been to leverage x86 development by tracking the
mainline kernel closely.

Being able to innovate, test, and explore bufferbloat, ipv6, and
security issues with two well understood 100% open source platforms
(x86 servers, mips routers) and push (the good) code back into the
mainline kernel, applications, and distributions such as openwrt,
ubuntu, redhat, etc., is what we are trying to do.

We get a lot of 'platform of the month' questions, and with our
resources, doing merely two platforms has been a mighty stretch
sometimes. Everything we do will end up upstream, in openwrt and
elsewhere, or so we hope.

In fact, it's generally easier to be working in the latest Linux
kernels on x86, prototyping things that may or may not work on weaker
hardware. The turn around times for new kernels and testing is
substantially less, and the debugging tools vastly superior.

As an example, the core new technologies in linux-3.3 relevant to
bufferbloat (BQL, SFQ, SFQRED, QFQ) are in, linux-3.3 and can run on
anything that can run 3.3.

Last quarter we worked out of net-next on x86.
Earlier this quarter the stuff that proved out landed in 3.3
A little bit later on, we got 3.3 running on mips, after doing a lot
of backporting to 3.2 as 3.3 stabilized....

There's nearly nothing out of tree in cerowrt-3.3 at this time except
some patches specific to the wndr3700 series of hardware, and a few
ideas under test.

As all the sources are available, please feel free to port cerowrt to
x86, or just wait a month or two as the patches that prove out migrate
upstream. But: Do give 3.3 a shot on a mainstream x86/x86_64
distribution with the debloat tool. AQM isn't just for routers
anymore, but desktops, wireless clients, and servers.

For more information, please see:

http://europa.lab.bufferbloat.net/cerowrt/ (yes, this is a web page on a router)

and the cerowrt-devel mailing list, and the cerowrt project wiki.

> Thank you.

De nada.
>
> --
> Christopher Sheats
> yawnbox at gmail.com
> _______________________________________________
> Bloat mailing list
> Bloat at lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/bloat



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