[Cerowrt-devel] stanford talk/deluged in hardware/yurtlab
dave.taht at gmail.com
Sun Feb 3 12:18:31 EST 2013
I spent most of the last week prepping a talk at stanford; trying to dispel
a few delusions.
The core points that I tried to make:
* Mice and ants account for a lot of traffic and sfq helps a lot on
* both inbound and outbound rate limiting, fq, and queue management are
* rrul is cool
While it was filmed, we (kathie nichols, eric dumazet, luigi rizzo, and the
students) really got distracted with some current issues and way off the
slides. It was not my finest hour. I think the modena talk went much
I should probably have brought out that slides 20 (pfifo_fast) and 23
(simple_qos, basically) were taken 1 minute apart, on the same
connection.... twice the utilization, 1/15th the latency and jitter... to
me these two slides should have had everyone in the room running out to
make this stuff work on their head ends and cpe a few seconds later.
The rest of the week I spent starting an internet draft, justin and I
building up the yurtlab testbed, upgrading hardware there, adding
instrumentation and trying to get to where I can track various data points
like power failures, packet drops, etc.
* Yes, I'm trying to do an internet draft requiring some form of fq + aqm
on the edge networks.
I'm really terrible at this.
The deployment (3.7.4-4 and 3.7.5-2 based on wndr3800s, pico station m2hp,
and nanostation M5) just survived a 24 hour dance party and total
saturation of the links... I really like the m2hps in particular.
I picked up a couple samsung 2955-dw laser printers. At 100 bucks each,
they were a steal. And it turned out they supported ipv6, and worked well
with linux out of the box. So, yea, I can fiddle with dhcpv6 on another
sort of box.
While I'm talking about hardware, these power controllers are excellent.
And I'd got nut in cerowrt to work with these so I'd be able to collect
Turned out really useful, had one flaky spot where the power was 90v or
Lastly, anybody want a couple picostation 2HPs? They only have 4MB of flash
and don't do ipv6 and are pretty useless for my purposes....
* the quest for new hardware
I continue to look for new hardware to support, as the wndr3800s remain
scarce. I rather appreciate the donations to cerowrt (yurtlab is a
different budget), it makes the quest for this stuff a lot easier, and
mentally "gives me a budget" for the quest.
I'm looking at this board:
I'm grumpy, as it doesn't have an esata interface internally, apparently.
It would have been good to have one of those hooked up to a good flash disk
for active measurements, as a core bottleneck on using a mini-sd card for
storage is the pathetic write speed on that sort of card. Aside from that,
it looked interesting. I also ordered a dreamplug, another raspberri pi...
but, darn it, I can't find the perfect hardware. I've been trying to sort
out what I want, perhaps I can't do it all on one piece of gear or have to
design something that does do things right. I like the zedboard a lot.
At the moment I hope that a guruplug or equivalent can be made to capture
8Mbit/sec of packet headers off a mirrored port.
Also, openwrt has added support for a new atheros reference board, a couple
Fixing bufferbloat with cerowrt:
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Cerowrt-devel