[Codel] proof of concept on embedded

[Codel] proof of concept on embedded

[Dave Taht]

I patched in the latest codel patch into cerowrt, bumping up the
control_law cache to 1000.

It worked. First time.


I was able to run ~230Mbit/sec ethernet traffic through it, which is
comparable to both sfq and pfifo_fast on this hardware (for reference
this is the netgear 3800 680Mhz mips box with a 16 bit memory
interface).

Might have been able to get more had I increased BQL's setting (which
is set to 3028, presently). With BQL in auto mode, BQL finds operating
points between 64k and 131k at these speeds.

RTTs stayed low.

I knew, based on the experiments back in november, that having the
dependency on timestamping was feasible, but it was another thing to
actually see it all work.

With htb on with a modified simple_qos.sh script to use codel rather
than sfqred, with htb set to rate limit at 2Mbit, using netperf via
wireless, I only got .86Mbit/sec for one stream and 1.4Mbit for 4. But
I'm currently willing to write that off to the unrelated bug 379.

root@codel:~# tc -s qdisc show dev ge00
qdisc htb 1: root refcnt 2 r2q 10 default 12 direct_packets_stat 9
 Sent 13446791 bytes 9738 pkt (dropped 0, overlimits 29695 requeues 0)
 backlog 0b 2p requeues 0
qdisc codel 110: parent 1:11 [Unknown qdisc, optlen=32]
 Sent 15667 bytes 129 pkt (dropped 0, overlimits 0 requeues 0)
 backlog 0b 0p requeues 0
qdisc codel 120: parent 1:12 [Unknown qdisc, optlen=32]
 Sent 13428070 bytes 9586 pkt (dropped 2710, overlimits 0 requeues 0)
 backlog 0b 0p requeues 0
qdisc codel 130: parent 1:13 [Unknown qdisc, optlen=32]
 Sent 0 bytes 0 pkt (dropped 0, overlimits 0 requeues 0)
 backlog 0b 0p requeues 0
qdisc ingress ffff: parent ffff:fff1 ----------------
 Sent 852437 bytes 10227 pkt (dropped 0, overlimits 0 requeues 0)
 backlog 0b 0p requeues 0


Regrettably, due to another bug I've been battling for a month I can't
run traffic at that speed through it for more than 2 minutes or so.
( see http://www.bufferbloat.net/issues/379 371 and 360 for details ).
And last I looked qfq was still messed up on this arch.

But I've run plenty of traffic through it at normal speeds, so it's usable.

A test build for those of you that have this router is up at:

http://huchra.bufferbloat.net/~cero1/3.3/dev/3.3.4-5/

(didn't get the new iproute2 stuff on it on this pass though)

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

Re: [Codel] proof of concept on embedded

[Dave Taht]

On Sat, May 5, 2012 at 1:47 PM, Dave Taht <dave.taht@gmail.com> wrote:

>
> Regrettably, due to another bug I've been battling for a month I can't
> run traffic at that speed through it for more than 2 minutes or so.

to clarify:

at *230+* Mbit for more than 2 minutes or so.

Thus far htb + codel 20Mit down, 2Mbit up, has been fine.


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