[Cerowrt-devel] Fwd: 3.3.6-2

Sebastian Moeller moeller0 at gmx.de
Wed May 23 20:48:04 PDT 2012


Dear All,

since Dave asked me to post out in the open here it goes:


Hi Dave,

hope you have had a great weekend, Maker Fair sounds sweet. And vacation sounds even better, I hope you have/had a great time there.
	I managed to give 3.3.6-2 a small test drive by now on my wndr3700v2. I have some observations I would like to report just to document them. All my tests are using a single 5GHz wireless client (running macosx 10.7.4) going to test sites on the internet over 30/4 MBit cable internet.
A) under moderate wireless stress I get a lot of allocation failures from slub, like:
[ 1221.664062] ath: skbuff alloc of size 1926 failed
In the routers dmesg. And every now and then the router crashes and reboots (I have not yet found a way to make this happen reliably, it seems to require some uptime)
	It seems that the UDP probes used by http://loki10.mpi-sws.mpg.de/bb/bb.php (short "bb") are quite likely to produce those skbuff failures and also occasionally cause the crashes. If I understand correctly this tool explicitly tries to overload the bottleneck link with UDP packages so it can estimate the worst-case buffering. Alas, the tool is rate limited to a few invocations per day, so I can not really test this hypothesis in any meaningful way. Interestingly both bb and netalyzr start reporting about 3 seconds upstream buffering on a fresh booted router which will change to around 38ms upstream buffering over the course of a day. And after that the router is prone to actually reboot during a run of the 20mbit alternative option of the bb test.  Yet concurrent interactive sessions are reactive no matter whether the reported queue is 3000 or 38 ms, so fq_codel is pure magic.
	My layman's hypothesis is that somehow the UDP stream reveals a bug in the atheros wireless driver, that occasionally takes down the router. So this might be a different aspect of bug 379?
	I will try to understand netsurf better and setup a UDP stream to see whether I can force the router to reboot reliably, as it is all I can report is a spurious reboot. Once I have a robust reproducer I will see whether I can make a recent openwork snapshot crash the same way.

	P.S.: I am currently reading up a bit on IPv6 and home security and it seems things are more complicated than I had hoped...

Best
	Sebastian


P.P.S.: I still wrangling netperf sources to hopefully be able to reproduce the issue (and test the UDP hypothesis)

best
	Sebastian


> 
> On May 14, 2012, at 1:59 PM, Dave Taht wrote:
> 
>> A test release of CeroWrt is now available that has support for Kathie
>> Nichols' and Van Jacobson's new AQM, Codel , and Eric Dumazet's new
>> fair queuing implementation on top of that, fq_codel.
>> 
>> fq_codel is enabled on all interfaces by default. It is vastly simpler
>> than what we were using before (sfqred) and draws upon and improves on
>> the same body of ideas (head drop, fq, timestamping) but is now tied
>> to Kathie and Van's blinding insights as to a good drop strategy, and
>> Eric's successor-to-sfqred ideas as towards head of queue behavior,
>> modern amounts of flows, and cache line optimizations.
>> 
>> There is a simple_qos.sh script that can be set to your uplink and
>> downlink speeds, but no uci interface for it as yet, nor gui. (help on
>> finishing aqm-scripts and the luci interface gladly accepted)
>> 
>> To see all the chocolately goodness of what fq_codel can do to wired
>> and wireless latency, it would be good for more to play with it.
>> 
>> Benchmarks have been very good thus far, and more benchmarks and
>> analysis are highly desired.
>> 
>> Caveat:
>> 
>> This release suffers from an unrelated bug ( #379 ) and should NOT be
>> installed as your main router. I would love to beat this bug because
>> it's the only prio 1 remaining but thus far, no luck. Under lighter
>> loads CeroWrt appears to work just fine, but that's for me. YMMV.
>> 
>> Get it here: http://huchra.bufferbloat.net/~cero1/3.3/3.3.6-2/
>> 
>> 
>> -- 
>> Dave Täht
>> SKYPE: davetaht
>> US Tel: 1-239-829-5608
>> http://www.bufferbloat.net
>> _______________________________________________
>> Cerowrt-devel mailing list
>> Cerowrt-devel at lists.bufferbloat.net
>> https://lists.bufferbloat.net/listinfo/cerowrt-devel
> 



More information about the Cerowrt-devel mailing list