[Cerowrt-devel] speed problems with later versions of cero?

[Dave Taht]

I've now had three private reports of "speed" problems with the latest
cerowrt releases.

All of them take serious uptime to occur. And unfortunately all the
bug report(s) were, "oh, it
got slow after everybody got home/ran a couple days, so I went back to
the prior router".

That's not helpful, and could well be new users blaming the change
rather than something else, but it is possible that this is happening,
so... IF that happens to anyone here, I'd like
to request a couple things in roughly this order

tcpdump -i ge00 -s 128 -w /tmp/myproblem.cap # for 10 seconds hit
cntrl-c Send me the file
pinging the box from the outside world # got packet loss?
dmesg > /tmp/dmesg.log # send me the file
logread > /tmp/logread.log # send me te file
run top and see if there is a daemon running away
tell me what ISP's network you are on...

restart the aqm system and see if it goes away
/etc/init.d/aqm restart

try nfq_codel and/or pie instead of fq_codel

1) I am aware of an old bug with odhcp6 where it would start hammering
the network with
requests for an ipv6 address in a tight loop. I don't know what
triggers it. (with AQM set right it is hardly noticeable but
"feelable". However if your bandwidth setting is even mildly wrong,
boom life gets bad and STAYS bad)

If you look at the packet capture it will be filled with dhcpv6
requests if this is what is happening.

Right now, I have a box dropping nearly every other packet for some
reason. It's not in a place I can easily get at until tomorrow. It's
been up for 90+ days and only started doing this recently

--- X.Y.Z.Q ping statistics ---
60 packets transmitted, 42 packets received, 30.0% packet loss
round-trip min/avg/max/stddev = 37.788/44.205/76.607/6.609 ms

It's my hope it's the above problem and is occurring now that comcast
is rolling out ipv6

Off the top of my head I don't remember what you need to disable to
turn it off. Certainly comment out the ge01 entry to turn it off.

2) Random number generation has changed.

rngd is still enabled, so we're not running out of random numbers, but
we could have other problems... check entropy?

3) There could be a daemon running away (run top)

4) Could have a new instruction trap problem under certain conditions

take a look at the /sys

dmesg
logead


Tony

-- 
Dave Täht