According to Lars Petterson: While burning my CPU.
>
> Hi all!
>
> > The solution seems to be the maxkeyup parameter in the
> > /etc/z8530drv.conf
> > file. After increasing the value from 7 to 15 all seems to work well.
>
> At sk6sa we uses 30 and still have problems. Exactly when/why the problems
> occure is hard to tell, sometimes we have to restart the computer
> once a day, but other times it takes much longer, we have had the
> computer running for 35 days without problems, so it is very hard
> to see a trend here... :-)
As i mentioned in my last post on this problem, there is (in my case) no
need to reboot the machine, just try ifconfig scc# down then up the
interface add a route, then use ping to test it, if your ping is returned,
you will then need to restart ax25d and netrom + other deamons like 'beacon'
etc. Thats what i do in a script which runs under CRON every # minutes.
This problem has been there for many kernel versions and indeed older driver
versions.
The only thing i can say about its cause is, there is in my case only one
factor i can see which kills the interface, netrom braodcasts when the
interface is busy, after a netrom broadcast of "many" nodes + 150 bingo,
down goes the interface.
When there is no traffic, ax25 or IP, (or few nodes to be braodcasted),
there "seems" to be no problem(s).
This would lead one tho point to a (memory) buffer problem, i use;
bufsize 2048
rxbuffers 20
txbuffers 20
Interface MTU 512 window 1888 also tryed with 944 same results.
Maxkey seems not to make any differance at all.
>
> > We
> > are using a clean 2.0.36 kernel, ax25-utils-42a and z8530drv-2.4c utils
> > without any patches or changes. The card is a PA0HZP opto SCC card (4
> > ports).
>
> We uses the same except the kernel, 2.0.33, with everything compiled in,
> i.e. no modules.
2.0.33 also on the problem server but patched with the nessasary axmodules
patch
>
> > It seems that the problem only occurs on very busy hours. On that port
> > we
> > also send unproto beacons from Xfbb and no node (national regulations).
>
> Is this the port that have the most traffic? We seem to have problems
> with the ports not having that much traffic. But the problem seem to
> show up when we have much traffic on another port. Xfbb sends data
> on all ports on one of the computers, on the other one we do not
> use Xfbb so the problem is not associated with Xfbb.
>
> > Also the TX and RX buffer is set very high at 25 and the bufsize has a
>
> We uses 16.
>
> > value
> > of 2048 bytes.
>
> Bufsize is 384.
>
> > Please correct me if I'm wrong, the value of maxkeyup is to prevent a
> > long
> > transmission time.
>
> Yes.
>
> > And when it must have a very high value to work,the
> > maxkeyup parameter doesn't have any meaning.
>
> It depends if you want to be channel hog or not :-)
>
> Remember that a packet at 1k2 takes approximatly 1.8 sec to send.
> If you uses a maxframe of 7 you may very well send continously for
> 12.6 sec. So 7 seem too low for a port using maxframe 7 (for tcp/ip-
> traffic we may very well send for even longer periods.)
I dont think its an interface speed issue here, i have ports on 1k2 4k8 and
9k6, any port using netrom "seems" to cause failures, no netrom and no
soft-dcd will allow the interfaces to stay up for eternity, or they do with
me.
Another system run by a friend has been up for,
*** Linux version 2.0.29 is up for 161 days, 14 hours, 54 minutes, 23
seconds.
He does not have this problem and only broadcasts 3 or 4 netrom nodes on
any 1 scc interface.
We both use PA0HZP cards.
As to the netrom broadcasts "being" the problem i cant say, all i am saying
is it "smells" like it.
I wonder if others who have this problem are transimitting large netrom node
broadcasts.
>
> 73 de Lars, sm6rpz
>
> --
> Lars E Pettersson | Chalmers University of Technology
> [EMAIL PROTECTED] | Gothenburg, SWEDEN
>
--
Regards Richard.
[EMAIL PROTECTED]
Merry Xmas to all, and may all your troubles be small (ones).
