Subject: Re: network hang
To: None <jmarin@pyy.jmp.fi>
From: maximum entropy <entropy@zippy.bernstein.com>
List: current-users
Date: 09/12/1997 06:24:04
>From: Jukka Marin <jmarin@pyy.jmp.fi>
>
>On Fri, Sep 12, 1997 at 01:38:30AM -0700, Wailer wrote:
>> I'm using netbsd as a gateway/workstation.  P133/32 with scsi disks.
>> I'm having an issue which I've seen off and on for a very long time
>> in which the network just clogs up and ceases to move.  I run ppp over
>> a v.34 modem attached to a 16550.  Lockups appear to occur with 
>> heavy network use, but not always.  This last one happened while a
>> client was running cu-seeme, so heavy udp use.
>
>So your PPP links locks up and starts working only after a modem disconnect
>and redial?  I see this every now and then on all my NetBSD 1.2 boxes.  I
>was hoping this problem had been solved by the com driver changes in
>-current.  I thought there was a handshake problem in the com driver which
>sometimes prevented data transmission.. but maybe it's not that, then.  I
>haven't used an RS tester to see the state of the handshake signals but
>I guess I should do that sometime.  My PPP link usually stays up for a few
>days in a row, but on a 486 system we see the problem a few times a day.
>
>The problem only affects one PPP link at a time, so I don't think it's
>a generic networking problem (Ethernet always works just fine).

I get this behavior a lot on my network, with a SLIP link between a
486 running NetBSD, and a DEC Alpha running OSF/1.  However I had
exactly the same problem in a previous configuration, with the same
Alpha and a 386 running Linux.  So I always assumed it was either the
Alpha's fault, or a configuration problem on my network.

I found by experimentation that manually telling all the hosts on my
SLIP-connected network to use a small MTU and small window sizes on
the route through the SLIP host made a drastic improvement in the time
the link would stay up.  The /etc/mygate on any of these hosts looks
like this:

206.20.83.193 -lockrest -mtu 296 -sendpipe 1024 -recvpipe 1024

By manually setting these to small values on all hosts on my local
network, these "freezes" of the SLIP line went from about 2 or 3 times
a day, to once or twice a week.

Maybe this will give someone with a clue some hints about what a real
solution might be...

Cheers,
entropy

--
entropy -- it's not just a good idea, it's the second law.