Subject: Re: No buffer space available (again)
To: S.J. Borrill <sjb42@cus.cam.ac.uk>
From: Neil A. Carson <neil@causality.com>
List: port-arm32
Date: 08/24/1998 13:37:37
Hi,

Jason or someone may be able to shed more light on this?... (see below
the quote)

S.J. Borrill wrote:
> 
> Yes, it's another mail from me about No buffer space available, this time
> on a different platform. I've just got a new CATS with Dec Ethernet.
> However, networking is proving to be a big problem. I've got it currently
> connected only to a RiscPC running 1.3F. I tried to FTP a file between the
> two machines, but it stalled. So I tried pinging from the CATS to the
> RiscPC. About 10 packets went out fine, but then it just stopped.  After a
> large time (up to 60 seconds) all the packets came back at the same
> time (i.e. no packets lost, but with 60000 ms ping times). It then chugged
> on for a bit more before halting and doing the same thing. This time ping
> reported no buffer space available. Doing an ifconfig de0 down; ifconfig
> de0 up briefly got it to work.
> 
> This isn't just a one-off either; I had exactly the same behaviour last
> week with a different RiscPC, with a different ethernet card and on a
> different network (but the same CATS).
> 
> systat mbufs on either of my CATS shows a full bar for data, whereas it
> looks a lot more sensible on the RiscPC. Note, my other CATS board which
> is using an older kernel works fine.
> 
> So, my question is 'why am I getting the same errors on all sorts of
> different hardware when I'm doing basic things?'. This is a serious
> problem, I need to rely on NetBSD as a server solution and I'm finding it
> increasing unreliable.

I've managed to recreate this. However, since I know nothing about the
networking code I don't know what the problem was, but maybe someone
more enlightened may be able to answer. The way I recreated it was as
follows...

On my network here, I have four machines: A Shark, a CATS, a PC and a
laptop PC, connected via 10BaseT on a cheap Frys hub. I recreated this
as follows. None of the machines have been up more than a couple of
hours... Telneting into the CATS, the mbuf usage is around 30 (our of
60), half way across the screen, for data. Headers and sockets were at
zero.

Flood ping the machine from my desktop PC, it goes up by 5 or so, on the
data bit. Flood ping my PC back from the CATS, and it goes up a wee bit
more. Now... Flood ping the CATS from my laptop as well, and suddenly
the data bar leaps up all the way across the screen, and the flood ping
from the CATS to the PC reports:

> talisker# ping -f oban
> PING oban.causality.com (192.168.3.23): 56 data bytes
> .
> ping: sendto: No buffer space available
> .
> ping: sendto: No buffer space available
> ......
> ping: sendto: No buffer space available
> ..
> ping: sendto: No buffer space available
> ..
> ping: sendto: No buffer space available
> ping: sendto: No buffer space available
> ..^C

I presume this isn't the defined behavour, but I don't know. Can someone
who knows more about this comment, ot at least tell me if I _have_
recreated the same problem so that we can start hunting?

	Chers,

	Neil

-- 
Neil A. Carson