[241] in linux-net channel archive
Re: Problem reported on c.o.l.d.s (was (fwd) Re: 1.2.5: Still losing 5-9 Megabytes MEMORY per day!!)
daemon@ATHENA.MIT.EDU (Marek Michalkiewicz)
Thu Apr 27 17:27:04 1995
From: Marek Michalkiewicz <marekm@i17linuxb.ists.pwr.wroc.pl>
To: linux-kernel@vger.rutgers.edu
Date: Thu, 27 Apr 1995 21:45:07 +0200 (MET DST)
Cc: linux-net@vger.rutgers.edu
In-Reply-To: <9504271314.AA07150@mail.foa.ska.com> from "Huw Rogers" at Apr 27, 95 03:14:09 pm
> This has been reported on c.o.l.d.s a number of times, with no replies from
> any developers. Are the relevant people aware of this?
Hi, I am the one who wrote about the need for a limit of memory
used by network buffers, and about sending ICMP Source Quench when
close to the limit. What do you think about that?
We already keep track how much memory is used by network buffers
(net_memory in linux/net/inet/skbuff.c), so it should be easy to
drop packets if network buffers use too much memory.
I don't know how hard would be to send ICMP Source Quench, though.
I am not very familiar with the networking code (yet). Alan?
Maybe at least the easy part (dropping packets when too much memory
is used by network buffers) could go into 1.2.x as a bug fix? :-)
This will not help if there is a memory leak somewhere, but at least
should prevent using all memory by packets forwarded from Ethernet
(fast) to PPP (slow).
BTW, nothing is printed by the following command:
grep net_memory linux/fs/proc/*.c
Maybe this info should be added to /proc - currently something like
shift-scroll lock seems the only way to access this information.
Regards,
-- Marek Michalkiewicz <marekm@i17linuxa.ists.pwr.wroc.pl>
