Date: Thu, 30 May 1996 14:05:02 +0300 (EET DST) From: "Andrew V. Stesin" <stesin@elvisti.kiev.ua> To: wilko@yedi.iaf.nl (Wilko Bulte) Cc: stesin@elvisti.kiev.ua, questions@freebsd.org, scsi@freebsd.org Subject: Re: Mysterious crashes of FreeBSD gateway -- caugh it(?) Message-ID: <199605301105.OAA17791@office.elvisti.kiev.ua> In-Reply-To: <199605292055.WAA11549@yedi.iaf.nl> from "Wilko Bulte" at May 29, 96 10:55:46 pm
next in thread | previous in thread | raw e-mail | index | archive | help
Hi again, # # As Andrew V. Stesin wrote... # # > When doing a massive TCP transfers to the 1.1.5-connected subnet, # > or even ping -f, a high network load was inspired on a gateway # > machine (receive a packet -- forvard it -- send redirect). # > NE2000 worked fine, but NCR driver started to through messages # > about I/O errors, "NCR dead", etc. Then gateway rebooted itself. # > # > What I want to ask. # > # > What might be the source of that trouble? Poor motherboard quality, when # > an overloaded (?! is it an overload?) either ISA or PCI bus forces NCR # > to go asleep? Or is it a bug in TCP/IP stack or IPfilter, or # > their interraction with NCR driver, tickled # > with the nessesity to process IP/ICMP packets at a very high rate? # # > I strongly suspect a hardware problem, but maybe there are other # > opinions? # # I have also seen this happen with a SCSI device that was not 'liked' # by the ncr driver. After changing the firmware of the _device_ (a tape) # the problem went away. I doubt here. Another (similar SCSI hardware) box is rock solid with the same configuration of NCR and Conner drive. The two differences between boxes are: 1) motherboards, 2) gateway has IPfilter inside. Is it possible, for example, that a complex set of filtering rules are processed at a high priority in the kernel while packets are "accepted"-"filtered on input"-"forvarded"-"filtered on output"-"output" ? and for the time of processing packets NCR interrupts are missed? (I don't know much about the things like splnet() etc. :( for a pity. Where can I read about them, BTW?) # A 'dead' ncr is a perfectly acceptable reason for the system to reboot # itself (at least it was on mine). Definitely. # I admit this does not correlate well with your ethernet board swapping, # but it is another datapoint on 'dead' ncrs # # Wilko # _ __________________________________________________________________________ # | / o / / _ Wilko Bulte email: wilko@yedi.iaf.nl # |/|/ / / /( (_) Private FreeBSD site - Arnhem - The Netherlands # -------------------------------------------------------------------------------- # -- With best regards -- Andrew Stesin. +380 (44) 2760188 +380 (44) 2713457 +380 (44) 2713560 "You may delegate authority, but not responsibility." Frank's Management Rule #1.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199605301105.OAA17791>