Re: memory & filesystem corruption under heavy load?
Robert Riggs (rriggs@tesser.com)
Sat, 06 Apr 1996 02:01:21 -0700 (MST)
On 06-Apr-96 Robert L Krawitz wrote:
>> cc: gzenie@hampshire.edu, linux-kernel@vger.rutgers.edu
> Date: Fri, 05 Apr 1996 17:09:18 -0500
> From: garth zenie <gzenie@espresso.hampshire.edu>
>
> On Fri, 5 Apr 1996 10:18:48 -0500 Robert L Krawitz wrote:
>
> > What does the corruption look like?
>
> from pgtable.h:
>
>That's much different from what I see. I see random bits flipped here
>and there.
>
> i failed to mention in my first message (i should fall out of the
> habit of making posts when i am on the verge of unconsciousness), that
> there is a strong possiblity that there is a SCSI factor to all of
> this:
>
> **Tagged Queuing** **Kernel Version** ** Corruption **
> off 1.2.13 never
> off 1.3.77 never
> on 1.2.13 never
> on 1.3.77 always
>
>My interface (53c810) doesn't support tagged queueing (well, it does,
>but the driver doesn't yet).
>
> P166 (brand new)
> 32MB EDO RAM
> 256k pipelined cache
> Micronics M54Hi Motherboard
> PCI 3c590 ethernet (vortex)
> PCI Buslogic BT-948
> PCI Triton IDE
> PCI Diamond Stealth Video VRAM
> ISA Soundblaster 16
> Conner CFP2107S SCSI drive
>
>Mine:
>
>P90
>32 MB FPDRAM
>Intel Plato motherboard
>NCR 53C810
>ATI Winboost (2 MB DRAM Mach64)
>Sound Blaster AWE32
>Seagate ST12400N & 32550N (2 1K filesystems on 12400 and 1 4K on
>32550, plus a couple of FAThead partitions, which presumably don't matter)
>WD AC2420 IDE (1 FAT partition)
Robert,
Can you try Gerard Roudier's NcrBsd-1.8 driver? I have had complete
SCSI locks using that driver (with 'make -j zImage'). It is a bit
faster on my system than the standard kernel driver and supports
tagged queuing. Corruption occurs more frequently with tagged
queuing enabled, but isn't required. Unfortunately I have absolutely
no kernel messages from when this occurs. The machine locks up solid.
The HD lights stays on. Pings don't work. Can't switch out of X. And
since it's the HD that's screwed, no syslog entries. I always have
a corrupted filesystem afterwards.
I never saw this problem prior to using Gerard's driver. I started
using it around the time 1.3.72 came out.
I'll try to set this up to send syslog entries to the networked
computer and get a little more info in the next few days. I'll
also see if I can somehow reproduce this with Drew's NCR driver.
Rob
(rriggs@tesser.com)