lkml.org 
[lkml]   [1997]   [Sep]   [27]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Real pre-10 krash boom bang.
On Fri, 26 Sep 1997, Daniel Ryde wrote:

> I tried the real release of Pre10 (linux-2.0.31-pre10.tar.gz) and
> unfortunatley it had only one hour of uptime, then black screen and total
> freeze. Sorry but I can't test this anymore, as these machines are in real

I did my best to stress 2.0.31-pre10 (the new snapshot from 2 days ago)
compiled SMP on a dual P120. With a make -j of the kernel, an emacs
session with 2 kernel source tars loaded, and 6 dd processes reading both
SCSI drives to /dev/null using various block sizes under 1k, I had the
load up at 10-25, nearly at the end of the kernel compile, I got a screen
full of complaints from cpp about closed pipes and the make terminated.

System is Tyan Tomcat II 512kb (bought after the COAST bugs were fixed) 2
Intel P120, 64mb true parity RAM, 60mb swap, IBM DALS DALS-3540, Seagate
ST3620N, Toshiba XM-5401CD, all on a NCR 810, Buslogic 946 with nothing on
it, ATI Mach64, SMC Elite16T.

I've reproduced this, and each time got a screen full of

/mnt/sdb1/linux-pre10-final-fdt/include/net/tcp.h:256: virtual memory
exhausted
In file included from tcp_output.c:40:
/mnt/sdb1/linux-pre10-final-fdt/include/net/tcp.h:206: virtual memory
exhausted
cpp: output pipe has been closed
cpp: output pipe has been closed

From what I've read, I suppose this might be normal since the emacs
session eats a good 20mb (most of which gets swapped out). The strange
thing is both times it's terminated at the same point. I redirect stdout
to a file, and I have:

-rw-r----- 1 root root 81136 Sep 26 16:26 out-j
-rw-r----- 1 root root 81136 Sep 27 01:03 out-j-2
-rw-r----- 1 root root 95137 Sep 26 13:28 out-sloth

out-sloth is a full successful compile. j and j-2 are make -j's of the
same .config that both failed shortly before completion.

The full procedure was first make -j (also set -j in Makefile) zImage
modules >out-j-2 & then 6 dd's to /dev/null with block sizes of 256, 512,
64 of each of the disks. The different block size runs were started
staggered over several minutes. emacs linux-2.0.31-pre10.tar.gz
linux-pre-2.0.31-9.tar.gz was started and stopped a few times and finally
left to sit in hexl-mode, and a ping -f -s 1024 of sloth was done from
another local system. sloth was also running gated...and after everything
finished it seemed to be doing fine. During the testing, interactive
response varied from good (immediate echo) to terrible (I could watch
the letters in "uptime" appear over the course of nearly a minute after
typing it), but gated (doing ospf) doesn't seem to have missed any
hello's.

I don't think I've beat this hard on the previous pre kernels, so I don't
know if they'd act differently. I still have several around though.
There were no "Couldn't get a free page" messages, and it was using the
default freepages settings (128 192 256).

------------------------------------------------------------------
Jon Lewis <jlewis@fdt.net> | Unsolicited commercial e-mail will
Network Administrator | be proof-read for $199/message.
Florida Digital Turnpike |
______http://inorganic5.fdt.net/~jlewis/pgp for PGP public key____



\
 
 \ /
  Last update: 2005-03-22 13:40    [W:0.081 / U:0.068 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site