Re: Linux Stability & cold.system

Doug Ledford (
Mon, 17 Nov 1997 05:09:51 -0600 (CST)

On 17-Nov-97 Alan Cox wrote:
>> difficulty testing wierd components. I suppose we could try to get each
>> developer to write a stress-tester for his/her subsystem.
>> Has anyone (LMP perhaps?) started this already? Will this work?
>> Thanks for the info.
>At the moment the main tests I run are
>o Crashme
>o ipsend
>o a small set of bugtraq exploit tools for kernel bugs
>People writing tests for subsystems is a good thing

Of course, the fact that several of us on the LMP group now have it
installed in production news server environments works the inode code fairly

I also run it on my personal workstation which does all of the following:

o Mars-NWE Netware services over IPX
o Samba services via TCP/IP
o Stronghold 2.1.1 SSL server ( is one
page on this machine)
o NFS server for installs of RedHat 4.2 here locally
o initrd based boot that sets up the 4 disk MD raid0 array as the
root ext2 filesystem
o Wordperfect v7.0 for Linux
o Wordperfect v6.0 for SCO Unix
o Netscape Communicator 4.04 for Linux
o CVS services here locally
o Bonnie tests on the drive subsystem (which also works buffer cache)
o Smaller local news server for a few global feeds and local tech
support news groups (uses inn-1.6b3 plus some patches)
o Accelerated X + Motif

These don't qualify as stress tests usually, but they do make a pretty
decent binary compatibility test. Plus, when you put them all together,
there are periods of time on the system when the stress load is pretty high,
such as during the nightly expire process and what not.

For stress testing, I'll usually do something like start a kernel compile
while simultaneously moving a set of dummy symbolic tags in my local kernel
CVS tree, plus playing a CD, plus checking out web pages using Netscape,
plus maybe a bonnie test. That gives a general load test, then there are
other tests I run to test specific subsystems, such as multiple bonnies
simultaneously so that it hits the drives really hard, and with nothing else
going on, dedicates the entire CPU usage to slamming those drives. Or, I
hit the system with a thousand teardrop packets in about 2 seconds to see
how it would handle that. Those kinds of things. What I stress usually
varies from kernel version to kernel version based upon what has been

E-Mail: Doug Ledford <>
Date: 17-Nov-97
Time: 05:09:52