Script for Undervolt Stress Testing

2007-04-12T23:16:21Z

Fb0Rnq:

This script helps in calibrating voltages when [[Pentium M undervolting and underclocking|undervolting a Pentium M processor]].

People have many different tolerances for how far they will undervolt their system. Some are eager to just run their Pentium-Ms at 700mV and abandon safety; they ramp their systems as far as they can without crashing their system, and maybe they pull the voltages up a margin from the failure point. However, this provides only a weak degree of security as a number of failures can occur that might not surface immediately. In the worst case, the system will fail months later, and the blame might be assigned to, say, a kernel upgrade or patch when really the system failed due to intermittent lack of power.

Many would like to guard themselves again such a failure and consequently have opted to run a prime number stress test such as [http://www.mersenne.org/prime.htm| MPrime] in a "torture test" mode, while they ramp down their voltages to find a comfortable margin from the failure point. However, as per recommendations from a [http://mailman.linux-thinkpad.org/pipermail/linux-thinkpad/2006-July/034806.html| thread of the Linux-Thinkpad mailing list], perhaps even more can be done. Following such advice, this script not only runs MPrime, but also toggles on and off a lot of power-demanding features of the laptop throughout the course of the test. The idea is to more rapidly expose corner cases in which the system might act up.

{{NOTE|Please feel ''very free'' to improve/fix this script. My intent for its posting is to make its ownership as public as possible. There's no need to try to E-mail me to validate your changes. If you feel they are in the best interest of the public, just make the changes. The script attempts to employ pre-conditions to intelligently apply functionality only to those laptops that appear to support it. Hopefully, its framework will allow for extension without heavy redesign.}}

{{WARN| There are ''very important'' warnings embedded into the comments of this script. I have left them there because if you copy this script to your system, I would want you to carry these warnings as comments with you. Please read these comments and the script very carefully. Stress testing an undervolted system is not a trivial undertaking and you need to be as accountable as possible for what a script like this does.}}

{{MoveToCode}}
<pre>
#!/bin/bash
#
# DESCRIPTION AND MOTIVATION
# --------------------------
# Designed for an undervolted laptops with frequency stepping, this script
# swings the system between aggressive and low power use, and also swings
# among the available frequencies.
#
# The idea is that such exteme use of the system will likely explore corner
# cases where the system might fail. Hopefully, such testing can curtail the
# time necessary to establish confidence in undervolted systems.
#
# In the background the MPrime program, a prime number search engine, runs in a
# "torture test" mode, in which it tests computations against known results and
# errs out if there's a discrepancy. Unless it errs out, this script runs
# forever.
#
# IMPLEMENTATION
# --------------
# The design of this script attempts to address laptops beyond the Thinkpad T42
# for which it was designed. Many of the function definitions are prepended
# with conditionals that check the system for functionality and either bail out
# or disable features accordingly.
#
# In particular, the nature of what "aggressive" constitutes is defined by a
# number of "toggle_" functions. The pre-pended conditional to these functions
# appends the function name to $AGGRESSIVE_TOGGLES if the system appears to
# support the feature. The toggle_aggression function then calls all the
# functions in $AGGRESSIVE_TOGGLES. Look at these "toggle_" functions for
# examples of how to extend this script for other possible stressing.
#
# EXTERNAL PROGRAMS EMPLOYED
# --------------------------
# Test system integriy (required): MPrime - http://www.mersenne.org/prime.htm
# Download files: curl - http://curl.haxx.se
# Read random sectors from CD: spew (for gorge) - http://spew.berlios.de
# Keep hard disk active: stress - http://weather.ou.edu/~apw/projects/stress/
#
# EXECUTION
# ---------
# Read this script including all the warnings below, and then make sure all the
# variables in the "Script Globals" section are appropriately set.
#
# This script uses the mprime binary with the "-t" switch for the MPrime
# "torture test." This test by default uses all the memory available on the
# system. However, if you run this system for many hours, your kernel may run
# out of memory, and kill mprime and this script. To spare yourself this
# problem, use the "NightMemory=" and "DayMemory=" parameters in MPrime's
# local.ini file, a file typically in the same directory as the mprime
# executable (read the MPrime documentation for specifics). The torture test
# by default uses the greater of these two settings, so just set them both a
# reasonable margin away from the total amount of memory available on your
# system. On a system with 512MB of RAM, I set these parameters both to 448,
# and had enough memory left over to run my normal set of background processes.
#
# The arguments of this script are "aggression" toggles to disable. Any
# function below that begins with "toggle_$OPTION" can be disabled by using
# $OPTION as one of the arguments of this script. Otherwise, all the stressing
# that a system supports are enabled by default.
#
# Because of Warning 3 below, I recommend you run this script as
#
# stress_test 2>

ThinkWiki - User contributions [en]

Script for Undervolt Stress Testing