-- Release 1.3.1pre1 (5 Aug 2005) sflory
	- Added some initial MCE support
	- Turn on kernel over commit
	- Added Charles Cazabon's memtester as default memory test.
	- Memtsts memory detection was broken switched to useing free.
	(Maybe the output of free will not change as much as /proc/meminfo)
	- Changed all the tests using badblocks to not need an end of 
	  device argument.  Badblocks does that by default.(sflory)
	- Fix 3ware test's new-burn option. (sflory)
	- Change messages and dmesg to not flag "error_exit".(sflory)   
	- Fix wait command so it doesn't exit if a timer goes off (jcollins)
        - Fix log location so it works on non-CWD TCF files (jcollins)
	- Reboot test added 
	- Support new cli syntax in 3ware test
	- Various PPC64 build changes
	- Split up destructive badblocks passes
	- Created new 3ware smart test smart-tw
        - Allow varibles to be used in on error
	- Add 3ware timeouts to messages
	- Add diffkernel to readme
	- Remove kdiff tests meanless 2st arg 
	- -D is used twice switch kdiff test to -w
	- Fixed Fast kernel compile test
	- Correct syntax error in dmesg test (extra space after
	  backslash), resulting in Test Results Inconclusive
	  for this test. (jcollins)
-- Release 1.3.0 (24 Nov 2003) sflory
	-Created a blocktest that uses badblock -n option
	-Created 3ware smart test.
	-Use smartmonutils syntax for smart tests.
	-Fix report script to use the log directory.
	-Added new test that does a diff of the kernel tree.  It's
	faster at finding filesystem corruption than the kcompile
	tests.
	- Compile cpuburn with -m32 this fixes amd64, and doesn't 
	seem to break x86 builds.  Cpuburn isn't 64 bit, but it's 
	better nothing.
-- Release 1.3.0pre10 (25 Apr 2003) sflory
	- Don't attempt to compile cpuburn on amd64
-- Release 1.3.0pre9 (21 Apr 2003) sflory
	- Converted blockrdtst-info to perl and fixed ydl issues.
	- Applied Daniel Roesen's regexp patterns fix to fpdisk.pm 
	on slackware
	- Changed deprecated -mpentiumpro to -mcpu=pentiumpro
	- Added x86_64 support
	- add hail mary kernel source check
	- fixed requirements check for kernel sources
	- added 3ware test requires tw_cli to be installed
	- Filehandle adjustment claims 4K when we do 512K.
	- minor update to compile on ia64
-- Release 1.3.0pre8 (4 Dec 2002) sflory
	- fix the /usr/src/linux-2.4 patch
	- Nice destructiveblocks test to 3.
	- Removed borrow percentage from memtst as it has not worked 
	for a long time.
	- Move log files to log directory and don't hide them
	- Nice cpu tests as their intent is really to suck down unused 
	cpu cycles.
        - Revert to old rw data test.
	- End jtc releases. Begin sflory releases
	- Updated FAQ entry for Linux Test Project.
	- Added the /usr/src/linux-2.4 patch.  Whew, we're back.
	- The general matching doesn't work for Linux swap.  So I reverted
	  it.  The tighter matching should work now on all locales since
	  I'm overriding it when parsing.
	- Keep users from shooting themselves in the foot by putting "."
	  or "/etc" in the root path.  Yes, someone did this.
	- More general matching for Linux partitions in the ident field
	  (get_linux_partitions())
	- Took some suggestions by Thomas Kniep.  fpdisk.pm now sets
	  locale to C before launching fdisk.
	- Added support for /dev/ida/* style RAID devices.  I'm beginning
	  to like BSD's /dev/disk...
	- Applied Linux-SPARC patch by David Miller (davem@redhat.com)
-- Release 1.3.0pre4 (6 Nov 2001) jtc
	- -p test now by default uses a 2 TB maximum file size limit.
	  If you need the old behaviour, pass the -2 option to newburn.
	- Increased the default patience of the test controller waiting
	  for possibly defunct test processes to terminate.
	- Rescan scsi bus as part of newburn, to insure /proc/scsi/scsi
	  is up-to-date on hot-swap systems.
	- Kernel compile does make vmlinux instead of bzImage because
	  bzImage only exists on x86.
	- Hide another USB warning that isn't really bad.
	- Excellent, ppc seems to be working.  Shut off the (useless)
	  attempt to ignore SIGCHLD in run.  Why'd I do that?
	- Changed setterm calls to redirect setterm to the console.
	  This way we will shut off the console screensaver regardless
	  of whether we're actually at the console. (newburn)
	- Removed all literal calls to fdisk from all scripts, using
	  the fpdisk library instead.  This is the first version I hope
	  will work on both Linux-ppc and Linux-x86.
	- For Linux on PPC and Darwin:  a new library, fpdisk.pm, attempts
	  to abstract away a lot of uglyness in the ways different unixen
	  handle the /dev structure.  I'll be testing this out and then
	  converting the rest of CTCS to use it instead of calling
	  fdisk directly, manipulating /dev filenames, etc.
	- Make newburn-generator avoid USB/SCSI floppies by default, or at 
	  least the one attached to my workstation.
	- Second attempt at devfs support.  We'll now aggressively
	  translate symlinks in device names/paths so that they do not
	  confuse the test generator.
	- Hide the warning about USB Mass Storage devices not having
	  guaranteed data integrity.  That way hotswapping USB devices that
	  aren't being tested won't flag errors in the log parsers.
	- The LVM patch will automatically ignore physical devices that
	  have LVM volumes for the badblocks test.  The devices will be 
	  tested instead through the LVM logical partitions.  This 
	  behaviour should be configurable, but is not (yet).
	- Applied Derrick's LVM support patch, merging in the conflict
	  with the new devfs support.
	- Removed as86 from the system requirements.  It isn't really
	  needed.
	- Removed a lot of (VA-)CTCS from the documentation.  I haven't
	  decided to remove it from everywhere yet or what my policy
	  is on that, since I am still grateful to VA but don't want
	  to use their marks inappropriately...
	- Updated the documentation for system requirements and to make
	  it a little clearer (I hope) when Cerberus will destroy your
	  machine.
	- First attempt at devfs support.  Sometimes I wonder what I was
	  smoking when I wrote the detection routines.
	- Added a new string to the log parser to catch "stuck on IPI wait"
	  and similar messages as failures.
-- Release 1.3.0pre3 (8 Aug 2001) jtc
	- What?  A release in less than a month?  Holy cow!  ;)
	- Updated the FAQ.
	- Created a simple method for autogening the CTCS spec file.
	  This program has "use autoconf Jason, for the love of God" 
	  written all over it.  Use "make spec" to create a spec file
	  from ctcs.spec.in that includes dynamically calculated package
	  dependencies.
	- Fixed a silly bug in the memory tester that made -2 always
	  fail.
	- Fixed up some more minor syntax errors introduced by
	  previous changes.
	- Removed the whole "kconfig" thing and replaced it with
	  yes.
	- Exported the shell code for the process tree killer
	  to a shell function library in lib/sh/ctcslib.sh.
	  This way we don't have twenty slightly different versions 
	  of the same thing lying around.
	- Changed default cpuburn for non-Intel CPUs to burnK7 from
	  burnK6.
	- Upgraded RR's cpuburn packages to version 1.4.
	- Change all #!/bin/sh to #!/bin/bash.  I used bash specific
	  code in too many places.
	- Forgot to add the selftest directory.  Currently it doesn't
	  contain much, maybe it will later.
	- Rewrote the check-requirements script in perl.  It will also
	  generate a list of RPMS that the package depends on when
	  you run it with the --withrpms flag.  This should be useful
	  for inserting into a spec file.
	- Fixed the syntax checker so it actually works right now.
	  The new cross-platform support for the memtst was confusing it.
	  I am in need of an autoconf-over.
-- Release 1.3.0pre2 (11 July 2001) jtc  "Slurpee"
	- It's been far too long.  My poor code has been languishing.
	  But, at last I have spent the time.  Cleaned up everything
	  for a release.
	- Added removal of backup files to dist-clean target.
	- Added some optimizations to the memory test, changed iterating 
	  variables in test functions to unsigned int and integrated bitpatts 
	  function so that the pattern is stored as an array.  Minor bugfixes
	  also. 
	- Updated test generator to reflect memtst generator changes.
	- Rewrote memtst module.  This _WILL_ break test generators,
	  command line options have changed.  Also, the new memtst
 	  correctly implements the byte-alignment restrictions,
	  actually tests everything you tell it to, and more.  This
	  is a MAJOR change and will probably screw up the test balancing.
	  Time for tuning.  See runin/README.tests for new test options.
	- Doc update for -V option.  Release.
	- Some comments changed/added in run to make understanding
	  it a little easier.
	- Made the "report" tool look a little nicer when there isn't
	  a report to be made.
	- New -V option restores the old VM thrashing behavior of the
	  old burn-in.
-- Release 1.3.0pre1 (09 Feb 2001) jtc
	- Changed the name of the KERNEL compile test to
	  KCOMPILE/FKCOMPILE (depending on fast vs. std).
	  That way our Q/A people won't confuse our kernel
	  developers in meetings...
	- Added a simple report generator that can be used
	  for e-mail.
	- Changed the memory test so that it frees the
	  block on every alignment pass.  Probably not
	  necessary, but will exercise malloc more.
	- OK, rewrote the memory reservation computation.
	  We now base the memory reservation on the system's reserve
	  percentages for buffer memory and page cache.
	- Adjusted the new memory reservation computation again.
	  Massive numbers of disks were resulting in memory
	  exhaustion.
	- Updated the FAQ for question 1.10 (which options
	  does VA use?)
	- Remove the hack-arounds to check syntax on
	  the nbench code.
	- New -I option in VA-CTCS to ignore < 10 backplane
	  resets in the DAC960 test.  This will be used in our
	  manufacturing facility to compensate for the 4450's
	  broken backplane that occasionally spews resets for
	  no good reason.  Normal behavior will still report
	  the errors without this option.
	- Fixed a bug in the new va_detect code.  We weren't
	  stripping spaces off the number of CPUs present.
	- Why didn't someone smack me?  I had two copies of the
	  same libraries, one used, one unused.  The used copy
	  is now all that is left.
	- Altered the memory reservation computation so it is
	  more aggressive on large memory machines.  10% of
	  16 GB (for example) was just a little too much.
	- Updated the FAQ with a new question I get asked a lot.
	- Changed command line parameters on newburn to reflect
	  the missing packages, and the fact they must be specified.
	- Changed makefiles and requirements around to accomodate the
	  missing packages.
	- Removing a lot of extra packages.  There's information
	  about it in the FAQ.
	- Fixed a bug in the DAC960 test that was making the test
	  constantly output errors even when I thought I had fixed
	  it.  So, the behavior should really have changed this
	  time.
	- Documentation updates for the DAC960 test.
	- flushb.c now compiles under 2.4. (string.h problem)
-- Release 1.2.15 (1 Dec 2000) jtc
	- Forgot to add some SGI ltp files as my tree checker script
	  was busted.
	- SGI ltp updated to 101200, some permission problems fixed.
	- Fixed up some problems with my requirements checker script,
	  including duplicates.
	- modem test no longer depends on chat being in /usr/sbin.
	- new DAC960 test only reports error when error
	  is present AND state has changed since the last
	  report.  See runin/README.tests, the dac960 test
	  has a new parameter.
	- Started work on automonkey-related projects for our factory,
	  including autoexit on TEST_COMPLETE + time specification.
	- Reduced overhead in the main CTCS loop by
	  only checking state every 20 seconds.
	- ataprint.c seems to cause gcc segfaults on RH7.0 when
	  running the syntax checker.  Excluding it for now.
	- Documentation update for the new automatic exit features.
	- New -D option and "on error" tcf command allow actions on
	  error only instead of any event.
	- Detect most error messages from the 3ware controller in dmesg,
	  messages test.
	- New stagger field in the autoexit-timeframe allows us to stagger
	  exits over a period of time.
-- Release 1.2.14 (11 Oct 2000) jtc
	- fastkernel and kernel tests had problems:
	- Fixed it so make mrproper happens BEFORE we try
	  to generate a VA .config (oops)
	- Fixed it so the generated VA .config actually gets
	  used.  (oops again)
	- Changed the timestamp test to sleep only 5 minutes,
	  that way we can get more granularity on HEARTBEAT
	  test failures.
	- Fixed the data test so that error reports make sense
	  (values in KB are in KB instead of bytes)
-- Release 1.2.13 (05 Oct 2000) jtc
	- maxalloc takes configurable ceiling, uses an efficient
	  binary search algorithm, and uses malloc since, I hope,
	  that will be safe.  Only impact should be that the test
	  starts up much much faster now on large memory machines.
-- Release 1.2.12 (25 Sep 2000) jtc
	- Updated documentation with a pointer to the ECC driver,
	  and information about the DAC960 test.
	- Added a question about automatic notification on upgrade
	  to the FAQ.
	- Added a DAC960 test to watch for errors tracked
	  by the driver but not noted in the system log (e.g.
	  continuous drive resets)
	- Added "lost interrupt" as an identified error in the
	  system log watcher.
	- traverseread now does a ls -la as part of the test
	  to bang on getattr a little bit.
	- traverseread can now read a specified amount from each file.
	- Changed the straverseread sleep delay to a random between
	  0 and 300 seconds to increase staggering.
	- data test now is more verbose about failures.  Will note
	  short/long(?) writes to assist in debugging Ari's problem.
	- New "traverseread" module will help us in beating up NFS
	  read-only exports, as well as other stuff.  See
	  README.TESTS, basically it just reads every file on
	  a given device and bombs out on an I/O error.
-- Release 1.2.11 (04 Sep 2000) jtc
	- as86 was missing from check-requirements.
	- Added the ability to display a "complete" or other
	  message at a specific time.  Use -z [timespec] to display
	  a "test complete" message, see the documentation.
	- Added the ability to tie a command (such as a CD-tray eject)
	  to a test event (FAILURE or the message mentioned above).
	  This way, when you have 300 machines on racks running Cerberus,
	  the CD trays pop out when they fail.  -d enables this option,
	  see the docs.
	- My new best friend, Kelley Spoon from rackspace.com, submitted
	  a patch that cleaned up my makefiles and added a spec file
	  so you can build an RPM with VA-CTCS.
	- %doc now works in the spec file (at least on SuSE 6.4, will
	  try at work tomorrow)
	- %files now uses a wildcard instead of the file-list generation
	  he had (more efficient)
-- Release 1.2.10 (19 Aug 2000) jtc
	- messages test ignores RPC, Exception 19 error handler errors.
	- dmesg test ignores RPC errors
	- detect ECC single bit error, multi-bit error
	  If you have the ECC module installed, the dmesg/messages tests
	  will pick up errors from it.
	- Chip has come to the rescue of the messages test.  The
	  messages test is now entirely in Perl.
-- Release 1.2.9 (10 Aug 2000) jtc
	- SGI has been assimilated.  Minimal impact to anything else.
	  (Linux Test Project from SGI, -g on the newburn commandline)
	- Fixed a bug in the makefiles -- crashme wasn't being compiled.
-- Release 1.2.8 (09 Aug 2000) jtc
	- Documentation updates.  This marks a change to a stable
	  release for the 1.2.x series.
-- Release 1.2.7 (07 Aug 2000) jtc
	- Jason's python infested mind forgot to initialize nint
	  in maxalloc.
	- Disabled the bigmem memory adding to the main memory.
	  That's my fault, I didn't understand that MemTotal
	  really means MemTotal.
-- Release 1.2.6 (06 Aug 2000) jtc
	- the run command now hides any error messages from chattr,
	  so that reiserfs users won't be upset by it.
	- the va_detect library (and newburn) now can find the total
	  swapspace.
	- the va_detect library (and newburn) now counts "bigmem" memory
	  as part of the total memory.
	- the color library can now return a string with "colorstr".
	- data test no longer requires usleep.  usleep also removed
	  from the requirements list in check-requirements.
	- Minor Makefile changes to accomodate maxalloc.
	- Added a God mode (-G) that will bypass what newburn-generator
	  normally considers to be fatal errors.  Don't use this except
	  for testing the generator.
	- Tuned the new algorithm to work in standard and simulated large
	  memory environments.  Should even work with pathological
	  mode.
	- Runaway swap/VM issue worked around.  For some reason,
	  occasionally the VM would attempt to swap out nearly the entire
	  memory tester process, causing a crash if you didn't have as
	  much swap as you had memory.  Even if it didn't crash, it
	  weakened the strength of the memory tester, because it caused
	  the system to swap so much instead of beating on the
	  memory.  newburn-generator now has a complex algorithm to deal
	  with this phenomenon by reserving physical RAM and attempting to
	  estimate the maximum swap necessary for a batch of tests.  New
	  messages are also displayed during newburn-generator that
	  indicate the results from the algorithm.  This should help in
	  testing it.
	- Created "maxalloc" in the runin/src/memtst.src directory.  This
	  program uses brute-force to attempt to discern what the maximum
	  possible size for a memory allocation is and prints it.  Used
	  in newburn-generator.
	- Fixed a silly syntax error in the messages test.
-- Release 1.2.5 (02 Aug 2000) jtc
	- usleep is a requirement for the data test.  Put it in the optional
	  requirements.
	- Really fixed the warning messages in -p.  Those messages were cosmetic,
	  but annoying.
-- Release 1.2.4 (01 Aug 2000) jtc
	- Memory test : The error consistency checker, while
	  good in theory, doesn't work when compiled.  Need to write an
	  assembly version.  Until then, it is disabled.
	- Memory test: Now does 15 passes, 5 passes focusing on the
	  cache RAM, 10 on the main.  Suggests which might have failed.
	  All of this helps reduce the amount of data printed by the
 	  memory test.
-- Release 1.2.3 (28 Jul 2000) jtc
	- A bug accidentally ALWAYS enabled S.M.A.R.T checking.  Fixed.
	- Support for a crash heartbeat started.
	- Change memtest ceiling to 1024 MB.  Some Linux boxes, depending
	  on configuration, can't allocate much more than that per process.
-- Release 1.2.2 (24 Jul 2000) jtc
	- Updated the FAQ.
	- Added S.M.A.R.T. checking from the ucsc-smartsuite to predict
	  drive failures before they happen. -a option in newburn enables
	  this checking for all hard drives.
	  S.M.A.R.T. checking will automatically disable itself if you
	  have no supporting devices (i.e. RAID).  You should also consider
	  adding the -D__NEW_HD_DRIVE_ID option to compile it -- I haven't had
	  much luck without this option, but it breaks compiling on current
	  VA loads.
	- The memory tester parses /proc/kcore to attempt to locate
	  bad pages when found.  Due to the way memtst works, 
          this actually has a fair shot at succeeding.
	- Fixed the lack of a dist-clean target in nbench.
	- Fixed lockfile messages in pathological mode.  Remember that
	  -p now _requires_ a command line option.
	- Added "make requirements" to run check-requirements for you.
	- Fixed up memtst so it has less tolerance of insufficient memory
	  and reports what it is doing more.
	- Errors in the test software are counted as test failures.
	- Errors during configuration of a test are counted as test failures.
	  Both of these are arguably wrong (since the _test_ itself was never
	  run), but I think this is the most reasonable thing to do in
	  this case.
	- Fixed an apparently long standing bug -- nonzero returns on test
	  configuration were supposed to be counted as test config failures,
	  not just values of "2".
	- COPYING file updated with information on the BYTE benchmark
	  suite.
	- Beefed up regular expressions in the kernel watcher to find
	  VM problems.
	- Hide warnings in crashme -- I don't want to deal with that right
	  now.
	- New modification to syntax checker should handle perl better (from
	  VA internal Cerberus distribution).
	- check-syntax now does line totals of the C, sh, and perl code,
	  excluding those parts of VA-CTCS not authored by VA.
	- Fixed check-syntax for Debian.  We weren't handling perl detection
	  right for them.
-- Release 1.2.1 (20 Jun 2000) jtc
	- File reorganization, put perl libraries in lib/perl.
	- Memory tester modifications to check for error consistency and
	  to report context around failures.  This greatly increases the
	  complexity of the error report.
	- Slight changes to the requirements checker so it reports problems
	  a bit more nicely.
	- CFLAGS for the memory tester force off unroll-loops optimizations.
-- Release 1.2.0 (09 Jun 2000) jtc
	- Syntax checker added for newburn.  This will hopefully
	  help check for corruption of Cerberus code before execution.
	- -K option now disables the slow kernel test automatically.
	- New -B option sets path for kernel test.  This will be used
	  in virgil phase II.
	- Added code/makefile syntax checker.  "make test" checks
	  everything for syntax errors.
	- Fixed a couple of syntax errors with my checker :)
	- Miscellaneous code rearranging (non-functional).
	- More documentation updates.
	- Added an optional section on the system requirements so things
	  like missing sensors will not cause you to barf.
	- Added an lm_sensors monitor module.
	- Added a command line option to newburn-generator (-l) to use
	  the sensors test.
	- Added a system check routine "check-requirements" that will
	  detect if you have installed the software right.
	- Added --force option to newburn (must be first option) to bypass
	  the system check.
	- Added -N option to burn specific devices/NFS shares.
	- Added mount command locking on the data test so that
	  it doesn't try to mount the same NFS share in different places.
	- Clean up core files on make clean/dist-clean.
	- Added a "allmessages" test that duplicates syslog.
	- Adjusted the messages and allmessages test to look for
	  /var/log/syslog so Debian users won't be without the system
	  log monitor.
	- Crashme, the BYTE benchmark, and the allmessages test now
	  configurable from the newburn command line.
	- Reorganized the newburn help. Should be easier to understand.
	- Made help work from the newburn
	  command, not just newburn-generator (hack)
	- Kernel compile test now is configurable on the command line.
	- -t option changed so it no longer affects the kernel test.
	  This means that to get the old -t behaviour, you have to use
	  -t and -k (1*numberofCPUs)
	- -p option changed, now limits itself to configuring the disk
	  R/W test.  -p requires a parameter now.
	- If there is a VERSION file, use that for the version reporting.
	- Updated makefiles to generate symlinks automatically.
	- Added "nbench" as a test module for Cerberus.  This is the "new"
	  BYTE UNIX benchmark suite.
	- FAQ and other docs updated again.  This time with more useful
	  information on interpreting failures, accuracy, etc.
	- burnreset utility added, paralleling the same utility in
	  the main Cerberus tree.  Makes it easier to use burn
	  outside of manufacturing.
	- Added a help screen and CVS checkout information to
	  newburn-generator.
	- Added the "crashme" test to ctcs.
-- Release 1.1.7 (20 April 2000) jtc
	- FAQ updated.
	- va_detect library now detects SONY and HP CD-RWs, and probably
	  a few other types too.
-- Release 1.1.6 (31 March 2000) jtc
	- va_detect library no longer loads ide-scsi.  Otherwise
	  the same device shows up as two different ones, and
	  I'm not prepared to deal with that yet.
	- Revised further to remove it if it is present.
	- burnCPU test uses primitive detection to run burnP6 on
	  Intel machines and burnK6 on others.
	- Several small bugs in the burn* test wrappers fixed.
-- Release 1.1.5 (28 March 2000) jtc
	- Number of filehandles in kernel increased to 512000.
	- The new kernel test that used to be in alt is now called
	  "fastkernel" and is now in the main directory.
	- Changed '-p' disk test behavior.  Now we use the "fastkernel"
	  test when I'm doing real disk tests with -p.
	- Integrated Robert Redelmeier's cpuburn-1.2.
	- Added '-m' switch to newburn to disable Larry's memory test.
	- Added '-r', '-R' switches to enable RR's cache thrasher
	  (burnBX).
	- DANGER WILL ROBINSON!! BUFFER FLUSHING via BLKFLSBUF is now on
	  by default.  Older (<2.2.13) kernels will definitely have
	  problems.  Comment out the execution of flushb.real in
	  runin/bin/flushb if you suspect difficulty due to this.
	- va_detect.pm now loads appropriate drivers to enable detection.
-- Release 1.1.4 (24 March 2000) jtc
	- kernel test changes in 1.1.3 reverted.  I put the new test in
	  "alt".
	- Documentation updates.
-- Release 1.1.3 (15 March 2000) jtc
	- kernel test improved.  Since we're no longer as paranoid as we
	  used to be, I optimized the kernel test and now we are once
	  again doing only one copy for multiple test runs.
	- runtest now has the capability to run cleanup code.
++ Release 1.1.2+1 (20 April 2000) jtc
	- CD-RW changes from 1.1.7
-- Release 1.1.2 (07 March 2000) jtc
	- Moved kernel compiles to /home/tmplinux (simple change to
	  generator)
-- Release 1.1.1 (06 March 2000) jtc
	- Makefile change to make compilation instructions more
	  consistent.
	- CD-RW avoidance support added.
-- Release 1.1.0 (28 February 2000) jtc
	- No changes to CTCS in this release, just update version number.
-- Release 1.1.0p4 (17 February 2000) jtc
	- Support for testing/avoiding IDE-floppy type devices.
	- va_detect library enhanced for the IDE floppy support.
	- va_detect cleanup.
-- Release 1.1.0p3 (10 February 2000) jtc
	- Compilation warnings in memtst cleared up.
	- Reversed the R/W test enable to shut it back off.  We're doing
	  pretty well with the existing tests.
-- Release 1.1.0p2 (2 February 2000) jtc
	- Larry's memtst adjusted so I can put a per-process ceiling
	  on the memtst program.
	- Per-process ceiling implemented in the test generator for
	  correct handling of very large memory machines.
	- Some more tweaking to the test generator/memtst to make sure
	  most all memory really does get tested.
	- Added the fcompare module to compare files on the media with
	  files somewhere in the filesystem.
-- Release 1.1.0p1 (27 January 2000) jtc
	- Compiler warning fix to chartst.
	- make clean now deletes flushb.real.
	- kconfig binary moved to runin/bin where it belongs.
	- test generator creates a more aggressive test for very large
	  memory machines.
	- added a thermal test mode to the test generator (-t) that uses
	  RR's cpu fryer.
	- Pathological mode made a little less pathological.  And made
	  the default.
	- data test nices itself slightly.
	- Fixed the fix for the race condition.  There was a thinko in the
	  code that disabled my fix.  :(
	- Enhancement to dmesg test to make it occupy less CPU time and store
	  a much smaller volume of log data.
	- Similar enhancements to kernel test greatly reduce debugging output.
	- Memory test no longer as verbose to reduce debugging output.
	- burn no longer uses mktemp to create new directories, now we use
	  a numbered system.
-- Release 1.0.3 (11 January 2000) jtc
	- Fix for race condition between test subprocess
	  termination and test process SIGINT delivery.  I have tested
	  enough to know it at least is no worse than current (<=1.0.2)
	  behaviour.
-- Release 1.0.2 (10 December 1999) jtc
	- Small bug in data test fixed -- now data-info doesn't get
	  confused when the same NFS directory is mounted in more than 
	  one place.  Remember that Linux df sometimes uses multiple
	  lines for a single device unless you specify -P.
	- Fixed race condition possibly causing multiple data tests to
	  fail when they are all trying to mount the same filesystem.
	- More workarounds for kernel compilation screwyness.
	- Niced up the badblocks tests so they don't take as much CPU
	  time from our CPU/RAM tests.
-- Release 1.0.1 (6 December 1999) jtc
	- run now distinguishes between tests interrupted with and without
	  errors on the final line in the report.
	- Manufacturing burn-in no longer has a default time limit.  One
	  can still be specified on the command line.
-- Release 1.0.0 (29 November 1999) jtc
	- Error message parser now sees the NMI "dazed and confused" message
	  as an error.  They could have at least put the word warning in it..
	- Error message parser will now see DAC960 drive critical error
	  messages as error messages.
-- Release 1.0.0p12 (19 November 1999) jtc
	- Kernel test now uses make -sw to reduce the size of test
	  logs info, but still provide some debugging output.
	- Kernel test now recognizes VA makefiles and adjusts accordingly.
	  Still works with Linus kernels, too.
	- Increase default sleepdelay for dmesg to 15 seconds from 10.
	  (reduces log file size)
-- Release 1.0.0p11 (17 November 1999) jtc
	- Adjustments to documentation per Bruce Benson's comments.
	- Tiny kernel test change.  Now uses mkdir -p to create
	  test directory.
	- Removed the copy sentinel from the kernel test.
	- Modified the kernel test to delete the VA makefile.  This
	  way kernel compiles behave The Right Way instead of the VA
	  way.
	- dmesg test stupid bug fixed that fills up the disk and throws
	  the test out of balance.  If I would only read my own docs...
	- Tiny bug in newburn fixed (was an unset bash variable, $x).
	- flushb now compiles right.  No longer using hdparm to
	  flush buffers in data test (even this doesn't activate
	  unless you set the environment variable, below)
	- run will eventually sigkill subprocesses now if they take too
	  long to exit.
-- Release 1.0.0p10 (12 November 1999) jtc
	- Small adjustment to data test.  It is more aggressive
	  about terminating subprocesses.
	- Monitor test now has a timeout.  The addition of this
	  parameter might break scripts, so read README.tests for
	  the new syntax.
	- Pathological mode no longer trashes ext2 (short term test).
	  Buffer cache flush disabled.
	- Set the environment variable "CERBERUS_FLUSH_BUFFERS" to try
	  with the flush-buffer system.  This WILL trash your system
	  currently, only use this if you're trying to debug why. (not
	  test environment variable type mentioned in README.TCF, but REAL 
	  environment variable.  Perhaps these should be one and the same 
	  some day...)
-- Release 1.0.0p9 (11 November 1999) jtc
	- Fixed small bugs in blocktst-info and friends that resulted
	  in incomprehensible messages when accessing dead devices.
-- Release 1.0.0p8 (11 November 1999) jtc
	- Forgot to update newburn-generator version on last cycle.
	  Fixed.
	- Minor newburn-generator internal architecture rework.  I've
	  exported some of the raw test generation features to a library
	  in preparation for a new batch of configurable test generators
	  (Thermal, Noise, Power, MTBF, Kernel Stability)
	- Serious rework of the data test.  The data test now will
	  go ahead and use large files in automatic mode.
	- The data test now overrides its internal calculations
	  if you specify a size.
	- newburn command only compares options now, not files, to
	  determine if a change has occurred in the control file.
	  This is necessary because the exact details on a data test will
	  change frequently/uselessly between passes.
	- Kernel test now uses kill -9 to kill subprocesses.
	- Kernel test now uses yes to insure that make oldconfig
	  asks no questions.
	- Kernel test can use a different build directory than /tmp/linux.
	- run-in test balance altered slightly to favor beating on
	  memory and CPU harder.
-- Release 1.0.0p7 (2 November 1999) jtc
	- Fixed stupid memory exhaustion in kernel test.
	- Added more GPL/copyright notification headers.
	- Added  Robert Redelmeier's CPU burn modules.  This
	  doesn't affect the manufacturing burn.
	- memtst is now configurable on the command line for
	  reserved memory.  Default changed to 128 MB.
	- FAQ updated.
	- Minor bugfix in makefiles.
	- Copyright banners added to source code.
-- Release 1.0.0p6 (29 October 1999) jtc
	- The "messages" kernel messages test is now case-insensitive
	  to pick up DAC960 error messages.
	- The kernel test has Yet Another Termination Bug fixed (more
	  like attacked with nuclear weapons).
	- Optimization in kernel test to prevent two copies occurring
	  on the first pass.  
	- Boosted the priority of the clock a bit more.  We must have
	  our pretty displays :)
	- kernel-info deleted as it is now obsolete.
	- A couple small things cleaned up in "run".
-- Release 1.0.0p5 (27 October 1999) jtc
	- Code cleanup and optimization of all the Test Control System
	  libraries.
	- va_detect.pm:  Rearranged the library to make the function
	  APIs less complicated and make more sense.  Altered
	  newburn-generator to compensate.
	- color.pm:  Debugging code removed.
	- timecalc.pm:  General cleanup/code deobfuscation.
	- Code cleanup and optimization of the runtest program.
	- runtest now executes only one external program (date), once,
	  per test (besides the test program, of course).
	- Extensive code cleanup and rework of the run program.
	- Behavior was not correct for timeouts occuring outside of the
	  wait command.  Fixed.
	- Several race conditions and extreme-case logic errors
	  eliminated.
	- Made our signal handlers safer, they now do much less
	  (especially for SIGINT).
	- Integrated foregound/background job handling into a single
	  subroutine.
	- Fixed multiple file handling so that error messages are reported
	  correctly.
	- Cleaned up the execution of runtest to prepare for its eventual
	  assimilation into the main program.
	- Eliminated the old reaper.
	- Fixed a minor display bug in the report generator.
	- We now attempt to set EXT2 attributes for synchronous writes 
	  on our logfiles. Failures to do this are not considered
	  critical, and are not reported.
	- Small cleanup on module color.pm.
	- General code cleanup on the newburn program.
	- Fixed a bug in newburn that would result in certain option
	  codes mishandled by the "echo" command.
	- General code cleanup on newburn-generator.
	- Internal modifications to newburn-generator to seperate the
	  detection routines into a perl module.
	- The inevitable miscellaneous documentation updates.
-- Release 1.0.0p4 (18 October 1999) jtc
	- Makefile updates.
	- Use hdparm instead of flushb.c to flush drive buffers.
	- No more kernel stupidity in this release.  While it is wasteful,
	  I now copy the whole kernel source tree every time from scratch
	  before it is compiled.  Hopefully this will correct those
	  pesky 3rd/4th pass errors.  I hate working around this, but
	  this does have the benefit of beating up the disks more.
	- Some pretty major internal reworking of the run command:
	- REAPER function eliminated and appropriate code reworked,
	  as Perl isn't reentrant.
	- There are some other signal handlers that we probably
	  should look at for the above reasons...
	- Chip set me on the right track to eliminating my strange
	  process termination difficulties.  All this should now be
	  fixed, the process termination handler is now in a single
	  location instead of being copied in different ways all over
	  the place, and finally debugging code commented out for release.
	- Reduced the number of process table entries per test by one.
	- Manipulating priorities to make the test control program
	  have higher priority, this should enable faster updates
          of the onscreen clock on large/IDE systems.
-- Release 1.0.0p3 (12 October 1999) jtc
	- Minor revision to kernel test -- apparently some of the
	  newer kernels require more cleanup of dependencies before
	  compiling in the loop. (mkdep.c dependency?)
-- Release 1.0.0p2 (11 October 1999) jtc
	- Sample tests now moved to the 'sample' directory.
	- Added command line options to burn in the CDROM, any JAZ drives,
	  or even (shudder) the floppy drive.  Can be combined with
	  pathological mode to be really, really, REALLY stupid.  See
	  docs for details.
	- Added a command line option to disable hard disk test (maybe
	  you just want to build kernels all day...)
	- Staggering randomization of badblocks/data tests has been
	  increased to a maximum 60 second delay.
	- JAZ/ZIP drive detection and avoidance code added to burnin
	  generator.
	- Makefiles added.
	- Code and documentation cleanup.
-- Release 1.0.0p1 (7 October 1999) jtc
	- Disabled the data test completely for non-pathological
	  mode.
	- Disabled VESA power save and screensaver during burn.
	- Enabled the Linux Magic SysRq key during burn.
	- Many cosmetic alterations to the burn system.  The burnin
	  generator reports more of what it is doing during setup.
	- Reprimand added for CTRL-C-happy operators.
	- The newburn command includes some state file management
	  intelligence now.  It can detect when you have changed options
	  and it asks if you want to eliminate the old test.
-- Release 0.9.7 (6 October 1999) jtc (internal only)
	- The newburn command now increases the available filehandles
	  to 65535, so pathological mode should no longer run
	  out of file handles.
	- Systems without CD-ROM no longer display an (unimportant)
	  error message.
-- Release 0.9.6 (5 October 1999) jtc
	- Minor logic bug in test generator fixed -- non-pathological mode
	  works right now even for big partitions.
	- Verbose is more verbose on the end report -- shows
	  iterations/timer for successful tests too.
-- Release 0.9.5 (30 September 1999) jtc
	- The end of the filehandle shortage.  No more perl programs
	  are launched inside of runtest, and a minimal number of external
	  modules are used generally.  runtest should now have much less
	  overhead.
	- The display should be colored correctly even under heavy load. 
-- Release 0.9.4 (29 September 1999) jtc
	- The default test generator now only launches one data
	  test per partition.  This should no longer exploit
	  the ext2 bug(s).
	- Pathological mode (-p) added to burnin, which engaged the
	  old behaviour.
-- Release 0.9.3 jtc
	- A small internal bug with the data test has been fixed,
	  no more errors should show up if it is interrupted.
	- The CRAP now has a verbosity feature.  Use burn -v to
	  activate, burn -v [time] is valid.  See docs for details.
	- Monitor test fixed, but switching VTs back to the original
	  terminal doesn't work anymore.  That's just for convienence,
	  so this will do until I figure out why open is blocking signals
	  for its subprocesses..
	- Burnin-generator now more aggressively/intelligently attacks
	  very large partitions.  Enough data tests are now launched
	  to insure good coverage.
	- data test now stays <= 6 processes per active test.  One
	  problem with Cerberus as currently implemented is the
	  possibility of process table overflow..
	- data test limits itself to 2 GB per file.  Otherwise on large
	  disks/arrays this gets pretty silly.
	- data test now uses a pseudo-random data generator rather than
	  dding from /proc/kcore.  The data generator does not use dd,
	  either, which improves the performance vastly.  75 MB/sec
	  of data can be generated on an unloaded PII/400, so we're
	  now really bound by hard drive speed, even for fast
	  RAID arrays.
-- Release 0.9.2 (16 September 1999) jtc
	- Kernel test modified now to support a parallelization
	  parameter.  The burn generator will now limit
	  parallelization to 2*CPUs, which while not optimal
	  for performance, will guarantee that everything will
	  be working hard, which is the goal.
	- Memtst insures at least 96 MB virtual memory free now 
          instead of 20 something.  This should
	  reduce out of memory errors on our Mongolian Horde style test.
	  However, previous increases have made little difference, so
	  this may be the wrong thing to be doing.
	- Fixed a stupid success accounting bug on the data test.  Now
	  it counts successes correctly in addition to failures.
-- Release 0.9.1 (9 September 1999) jtc
	- Added a clock that keeps the total test time on screen.
	- Cleaned up debugging code in a lot of places so that extra
	  output only shows up when a problem with the test suite is
	  suspected..
	- Changed handling of foreground processes in 'run' so that
	  it is similar to the background process handling (allows
	  clock to run during foreground processes)
	- Added a burn time set parameter for "22h" or "1h" or "340m" or
	  whatever..
-- Release 0.9.0 (2 September 1999) jtc
	- Changed release number to reflect our new beta status.
	- Minor change to the RAID code in the test generator.
	- Rework of the test generator for a 24 hour burnin, similar to
	  that which we use...
	- Exported some good helper functions out of run.
	- Added s-tests, which delay random amounts before starting.
	  This allows me to stagger disk tests in a better way
	- Added purple failure messages for bad test configuration (i.e.
	  burn script errors!)
	- runtest now has TTF support.
	- Adjusted TTF code so that it is more accurate & consistent in
	  reporting.
	- kernel and syslog watchers now implemented....
-- Release 5 (31 August 1999) jtc
	- Burn generator added, newburn script added.. a flurry of other
	  minor changes and bugfixes.
-- Release 4 (31 August 1999) jtc
	- Added more colors, clearer messages
	- Last release as runtest, without any burn capability
-- Release 3.1 (31 August 1999) jtc
	- Added color to the error reporting.
	- Removed some debugging output.
	- Runtest now outputs human-readable (color) and loggable output
	- Multiple timer blocks are now legal, things like looping inside
	  timer blocks should now be possible..
	- The only output during test runs (after initialization) should
	  be errors, now!
-- Release 3 (14 Jun 1999) jtc
        - Fixed some bugs in the data test that didn't guarantee termination
	  on interrupt.  Added some code for runtest too to support this.
	- Fixed the block test info systems so that they parse fdisk
	  output a bit more smartly.
-- Release 2 (11 Jun 1999) jtc
	- Fixed bugs in the looping construct that were causing the whole
	  file to loop, rather than just the loop area.
	- Fixed a small bug in the 'set' command.
	- Beautified the code for 'run' and 'runtest' so that it is more
	  readable.
-- Release 1 (09 Jun 1999) jtc
	- Initial release
	
