From: gossamer@glasswings.com.au (Bek Oberin)
Newsgroups: alt.sysadmin.recovery
Subject: Re: SysAdmin's Dream OS (Was: It's just NOT FAIR...)

In article (4gdsuq$18f@osceola.cs.ucf.edu), Ethan Frolich wrote:
:petro (petro@suba.com) wrote:
:: Ok, to begin the cannonical list of all OS's that suck:
:: VMS sucks.
:: Solaris sucks.
:: SunOS sucks.
:: MS-DOS doesn't qualify, but sucks.
[...]
:: Windows 95 doesn't suck, it blows.
:: Plan 9 sucks on lots of machines at once.
:: Oberon is considered harmful, and sucks.
:Okay. So what would you people consider to be a non-sucking OS?

I'd like to submit Penix as the cannonical example of a non-sucking OS, I happen to have a stack of man pages from it sitting just here[1], but here's the development document for inspection:

=== Okay, here are some general notes on Penix, compiled from a stack of notes left over by the Penix Development team just before they all went to lunch about two months ago and never returned. if anyone out there has seen them, or may know of their whereabouts, or knows what happened to the payroll, or even if you'd like to join the Penix Development team, please mail me at arifel@melbourne.dialix.oz.au.

Please note: there is no social stigma attached to joining the Penix Development Team, and i don't want to hear any more lame jokes about guys who need their Penixes developed. right? right.

selected MAN entries:

cfts completely fuck the system

rcatpar randomly change all the passwords at random

wi whatyouseeisprettypathetic interactiveEditor

edmond page-oriented editor. a step down from a line-editor,
this one treats text as page-sized chunks.

rofl roll on floor laughing. not so much a system function as
what the support staff do when you ask them if they
still have last week's backups on tape somewhere.

ispi shows some of the processes running, some of the users on
the other lines, and sometimes even gives statistics on
CPU usage. it's very context-sensitive.

sssl suddenly swap serial lines with another user

banan display a banana on the screen. taken from
uselss/lib/ansi_art.

crp crap filter - removes tildes from your SerIOus stream

bollx `bollocks!' - ignore ctrl-x, -y and -z `break' flags

stuffio use LZH compression for inter-task message packets (to
save space)

drump dump the rump file

clrump clear the rump file

flrump flub the rump file

doobrie Do Operational bisynchronous reads (input) extended
(we're not sure about this one ourselves)

ftang find textfile and negative-gouge it

gouge remove all the carriage returns from a file ported from
MS-DOS

pong let the other machines on the net know that you're there

brap break all processes - send a ctrl-z to all active
daemons, nyemphs and maenads. this isn't so much av
`terminate the process' function, but rather a method of
signalling to processes that aren't actually attached to
serIOus lines, just to let them know that we are keeping
an eye on them.

exorcise terminate a process completely and utterly. kill it off,
every last trace. we're not kidding here. finito.

General Penix terms:

walnut the heart of the Penix OS. at compilation, it is
approximately 71 megabytes in size (depending on whether
you have the development system or not), but this figure
regularly oscillates between 67 and 83 meg, as the system
`breathes'. It was written in interpreted BASIC, for
ease of modification.

daemon a process that detaches itself from your console session.
unless you redirect the output from this daemon into a
specified file, any output produced will appear out of
various serIOus lines on the system at the daemon's
discretion.

aengel like a daemon, except a bit more polite. an aengel
process will ask before it directs output to your serIOus
line. it will generally do it anyway, but at least it
asks first.

nyemph a daemon process that can migrate from one file system to
another, and will do so if its output has filled the
filesystem that it started on. slippery little buggers.

maenad like a nyemph task, but a little bit more attractive.

sprite a very simple background task that does menial things like
flash the cursor colours, play the sampled `beep' sound when
you press control-g on the keyboard, and make that annoying
click on the floppy drive as the OS checks to see if you have
put in a disk yet.

goelem a very powerful background task. also very stupid. they
are usually started by the system itself, when the need
arises to do something labour-intensive and boring - so
boring, in fact, than none of the usual background tasks
will touch it. you occasionally find goelem tasks
wandering around after they have finished their task, too
stupid to realise that they HAVE finished; pretty much
all you can do about it is allocate them a very low
time-slice (see the `motu' command) and hope that they
don't get into any trouble, or try to kill them with the
exorcise function.

ogre like goelems, but nastier, harder to kill, a bit more
intelligent and, like nyemphs, able to migrate to
different filesystems as required.

caenobites
this is something we aren't particularly proud of here at
Sunburne, but we think that you should know about, in the
interests of safety. we will admit that the circumstances
that led us to developing Penix are slightly unusual, and
that there is quite a bit about this system that was, er,
inherited. We're not saying that we stole it. but there
are large areas of this system that we know nothing about,
but which, when we remove them, the system stops. we don't
know exactly what caenobite tasks do, but we have found
that you can't kill them; you can't stop them; you can't
even slow them down. whenever we use the motu command to
alter their timeslice, they completely ignore it (this may
be due to the fact that we think motu's orders are passed
to the rest of the system by a caenobite task). they can
be incited to remove other tasks by passing them a
BOX_OPEN message packet, but they are best left alone.
thank god there are usually only about four of them.

networks:
the Penix network philosphy is based on a `Master/Slave'
relationship. Systems can be linked into networks that
include hundred of machines, but generally, inter-machine
communication is dominant/submissive. One machine makes a
begging, whimpering request for a file from its `Dom',
which must then establish which of its `Subs' has the
file. It then issues a spank_IO order for the file; if the
data packet isn't sent within the system default timeout
period, the Dom issues a whip_IO. If THAT doesn't work,
the Dom grabs the nearest (by address) Sub by its bus,
shakes it silly, and applies the logical equivalent of nipple-clips.

serviette:
a server with less than 300 meg of hard disk space.

drooler:
printing is controlled by the drooler daemon, who dribbles
text out of the assigned printer ports when it feels like it.
if the drooling process is going too slow for your liking,
it can be prompted by invoking any of the currently running
caenobites.

LNCs: Logical Nipple Clips. This is actually a variety of Worm
process specifically designed to chase between machines on
the net, prompting Subs to complete the most recently issued
Spank_IO or Whip_IO command.

Booting:
one does not re-BOOT a Penix system; it is `re-shoed'. If you boot a Penix system, it will fall apart. Reshoe-ing involves locating the sock partition, socking it, then looking about for something that resembles a shoe-block (a disk-block with startup information on it). once located, the shoe-block is overlaid onto the sock partition and is `laced'. At the request of our user, we have incorporated dual shoe partitions, to enable disk mirroring. the System must always be re-shoed from the same partition; if you get the left-shoe and the right-shoe partition mixed up, then after re-shoeing, the system will take a few tottering steps and will then fall over. be extremely careful about modifying the lacing procedure, as it is quite easy to lace the two partitions together... as you can imagine, this can be VERY dangerous, and any system with its shoes laced together won't run very well.

scanning the sock partition for a shoe block can be a time- consuming process; if the entire hard disk had to be examined, it could take days. an algorithm has been implemented that scans disk blocks at random, on the assumption that there will be more than one shoe block on the disk (this is a standard security feature - in fact, more than half of the disk's blocks are usually allocated as shoe-blocks, to save time when looking for one to shoe from).

example of console messages on booting:

Penix v37.019a ROMrev 76:e_beta (development)
shoe in progress
locating sock partition
scanning SCSI bus *******
scanning st506 devices **
getting desperate
scanning floppy drives **
getting really desperate
scanning tape drive *
sock located on tape drive

*** CLEAN SOCK LOCATED ***

scanning SCSI bus again ******
loading SCSI device driver

scanning for shoe
block at RANDOM ....................................
....................................
....................................
.................... bingo

left shoe located at cyl:00.head:15.sec:16

loading shoe block
processing ......................................
............. ignoring checksum errror
overlaying sock partition
lacing **** shoe partition LACED, Booting
--------------------------------------------------------------------
walnut located, processing ..... done
system boot, 17:29:04 25 March 1992
system ID: Dom:Rhonda Galoot_Cologne

do you want to inhibit DAEMON processes at this time (default=n)? n
do you want to inhibit AENGEL processes at this time (default=n)? n
do you want to inhibit NYEMPH processes at this time (default=n)? n
do you want to inhibit MAENAD processes at this time (default=n)? n
do you want to inhibit SPRITE processes at this time (default=n)? n
do you want to inhibit GOELEM processes at this time (default=n)? n
do you want to inhibit CAENOBITE processes at this time (default=y)? y
too bad. you can't.

starting Daemons: spon spoo furph wrack wruin hobble
castrate whats_all_this_then drooler

starting Aengels: moroni goliad mikayel cefiar lucifuge_rofocale

starting Nyemphs: dom_net_prep dom_net_enquire TOPY_server
neubauten_server spk_server DVA_server NIN_server
skinny_pup_server

starting Maenads: *** Maenad table is empty

starting Sprites: cursor_flash keyboard_click
annoying_floppy_drive_noise

starting Goelems: lif_dat_wait tote_dat_bail get_a_little_drunk

starting Caenobites: pinhead butterball chatterer face atkins
balberith channard

this is a DOM net_device: initialising SpankIO
WhipIO
locating LNCs... done.
locating sPURs... done.
locating CONT... done.

establishing cat_o_nine_tails netlink... done.
locating subs:

Sub:Harry_as_a_boy Galoot_Cologne
Sub:Noxiuma_tapioca Galoot_Cologne
Sub:Algonquin_Jane_Cowhoon Galoot_Cologne
Sub:Obdullah_X Galoot_Cologne

*** SpankIO Warning:
no response from Sub:Harry_as_a_boy Galoot_Cologne

*** Console Message from Caenobite Process ::pinhead::

:: the system... you booted it... we initialised, now you must
:: work with us, taste OUR pleasure!
::
:: we have such sights to show you!

initialising serIOus lines 000 001 002 003

...and so on.

Penix directories:

/typical - defaults for various things like
shell initialisation, flags to say
if the system is a DOM or a SUB, and
various other oddities like the
penix_kickoff script file.
/keymaps
/keymaps
/keymaps
/keymaps - system default keymap for wi

/lib - library files here, strangely enough

/executeables - executeables

/uselss
/der - text files/notices here
/executeables - additional executeables live here
/mungle - not sure what this directory is for, but the
system won't run without it.

/whoops_a_daisy - unallocated blocks here
/ho_hum - logfiles here
/trash - schedule tables for backgrounders
/ephemeral - temporary files
/drooler
/dribble - temp dribble files from drooler process
live here

/help_files - the man entries would be here, but our user
thinks that `man' is sexist, so we called it
help_files.

Sunburne Computer, Inc. 25th March 1992