Date: Mon, 28 Dec 1998 15:00:48 -0500
From: Tom Limoncelli 
Cc: lisa99-discuss@usenix.org
Subject: Re: Call For Papers topics

[this is background info to clarify my next message to the mailing list]

Tom Explains Theoretical System Administration
----------------------------------------------
draft dated 12/28

As the person that coined the phrase "theoretical system administration" let me
clarify.  My goal is to encourage papers that are "analytical" and discuss
"paradigms."  While linear improvements are made by creating better tools,
quantum leaps can be made by developing models, figuring out what we do, and,
hopefully, this will lead us to entirely new ways of doing things.

"Theory" isn't the right word, but it sounds better than "paradigmic system
administration" or "analytical system administration".  However, with analysis
and paradigms we build theory.  I'm open to suggestions.

An example:

For example, take Remy's paper on the lifecycle of a host.  The analysis
concludes that machines loop through cycles:

        initial OS install
        LOOP
                OS degrades
                OS is reloaded
        UNTIL sick of the machine
        decommission

With that, we can make conclusions like "we want certain phases to last a long
time (degrades), and others to last a short time (install, reload).  It also  
helps us come to terms with our psycological discomfort about decommisioning
machines (we can accept that "Every OS goes to heaven").  We can make checklists
of things to do before and after you hit each stage (therefore, we can be more  
methodical, less amaturish about what we do). 

Comparative papers can try to figure out why the 'OS degrades' phase of certain
machines/OS is longer than other.  We can also classify types of degradation,  
which will lead us to be able to prevent certain causes.  We can take this info
back to vendors and say things like, "Because you break Smiths 3rd rule of 
degradation prevention, your OS sucks; please fix it" rather than "Your OS
sucks."

The state of the art is to say, "Microsoft SUCKS!  UNIX RULES,
DUDE!!!".  We sound like children.  I think it would be much better if
we could say, "UNIX compares better to NT at all stages of the Remy
model, leading to higher cost and less reliability."

We can also use this as a building block for other papers.  It is very difficult
to calculate total cost of ownership (TCO).  However, with the above model we
can describe one facit of cost and figure out how we can reduce cost.

What makes it truly "theoretical" is that it can be proven wrong.  Someone could
write a paper saying that the model is broken for particular classes of OS.  Or,
one might say that the conclusions are wrong... we want 'degrade phase' to be as
violent as possible because, in theory, users benefit from flexibility and in a 
perfect world the OS would reload itself after a user logs out."

--tal