M.S. Project idea

[George.H.Harry.Rich]

In article <6993 at cps3xx.UUCP> ramchan at cpsin3.cps.msu.edu (J. Sunil Ramchandani) writes:
>Hi,
>
>As part of my degree requirements (M.S. in CS) I am supposed to do a
>project which has to be in my major area of specialisation (Operating
>Sys.). I was wondering if any of you knowledgeable sources out there
>could give me an idea or suggest a topic for my project which should be
>done on UNIX in either C/C++. The aim is to get a very good idea of the
>UNIX internals as part of the project. The project is worth 6 credit
>hours and should be spread over 2 terms (2 quarters). 
...

I'll give you a couple of problems associated with Unix.  How much they
involve you in internals depends on what you do with them.  I think
broad as opposed to single implementation oriented solutions to either
problem would be worthy of a dissertation topic since it would involve
significant contribution to theory.  (Your faculty might not agree with
me on this; both also address real, practical problems).  Meanwhile,
you might search inside them for an appropriate master's project.


(1)	Security
	--------

There are a number of security weaknesses in Unix systems as they are
normally installed.  In my belief, they come up not because they
basic security scheme built into the kernel is weak, but because in
the process of implementing and integrating tools for maintaining and
administering the system, leaks have been added, rather like putting
up a strong burgler resistant front door and leaving the key under the
doormat because the cleaning woman has to get in.

I see several items worthy of study.  (1)  Using the existing base
security scheme, is there a reasonable approach to making the necessary
tools available without the well-known security weaknesses?  What should
be the implementation guidelines for future tools based on what is learned
in this study?   (2) Is it possible to use the existing security scheme as a
sound basis for a broader model such as produced by the IEEE?  (3) If
modification of the kernel is necessary to adequately layer on broader models,
what is the minimum necessary kernel change to do this?


(2)	Performance Analysis
        --------------------

It has proved difficult to forecast how a Unix system is going to behave
with a given application set.  I would like to see a system which accumulates
data on applications as they run, whether together or in isolation, and uses
this data to forecast the effect of proposed load changes on the system,  i.e.
if I add two more users of application A to the peak load: am I going to go
into a devastating page thrash?;  how much am I going to be hurting the
response on application B?

Obviously, there are several sub-projects in this which are themselves
quite challenging. 

Luck,

	Harry Rich