On the Duality of Operating System Structures

This 1978 paper from Xerox PARC argues that Operating Systems design has taken
two paths up to this point, that of the message-oriented system and that of the
procedure-oriented system.  The paper claims that these two different paradigms
are duals, and that they are logically identical.  The paper goes on to
empirically examine existing message-oriented and procedure-oriented systems and
it then derives a canonical, idealized model for each type of system.

For message-passing systems, it notices that the process structure and hierarchy
of the system is fairly fixed.  You have a number of long-term processes
communicating through existing channels, passing messages.  You have a
synchronized mailboxes that are processes in sequential order.

Meanwhile, in a procedure-based system, you have lots of ephemeral processes
that are quick and dirty to create.  Processes have access to everything, and
there is an emphasis on locking to protect shared data.  Where in
message-passing relative state seemed more important, in process-based systems,
globally-scoped data structures are the norm.

From these two sets of data, the authors derived idealized models of both
systems.  In message-passing, you have a uniquely-identifiable message with a
header and some variable data portion.  These messages are transmitted via some
communication medium and may be subject to queueing.  Processes are
constructed around inter-process communication.

In procedure-based systems, everything is much more modularized and
self-contained, with the ability to access any global state (with appropriate
consistency protection mechanisms in place).  Processes are created and then
waited upon.  Where message-based systems can be think of as locally
distributed, procedure-based systems are instead localized, modular code that
gathers the resources it needs and then does everything for itself.

The coolest part about the paper was the argument that these systems are
essentially logically equivalent, and are perhaps even easily transformed from
one into the other.  As people had spent decades designing and arguing about the
design and implementation of operating systems, this paper unifies those two
tracks of thought and says "It's alright -- you're both doing ok, and you're
not that different."  If considered in modern times, the authors might argue
that there is no point in arguing about the difference between the MacOS and
Windows, up to a constant factor of crashes.