Title
A Note on Distributed Computing

Author
Jim Waldo and Geoff Wyant and Ann Wollrath and Sam Kendall

Distibuted Systems are intrinsically different from single address
space systems for four main reasons:
1) unknown latency (must take latency into account)
2) different model of memory access (no globals or shared memory)
3) concurrency
4) partial failure.  In local systems, failure is almost always
complete.

One postulated approach to distributed programming is just designing
fixed interfaces and then only using those to ensure the design,
dividing these among different machines when convenient.  This has the
difficulties of not only being inefficient -- both programming to
fixed interfaces can be slow and not "cheating" when possible leads to
slower programs -- but also of being incorrect through not taking into
account that objects can fail to respond even when you follow their
interfaces.

Waldo argues that object interfaces must explicitly expose the fact
that they are being conducted over a network.

He also discusses how changes to programming languages to simplify
communication does not change the essential problem.  He notes that
changes to languages have occurred several times, but none of them
has, or can, eliminate the essential complexity inherent in
distributed systems.

He argues that latency and memory access problems could be masked, but
the problems of concurrency and partial failure remain, and to some
extent are what define distributed computing.

He also argues that multithreaded programming and distributed
programming are different, because there is no indeterminancy in the
invocation of operations.