The Process Group Approach To Reliable Distributed Computing
Kenneth P. Birman

Paper's Contribution: 

Essentially a marketing pitch for the ISIS Toolkit which is an
application programming interface (API toolkit) that provides
communication channel mechanisms to aid programming for distributed
process group applications, including features such as correct
communication ordering, fault tolerance, data replication, etc...

Paper's Critical Ideas:

Actually there's very few new ideas, in terms of distributed theories
in this paper, instead it builds upon the theories of previously
published works to provide an API toolkit, ISIS, that applications
programmers can use to develop their distributed applications.  One of
the features the ISIS toolkit provides is correct message delivery
ordering, and the algorithms used to achieve this were built upon
Lamport's work on causal ordering of events in distributed systems,
especially with regards to ISIS's use of "virtual synchrony" that
allows for higher performance than "close synchrony" since it takes
advantage of not requiring processors to block for concurrent messages
that don't have causal relationships.  This is especially important in
heterogeneous environments where you don't want a set of "fast"
processors continually waiting for a slow one.

The paper investigates many of the problems that plague distributed
applications are presented and touts how ISIS helps programmers deal
with them.  In an interesting example it presents a security brokerage
distributed application that demonstrates many of the crucial issues
that a distributed application must consider, such as unreliable
communication channels, state transfers, fault-tolerance and recovery.
Their fail-stop model, of not allowing a computer that has experienced
a failure to try to recover its state, but rather must rejoin the
process group from the "start" state, is certainly a nice
simplification, but could be an oversimplification if failures aren't
rare.

Paper's Flaws:

My personal peeve about this paper is the fact it never defines what
the ISIS acronym stands for.  I hate that.  But beyond that this paper
doesn't bring a whole lot to the table other than acting "coming out"
party for ISIS.  The paper may have been more interesting if it delved
into some examples of whether these communication channel mechanisms
actually produce success, are efficient in real applications

Relevance to modern systems:

Historically one of the largest road-blocks to further widespread
adoption of distributed systems and parallel processing has been the
subtle issues, relating to ordering and synchronization, involved in
application development, making these types of systems too difficult
and costly to implement reliably.  By providing programmers with an
easy to use set of tools that help manage these issues for them,
there's a good chance the barriers would be removed and a new era of
prevalent distributed/parallel programming would ensue.  ISIS is an
attempt to be the easy-to-use toolkit for distributed application
development.