Distributed Snapshots: Determining Global States of Distributed Systems
Chandy and Lamport

This paper presents taking snapshots of distributed systems into
global states to determine if the distributed process has reached a
stable property.  A stable property is reached when each snapshot of
the global state can be resolved to true for the particular property
you’re checking.  Stable property examples are when process
computation has ended and system is deadlocked.

Chandy and Lamport present a way to capture a global state by
processes capturing their own state, and then using a "marker" message
to tell other processes to capture their state.  When analyzing the
global snap shots the events before the marker is sent and received
represent that global state.  By observing consistent global states
the system can be evaluated to have a stable property.

Assumptions are assumed that messages sent are received in the same
order making this algorithm simpler.  Removing these assumptions will
make things more interesting.

Chandy and Lamport tackle a hard problem of analyzing the global state
of a distributed system.  This paper providing assumptions proves how
this can be solved and what value this solution can be to a
distributed system.