Failure and Fault Tolerance in DpiF.
Adrian Francalanza
Abstract
In this talk I present the behavioural study of distributed
programs in the presence of location failure. I develop a distributed
pi-calculus called DpiF, where the terms are configurations with
network state information. Two novel constructs in this language are a
kill constuct that changes the state of a location during reduction and
a synchronous move between locations that can be used as a "ping" to
determine the current state of a location.
I overview the reduction semantics of this language together with a
reduction barbed congurence for DpiF configurations. Subsequently, I
present an LTS together with a bisimulation and show that the
reduction barbed congurence concides with the bisimulation.
In the second part of the talk, I discuss fault tolerance for DpiF
terms. I develop a notion of "partial views" for DpiF
configurations and use this to formalise the definition of fault
tolerance. I then develop tractable techniques for determining
fault tolerance in DpiF configurations.
This talk is based on ongoing work, supervised by Prof. Matthew
Hennessy.