book cover Chapter summaries

Part I

Chapter 1

Building a Simple Sequential Process

We introduce the syntax that CSP uses to create basic sequential processes:  prefixing, recursion and conditional, external and internal choice. The ideas and syntax of events and channels is introduced.  We learn the distinction between deterministic and nondeterministic choice. Examples include buffer and counter processes, and processes that describe a human's life in terms of constituent events. We see the ideas of traces and trace refinement. The FDR tool is introduced, as is its input language CSPM.

Chapter 2
Understanding CSP

This introduces the three types of formalism that this book uses to study CSP.  The first of these is algebra: we give and explain many examples of algebraic laws that CSP operators satisfy as well as comparing these to familiar laws  from arithmetic and set theory. Denotational semantics is based on behavioural  models such as traces: we introduce rules for computing traces and fixed points, as well as the Unique Fixed Point (UFP) principle for guarded recursions. Operational semantics are motivated by LTSs and the operation of FDR. We discuss trace refinement in depth and understand how to use it in formulating and running specifications for FDR.

Chapter 3
Parallel Operators

We meet the main parallel operators of CSP: synchronous, alphabetised, interleaving and generalised parallel. We see how processes can influence each other and transfer data along channels by synchronising on all or some of their events. Examples include people trying to agree on joint lives, bargaining in a shop, and the five dining philosophers who can deadlock through contention for resources. All of these are implemented in the accompanying example files, and we explain how these new operators affect FDR.

Chapter 4
Case Studies

This chapter introduces a number of case studies that we will meet many times in the rest of the book. The first of these shows you how you can code Sudoku puzzles in CSP and use FDR to solve them. The second shows how designing a routing network can easily lead to deadlock but how well designed networks can work reliably: we offer a number of alternatives for this in trees, general networks, rings and cartesian grids. The final example describes ways of overcoming an erroneous communication medium. As well as introducing the Alternating Bit Protocol and other ways of gaining reliability, we also introduce techniques for modelling error-prone systems in CSP.

Chapter 5
Hiding and Renaming

We meet some new operators. Hiding (P\X) allows you to conceal the internal actions of a process from the outside world. Renaming (P[[R]]) gives a flexible way of changing the labels on the events that a process communicates into different labels seen by the outside world. We see how hiding can create more natural models of parallel systems but can create a few theoretical difficulties, and how renaming combined with parallelism can create seemingly magical effects such as allowing Adam and Eve to achieve a variable renaming so each can have their own choice of apple. We introduce link parallel: a combination of parallel, hiding and renaming that makes constructing practical networks comparatively simple, and see how it can be used in building some natural dynamic networks such as one for implementing mergesort. We find that renaming can help FDR solve Sudoku puzzles faster.

Chapter 6
Beyond Traces

In previous chapters we have already discovered that traces give an incomplete picture of how processes behave, for example by failing to distinguish deterministic from nondeterministic behaviour and failing to capture deadlock properly. In this chapter we introduce the ideas of failures and divergences, which allow us to develop models that do capture these phenomena accurately. We see how these models allow to formulate and verify richer specifications on FDR. The phenomenon of divergence is described and we show one way to ensure it does not happen. We introduce lazy abstraction as a variant on hiding and show how it can be applied to capturing fault tolerance and computer security.

Chapter 7
Further Operators

This chapter completes the CSP language by introducing three operators that allow one process to hand control on to another. In sequential composition P;Q this happens when the first process terminates successfully via a special  action. In interrupt  P/\Q  this happens when the environment communicates an initial action of Q, and in throw  P[|A|>Q  any occurrence of an action in A within P counts as an exception which hands control to Q. We study in  particular the relationship between termination and parallel composition, and the effect of termination on the semantic models seen in earlier chapters.

Chapter 8
Using FDR

This chapter describes FDR and how to use this tool effectively. We describe how to interact with the FDR user interface, and how to program in CSPM  and its functional sub-language which is roughly equivalent to Haskell. We see what FDR can and cannot do, and how to play to its strengths. For example we see -- in an example based on finding a Hamiltonian path -- how it is better to code complex systems as parallel compositions rather than as large sequential processes.  We show in outline how FDR actually carries out a refinement check, and how this algorithm can be adapted to check determinism. We discuss the relative advantages of depth-first and breadth-first search, and finally show how compression operators that FDR supplies can enable one to (at least partially) overcome the state explosion problem that limits the size of systems the tool can handle.

Part II

Chapter 9
Operational Semantics

This is an in-depth study of the operational semantics of CSP and of the transition systems these are based on. We study the difference between finitely and infinitely branching transition systems, and between ordinary LTSs and ones where there may be acceptance or divergence information in additional labels on states. We show how CSP can be given an operational semantics in either the traditional Structured Operational Semantics (SOS) style or in a less flexible Combinator style that captures the spirit of CSP. We show what any operator with a combinator operational semantics can be expressed in CSP. Combinators lead to Supercombinators, the technique that FDR uses to implement transition systems effectively. Finally we show how formal "observations" of transition systems allow us to deduce what a process's traces, failures and divergences are.

Chapter 10
Denotational Semantics and Behavioural Models

This chapter studies denotational semantics in more depth than in the introductory chapters. We discuss the nature of behavioural models and what is required of them. We see how CSP definitions of operators naturally lead to distributive operators, and discuss topics such as full abstraction and congruence with operational semantics. We look in detail at each of the traces, stable failures and failures divergences model, and understand their healthiness conditions. We are able to specify determinism formally and study its relationship with the related concept of confluence.

Chapter 11
Finite Behaviour Models

There are two dimensions to study when we try to understand what behavioural models for CSP exist: what finitely observable things to record and what infinitely observable things to add to these. In this chapter we study what finite observation models exist. We examine the long-understood acceptances or ready-sets and refusal testing models. We discover two newer models: the acceptance-traces model of all finite linear observations, which is the finest possible model in this category, and the revivals model which sits above traces and stable failures, and below all other models. The structural result that proves this last fact puts ideas such as full abstraction in a completely new light. For each model we see what sorts of specification are best cast in
terms of it.

Chapter 12
Infinite Behaviour Models

Here we study how infinite behaviours can be added to these models. We get three classes of models in addition to the finite observation ones. The first of these, adding only divergence, can cope with finitely nondeterministic CSP with divergence strictness. The second, in which infinite traces and similar behaviours are added, can handle infinitely nondeterministic but divergence strict CSP. In the final class divergence strictness, or at least most of it, is removed. The resulting models are very natural except for requiring an exotic fixed point theory. We are able to get similar structural results to those in Chapter 11, for the divergence-strict models.

Chapter 13
The Algebra of CSP

We introduce a formal algebraic semantics for CSP. First we show how algebraic laws can systematically reduce all finite CSP terms to a head normal form, and therefore create an Algebraic Operational Semantics. Then we show how to reduce this to a normal form for the finest CSP model as developed in Chapter 12. We can then show how a small selection of optional laws such as  P=P [] P allow this normal form to be transformed into normal forms for any of the main models of CSP. This provides an elegant explanation for why the hierarchy of models developed in the previous two chapters is as it is.

Part III

Chapter 14
Timed Systems 1: tock-Time

This chapter develops the tock-CSP model of CSP introduced in Chapter 14 of TPC. We see how incorporating time into CSP models necessarily changes our understanding of the processes represented in this dialect. We develop a major  new case study: the Bully algorithm for leadership election, and explain the use of the tau-priority model that FDR uses to support timed analysis,  implementing the principle of maximal progress. We create timed versions of some of the routing algorithms described in Section 4.2 and compare their timedperformance.

Chapter 15
Timed Systems 2: Discrete Timed CSP

Here we look at the other CSP view of timed systems, namely Timed CSP in which time appears implicitly rather than as an explicit tock. We concentrate on the discrete variant of this, based on a semantic model involving tock to record the passage of time, and show how it can be implemented in a variant of tock-CSP. We use the alternating bit protocol as an example, and show how it relates to the more traditional continuous timed CSP through Ouaknine's theory of digitisation.

More about FDR

This chapter covers a number of more advanced topics in the implementation and  use of FDR. The first of these are in-depth studies of normalisation and the other compression operators that FDR uses, the latter including types of bisimulation and the CSP-specific diamond operator. We then look at ideas for partial order reduction and "lazy compression" including the chase operator.  We look at options for using parallel execution and improved memory usage in FDR. Finally we look at ways of making FDR handle a wider range of specifications, including ones that are not refinement-closed an distributive, certain types of infinite-state specifications, and LTL.

Chapter 17
State Explosion and Parameterised Verification

Here we look in depth at the state explosion problem and the related parameterised verification problem: how to prove a property of a general class of processes or networks. We look at induction and data independence separately.and combine them into data independent induction, which can handle many networks that neither parent method can. These include networks based on one of the routing examples from Chapter 4, which are proved correct in general. We tudy the topic of buffer tolerance: when can we prove a property of a system
with buffered channels by studying the corresponding unbuffered network, typically with a much smaller state space. This is closely related to the data-flow model of concurrency. Finally, we study how Counter-Example Guided Abstraction Refinement (CEGAR) can be integrated naturally into FDR

Part IV

Chapter 18
Shared-Variable Programs

This chapter shows how one can write a compiler in CSPM  and base a tool around the result. In this case that compiler is for shared variable programs, so this chapter is also an introduction to that subject. We study this through examples such as mutual exclusion, including Lamport's bakery algorithm, and the dining philosophers. The compiler is based on programs described in a CSP data type,  which is generated and interpreted by the front end of the tool, which we call SVA. SVA, a front end for FDR, is available from this book's web site.

Chapter 19
Understanding Shared Variable Concurrency

Here we look at shared variables in depth, including a study of dirty variabless via the bakery algorithm and Simpson's 4-slot algorithm (where we propose a version that avoids known difficulties arising from dirty flag variables). We also define what it means for one shared variable program to refine another one,and introduce various ideas necessary to create finite model-checked proofs of programs (such as the bakery algorithm) that involve unbounded integer types.

Chapter 20
Priority and Mobility

We look at two ideas which are not normally thought of as within the scope of CSP, namely priority and mobility. These are motivated by a CSP study of finding a (chess) knight's tour. We introduce a priority operator that is consistent with some of the standard models of CSP. As examples of priority we study Statecharts via a burglar alarm case study, and a simplified form of  these, namely two-phase synchronous automata. We contemplate how to add mobility into CSP's alphabet-based model of interaction and make a few tentative proposals, including a "closed-world" form of mobile parallel composition. This is illustrated via a model of a simple telephone network, which is also simulated in standard CSPM.