Date: Sun, 09 Apr 1995 18:17:10 EST
From: "Rob Slade, Social Convener to the Net" <roberts@mukluk.decus.ca>
Subject: "Metrics and Models in Software Quality Engineering" by Kan

BKMAMSQE.RVW   940221
 
"Metrics and Models in Software Quality Engineering", Kan, 1995, 0-201-63339-6,
U$39.75
%A   Stephen Kan
%C   1 Jacob Way, Reading, MA   01867-9984
%D   1995
%G   0-201-63339-6
%I   Addison-Wesley
%O   U$39.75 800-822-6339 617-944-3700 Fax: (617) 944-7273
%P   344
%T   "Metrics and Models in Software Quality Engineering"
 
This is an introduction to quality enhancement methods and measurements.  It is
broadly based, but also gives depth to the discussion in many areas.  Areas
such as the definition of quality and statistical modelling, which may be new
to software developers and managers, start from simple and clearly explained
concepts.
 
Chapters cover the definition of quality development and quality management
models, measurement theory, software quality measures (metrics), quality
identification tools, defect removal, the Rayleigh model and expectation of
defects, exponential models for expectations, the use of Rayleigh and other
models in management, complexity, and customer satisfaction measurement.  There
is very thorough bibliographic material at the end of each chapter.
 
The quality of this work, in one sense, is beyond question.  In terms of the
book's own definition of "fitness for use," however, there are some points to
be raised.
 
Software, as we all know, is digital.  It is subject to catastrophic failures
which cannot be predicted by statistical models of "errors per thousand lines
of code".
 
The need for customer input is repeatedly stressed.  The Patriot missile system
failure, mentioned in the book, was based on a customer requirement for a
maximum fourteen-hour run time.
 
Customer specifications, obviously, required an error recall function for the
Therac 25.  Unfortunately, neither the customer nor the company thought it
necessary to have a final check display of the parameters taken by the machine,
assuming the recall would work in all cases.
 
Mention is made of "object orientation" and the reuse of software.  It was
reuse of an object, valid in its own environment, which led to the Pentium
fiasco.
 
Finally, there is the *cost* of quality, pointed out in "Four Days With Dr.
Deming" (cf. BKDEMING.RVW).  Kan's work, voluminous as it may be, is only an
introduction.  Thus, it would be primarily of use in a very large effort where
the project manager had technical and statistical assistance simply for project
management.  As an example, I give you the Internet.  The "quality standard" of
"rough consensus and running code" must make quality specialists shudder. 
Without it, however, we'd be looking at a network based on OSI.  (No sniggering
>from you UDP hackers in the back, there.)
 
copyright Robert M. Slade, 1995   BKMAMSQE.RVW   950221

==============
Vancouver      ROBERTS@decus.ca                  | "90% of infections
Institute for  Robert_Slade@sfu.ca               |  are Stoned."
Research into  Rob_Slade@mindlink.bc.ca          | 
User           Rob.Slade@f733.n153.z1.fidonet.org|      - the viral corollary
Security       Canada V7K 2G6                    |        to Sturgeon's Law