* Re: software metrics
1995-03-08 17:08 software metrics Patrizia Sgubbi
2014-08-19 22:08 ` sameen.arshad
@ 2014-11-15 2:11 ` rriehle
2014-11-15 14:24 ` brbarkstrom
2 siblings, 0 replies; 7+ messages in thread
From: rriehle @ 2014-11-15 2:11 UTC (permalink / raw)
On Wednesday, March 8, 1995 12:15:51 PM UTC-8, Patrizia Sgubbi wrote:
> Hi I am working on a graduation paper about software metrics. more
> precisely, I am articulating my work in 3 phases. The first is just a
> review of existing techniques, starting from the basic papers towards
> modern object-oriented approaches. In the second phase I have to find a
> way to apply metrics to a special case: formal specifications. I am
> examinating the design phase of a software project in different
> versions, as about ten groups of students have written them. The
> language adopted is Larch. This design phase follows an analysis phase
> developed using Z language.
> I am not interested about the implementation level. My final goal would
> be to develop a tool which automatically takes larch scripts as inputs
> and gives as output an estimation of their quality.
>
> IMPORTANT: I am looking for papers about software metrics in general and more
> specifically about software metrics in Object oriented environment and
> software metrics applied to formal specification or software metrics
> related to analysis and design phase of the software life cycle.
>
>
> Any suggestion?
>
> Please e-mail to
> sgubbi@cs.unibo.it
>
>
> Thank you all,
> Patrizia
>
>
> ---------------------------------------------------------------------
> - Patrizia Sgubbi "... like a true nature's child -
> - e-mail: sgubbi@cs.unibo.it " we were born, born to be wild..." -
> - tel: --39/-51/372917 (Steppenwolf) -
> ---------------------------------------------------------------------
>
> --
> ---------------------------------------------------------------------
> - Patrizia Sgubbi "... like a true nature's child -
> - e-mail: sgubbi@cs.unibo.it " we were born, born to be wild..." -
> - tel: --39/-51/372917 (Steppenwolf) -
> ---------------------------------------------------------------------
Quite a few books have been written on this topic. The question we need to ask is, "What do we mean by metrics?"
We have metrics for processes, for products, for performance, etc. That leads to the question, "What do we want to measure?"
Consider the product metrics. In the world of physical engineering, we have design metrics based on the laws of physics. In software, we need to invent those metrics. In physical engineering, we have "sensitivity" metrics that involve gradual changes; software tends to be discrete. Gradual changes, while possible, are not quite the same. A single bit with a zero instead of a 1 can crash an otherwise robust design.
In physical engineering, we can easily design to tolerances. Not so easy (but still possible) in software engineering. Do you want to study design metrics?
In process metrics, we have a large number of things we can measure. Often overlooked, but of vital importance in modern software engineering is the area of risk metrics. How do we measure the various kinds of risks? How do we learn from past risks? In my software risk management course, I emphasis Bayesian probability as one of the tools for moving toward Level Five in the CMM/CMMi scale. It is an exciting area of risk management (OK, only exciting for a really boring person such as me).
What do we measure? Size? Progress? Performance? Failure rate? Defect rate? Project activities and targets? SLOC ratios? Function point ratios? Customer satisfaction (seldom really measured by technologists)? Amount of time and money for adaptation? Conformity to standards? On and on and on?
What kind of metrics of value to you on any given project, for any given software product?
Important!! Metrics are no good unless they allow you to make decisions based on those metrics. A lot of metrics are what I call "Hmmmmmmmmmmmmm" metrics. You gather all those numbers, look at them and say, "Hmmmmmmmmmmm!" If the metrics program does not lead to valuable decision-making, you have wasted your time.
Hope this is helpful.
Richard Riehle, PhD, International Technological University, San Jose, CA
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: software metrics
1995-03-08 17:08 software metrics Patrizia Sgubbi
2014-08-19 22:08 ` sameen.arshad
2014-11-15 2:11 ` rriehle
@ 2014-11-15 14:24 ` brbarkstrom
2014-11-15 19:40 ` Brian Drummond
2 siblings, 1 reply; 7+ messages in thread
From: brbarkstrom @ 2014-11-15 14:24 UTC (permalink / raw)
>
>
> Thank you all,
> Patrizia
>
>
You might take a look at Barry Boehm's books on Software Cost Estimation
[the classic is Software Engineering Economics, 1981; the newer version
is Software Cost Estimation with COCOMO II, 2000]. The competitor is
Capers-Jones with his Function Point Analysis. He was big on metrics
that would demonstrate his approach was better than anybody else's.
You might find McConnell, S., 1996: Rapid Development: Taming Wild Software
Schedules, Microsoft Press of some help. You might also take a look at the
work Harlan Mills did (either Google or try Amazon to get a book list).
Metrics that relate to the number of errors are probably useful. There is
some work on that for Ada versus other languages, but I don't have a specific
reference. You might try `software error rate' or similar phrases on Google.
As far as I can tell, software metrics has become a rather passé topic.
OTOH, the GNAT GPS tool computes a relatively large number of metrics.
I haven't experimented with languages other than Ada, but I occasionally
track my own programming by running the metrics to get SLOC and number of lines
per unit. There are others, such as cyclometric complexity and related
measures, but I'm not doing really big projects right now.
Bruce B.
^ permalink raw reply [flat|nested] 7+ messages in thread