Defining "firmware quality"

chinese + hindu programmers in this project divided by total number of programmers in this project.

VLV

Up time vs. problem time is a common one -- i.e. how many hours you can drive your car without the throttle sticking wide open, or how long your pacemaker works without skipping a beat.

If the system provides discrete events, like readings or firings or whatever, the number of attempts vs. the number of failures.

--
Tim Wescott
Control system and signal processing consulting
www.wescottdesign.com

Tim's response is excellent. In other words, you and your customer will define quality. Your job as the firmware QA engineer is to define the metrics of quality and then go ahead and measure it.

"If you can't measure it, you can't manage it."

JJS

--- news://freenews.netfront.net/ - complaints: snipped-for-privacy@netfront.net ---

Nel suo scritto precedente, John Speth ha sostenuto :

Measuring "something" is the very first step, that's right. Here's the question: are we sure that we're measuring the right thing? If we were speaking about a mechanical components (say an aircraft wing, for example) we could define a set of mechanical test cases because we all know what we're looking for (a wing that can make an airplan fly). As a matter of fact, these tests don't care about the customer definition of quality (ok... customers like not to crash while flying to Hawaii

-.- )

But when we talk about firmware, are we able to define something like this? Or we have to lean only on customer aspectatives? Wouldn't it be better if we can define a metric that allows to compare initial requisites with produced firmware? Are we falling back to bugs counting?

Regards!

--
http://www.grgmeda.it

I think you're confusing the question "What is quality?" with the question "How do we accurately express, predict and measure quality?".

I'd expand on that, but about all I can say at this point is "and that's a hard question to answer!".

--
Tim Wescott
Control system and signal processing consulting
www.wescottdesign.com

You have to define a line-of-code (some will argue about the criteria to use, there -- and how it can be "manipulated" by folks trying to "improve quality" by manipulating the LoC metric :>).

E.g., do you include "dead code" in your LoC metric?

You could also use function points, etc.

And, how do you define a "bug"? Are all bugs "equal"? (I would argue that they are NOT)

Perception implicitly assigns some "level of quality" to the perceiver. :> E.g., if the user isn't very adept, does that give you a false sense of having produced a "quality" product? What happens if the user becomes more "advanced" with experience? Suddenly, the same product in the same user's hands has *less* quality than it did when the user was inexperienced/naive?

*If* you had good specifications (exhaustive), quality could be measured in terms of "deviations from specification". Few pieces of code are written with such specification detail, though (perhaps the military?).

So, you then have to address "specification quality" :-/

And, regardless, what are you going to *do* about that quality (or lack thereof) *if* you can find a way to actually objectively measure it?

Presumably, those tests are designed to verify that the wing conforms to some *specification* regarding its weight, mechanical strength, conformance of the airfoil to "ideal" contour, etc.

Those specifications were, in turn, derived from other specifications: "We have to be able to carry X passengers and Y cargo over adistance of M miles with a fuel efficiency of E...".

You need specifications ("requisites") against which to measure. Quality, in the software sense, is how well you conform to your requirements (how pretty your code looks might be nice, too, but that doesn't directly indicate how well it does what it is *designed* to do)

How do you define the "quality" of a homebuilder:

- Number of nail-pops_in_the_drywall after 6 months?

- Number of floor_creaks?

- Ounces_of_water_per_inch_penetrating_the_roof_per_inch_of_rain? etc.

Many industries have "invented" (seemingly) meaningless metrics simply because you have to count *something*...

I suspect most (software) products are now evaluated solely in terms of "number of units sold". :< And, as long as that number is high enough to keep the company profitable, they keep doing what they are doing.

Well in the "software sense" we've advanced a lot more than that. We have now the "SQuaRE series" of international standards (ISO/IEC

25000-ISO/IEC 25051).

In fact they address some of the subtleties written earlier like the question of the wing 'quality' versus the passenger's perceived attributes for a plane quality.

Ditto.

Or because those 'somethings' are meaningful in the chain of supply becoming antecedents for the attributes ultimately perceived by the final user?

Yes. . . the good enough SW that made some Redmond company the wealthiest in the Globe ;-)

--
Cesar Rabak
GNU/Linux User 52247.
Get counted: http://counter.li.org/

I think you mean bugs per KLOC, as in kilo or thousands of lines of code. Bugs per LOC indicates a complete lack of quality.

As others here have indicated, quality is a rather subjective matter. We work with firmware used in Safety-Critical applications, medical, avionics, etc. In these quality is defined more objectively in terms of software failures per hours of operation, or indirectly by the type of verification and validation required by a given standard. In both cases the measurement is for a speicified safety level which range from not considered a safety concern to life critical.

--
Scott Nowell
Validated Software
Lafayette, CO 



__________ Information from ESET NOD32 Antivirus, version of virus signature
database 4935 (20100311) __________

The message was checked by ESET NOD32 Antivirus.

http://www.eset.com

We

s,

n

ment

ure database 4935 (20100311) __________

Isn't that an oxymoron? Like military intelligence, sorry coundn't resist.