Re: Predicting Software Error Density

[mlo@cherry14.cray.com (Mick Oyer)]

In article <759@ajpo.sei.cmu.edu>, dyer@ajpo.sei.cmu.edu (Richard Dye) writes:
|> 
|> I am interested in references to any articles that would help me
|> predict the error density in fielded software.  Has anyone done any
|> work or published any papers that describe the defect discovery and
|> fix curves over time?

I'm involved in doing some of this type of work here at Cray.  One or two
items of interest:

    There's an article in Business Week, 2/11/91 that talks about Japanese
    vs. American Fault Density.  According to the article, the Japanese are
    producing 12,447 lines of debugged source code per man-year, with an
    error density of 1.96 "technical failures per 1000 lines of source code".
    Us lowly Americans are producing code at a density of 4.44 TF/Kline of code.
    at a rate of 7,290 lines of code per man-year.  I especially like the last
    sentence of the article - "Luckily, U.S. software experts have been monitoring
    developments in Japan for years and still have time to learn some Japanese
    tricks before it's too late".  Where have I heard that before?

    This article is written in conjunction with a new book on the shelves:
    "Japan's Software Factories" by Michael A. Cusumano, Associate Professor
    of Mgmt. at MIT.  Looks like interesting reading.  I've got one on order,
    hopefully it will answer my question - "Where do some of these numbers come
    from?"

   | Michael (Mick) Oyer   MAIL   :  mlo@cray.com  ||  uunet!cray!mlo     |
   | Sr. Software Analyst  AT&T   :  work: (612)-683-5855                 |
   | Cray Research, Inc.   USNAIL :  655F Lone Oak Dr., Eagan, MN 55120   |
--
   | Michael (Mick) Oyer   MAIL   :  mlo@cray.com  ||  uunet!cray!mlo     |
   | Sr. Software Analyst  AT&T   :  work: (612)-683-5855                 |
   | Cray Research, Inc.   USNAIL :  655F Lone Oak Dr., Eagan, MN 55120   |