2008 Woodie Awards
Research, industry differ in priorities
Aaron Fetterman
Issue date: 2/18/08 Section: News
Software maintenance can take up to 85% of the budget for a software product. Understandably, then, it has been a large focus for research: there are methods to refactor code, find repeated sections, do program slicing and aspect analysis, and automatically migrate the language, technology, or platform. Unfortunately, these methods are rarely implemented in the industry.
In a presentation Wednesday, February 13, James R. Cordy from Queen's University in Ontario, Canada explained, "why some things that my colleagues do in research won't ever show up in industry." Cordy was vice president and chief research scientist at Legasys for six years.
The most important thing for the academics to do, says Cordy, is to understand the industry, and to create services that assist the developers instead of tools that would replace them. Cordy broke his understanding into four main reasons: the business risk reality, the technological reality, the social reality, and the economic reality.
The business reality is this: large software systems serving industries like the financial sector cannot fail. How often does an ATM not work? An hour of downtime could cost a bank billions of dollars. The main focus for these companies is not the cost of software maintenance, but the risk of failure. Therefore, of the software maintenance budget, they spend 70-80% on testing. Because everything is tested so thoroughly, they attempt to make as few changes as possible, so that it affects the fewest lines.
Cordy used the Y2K bug as an example throughout his presentation. Because dates were represented with two digits, years after the millennium (01, 02) seemed smaller to the computer than years before the millennium (97, 98). The obvious solution would be to use four digits for the date, but that would have changed 15% of the source lines. Every one of those lines would then have to be re-tested. Using an imperfect windowing solution, only 0.3% of the lines needed to change.
In a presentation Wednesday, February 13, James R. Cordy from Queen's University in Ontario, Canada explained, "why some things that my colleagues do in research won't ever show up in industry." Cordy was vice president and chief research scientist at Legasys for six years.
The most important thing for the academics to do, says Cordy, is to understand the industry, and to create services that assist the developers instead of tools that would replace them. Cordy broke his understanding into four main reasons: the business risk reality, the technological reality, the social reality, and the economic reality.
The business reality is this: large software systems serving industries like the financial sector cannot fail. How often does an ATM not work? An hour of downtime could cost a bank billions of dollars. The main focus for these companies is not the cost of software maintenance, but the risk of failure. Therefore, of the software maintenance budget, they spend 70-80% on testing. Because everything is tested so thoroughly, they attempt to make as few changes as possible, so that it affects the fewest lines.
Cordy used the Y2K bug as an example throughout his presentation. Because dates were represented with two digits, years after the millennium (01, 02) seemed smaller to the computer than years before the millennium (97, 98). The obvious solution would be to use four digits for the date, but that would have changed 15% of the source lines. Every one of those lines would then have to be re-tested. Using an imperfect windowing solution, only 0.3% of the lines needed to change.
Be the first to comment on this story