“In the past year, alone, we saw Lufthansa cancel the flights of over 3,000 booked passengers due to a computer fault in the check-in systems. We’ve seen insufficient testing lead to the failure of Nike’s i2 demand forecasting in 2000, costing the company more than a forecasted $100 million in sales. We’ve seen a failure to test for specific conditions contribute to the August 2003 blackout that affected the Northeastern United States and Canada. And beginning in June of last year bugs in connections between Hewlett-Packard’s legacy order-entry system and SAP systems caused a backlog of customers orders ultimately costing the company $40 million is lost revenue (CIO, 11/15/05).”
In fact, software errors, or bugs, are so prevalent and so detrimental that, according to a study published in 2002 by America's National Institute of Standards and Technology (NIST) software bugs cost the American economy $60 billion a year or about 0.6% of its gross domestic product.
Missed deployment and delivery dates, budget overruns, failure to comply with industry regulations, interrupted workflows and frustrated customers are the natural byproducts of application flaws. The current approach to resolving application problems, and ensuring software quality throughout the entire useful life span of the application, is simply not getting the job done. The time has come for a sea change in how we manage problem resolution, and ensure quality for complex, composite new applications, as well as aging legacy systems.
The Current Zeitgeist: Testing = Quality
Researchers from the Carnegie Mellon University estimate that programmers make between 100 and 150 errors per 1,000 lines of code. A plethora of similar research has clearly established the legitimacy of Quality Assurance as a fundamental life cycle process. Gone are the days when the length of the quality assurance phase was axiomatically the number of days between the end of coding and the release date. There are still some laggard organizations that task the QA team to “make-up” time lost in development, but the zeitgeist in software quality is still rooted in the concept that quality can be “tested in” to software applications. There are some solid economic roots to that thinking. In his 1981 software classic, Barry Boehm published some, at the time, eye-opening information to demonstrate how the cost for removing a software defect grows exponentially with the phase of the development lifecycle in which it is discovered.
Forward looking organizations realize that the new paradigm for life cycle application quality is to build quality assurance into every phase of the development life cycle. There’s no doubt that applying more rigour to the definition of business requirement and investing more in design reviews will result in software applications that are more aligned to the business needs (one dimension of total software quality). But many organizations are still locked in struggle to ensure the software works, as designed, and performs as expected, after deployment. To address that challenge, they are optimizing their application life cycle quality practices.
Optimizing Life Cycle Quality Practices
Optimizing quality across the application life cycle requires the integration of people, process and technology in an optimal organizational structure to deliver a very straightforward result – to help the development process deliver quality software faster, at a total lower cost, and to maintain high quality levels through the application’s productive life.
Leading companies are making organizational decisions to create separate QA organizations; they are elevating the role and stature of the QA engineer; they are implementing best practice processes with rigour. However, it is surprising that with the maturity and wide availability of testing technologies, the technology dimension of many QA strategies is still so undeveloped. The technologies available for testing are usually categorized into two categories – tools that support the manual testing process and tools that execute some automated testing.
The current ‘sweet spot’ for QA technology – the area where technology can yield significant ROI lies in automating as many aspects of the manual testing process as possible. According to the IT industry consultancy, Gartner, 80% of all applications are either not tested or are only tested manually before being delivered into production. There are a number of solutions that manage the administrative process of testing (test planning, execution tracking, defect tracking) such as Test Director from Mercury Interactive and ClearQuest from IBM’s Rational division.
But, many companies are beginning to realize the powerful advantage that comes from automating the tasks of the QA Engineer, not just the QA Manager. For example, consider what happens when the QA Engineer does discover a problem – the process is particularly cumbersome. First, the problem must be replicated (to ensure that it is reproducible.) Then, detailed documentation is prepared to record the exact sequence of steps executed, data entered, etc. The process of replicating and documenting problems found in the QA process can consume 25-50% of the total QA effort.
This documentation is sent to the development team for analysis, which usually starts with the developer’s attempt to recreate or replicate the problem situation so that traditional debugging tools can be deployed. This generally requires the developer to try to recreate the application environment in which the problem occurred and recreate the problem using the information sent by the QA team.
Unfortunately, although a lot of time is spent on documenting problems found in QA, many problems are incorrectly communicated or insufficiently described, which results in development not being able to recreate the problem to analyze for root cause. Despite a significant amount of “ping pong” communication between developers and testers, a troubling 22% of problems found in QA are not reproducible by development. With more complex composite applications, that number is rising towards 30%. The net result is a significant number of problems that were actually found in the QA process are not fixed before the product is released. As we learned from Barry Boehm’s 1981 research, fixing a problem before the product reaches general availability can reduce the cost of fixing it by orders of magnitude.
One innovative organization, Cerner Corporation has implemented an application problem resolution system from Identify Software, called AppSight, to automate the problem handling aspects of software bugs discovered during QA. As a world leader in healthcare software solutions, Cerner is replacing paper charts with intelligent, interactive electronic forms designed to improve patient care and business management. Until recently, Cerner took the industry-standard approach to manual QA testing, with industry-typical costs and results. ”Given the labor it takes to assure the software quality demanded in healthcare, where lives are literally at stake, we were immediately intrigued by AppSight,” commented Owen Straub Cerner’s VP of Engineering.
Beyond the Testing Mandate – Dealing with the software quality “afterlife” in production
Any organization that is serious about software quality needs to internalize the concept no matter how effective your quality testing processes are, they can not completely ensure that your software is free of bugs. As is often observed, testing can only demonstrate the presence of defects, not their absence.
Today’s applications are complex systems, comprised of numerous loosely coupled processes owned and managed by multiple parties. There has never been an instance, in recorded history, of a complex system (organic or man-made) that doesn’t sometimes fail – that’s why we have doctors, car service workshops and a hole in the ozone layer. In testing, you simply do not get the functional "stress test" of hundreds or thousands of users performing functions in unexpected combinations. And, research indicates that approximately 40% of application problems are not caused by a software bug at all, but by some condition in the system environment in which the application is deployed. But these very real problems still need to be solved -- hopefully without needlessly escalating the issue to the development team, a practice that significantly affects downstream development dates. Any organization that is serious about software quality needs to extend their quality business system to quickly fix problems that occur once the application is deployed.
CGI, a large systems integration house, has also deployed AppSight – this time as a strategy to manage quality in applications it has developed and deployed for its remote customers. In one example, the Toronto business unit of CGI, which serves some of the largest telecommunications companies in Canada, had a major performance problem with a web-based cell phone activation application. The commitment in the service level agreement (SLA) was for a 10-minute transaction completion time, but the transaction was actually taking 18 minutes and developers at CGI’s Centre for Testing and Quality couldn’t figure out why. Using Mercury Interactive Corp.’s Load Runner didn’t help much either — the load performance and stress tool, which displays information on CPU usage, memory and the top 10 transactions didn’t uncover any problems in the pre-production environment. However, using AppSight’s Black Box technology, CGI was able to record the actual application behavior as the application ran in the live production environment. AppSight consolidated all the relevant information from several levels onto one screen so that CGI could see a backlog of pending requests and understand the need to increase the number of threads. "Armed with the insight captured by AppSight in the live environment, the CGI team was quickly able to implement a simple fix to increase the end-to-end throughput by 50%, slashing average phone activation times from 18 minutes to 9 minutes. ” AppSight allows us to analyze production problems we’d never find through performance testing and a traditional QA process, so we can recover quickly and continually deliver the highest quality applications to our clients,” said David O’Leary, Director, CGI Centre for Testing and Quality.
Forward thinking companies are far beyond thinking of quality assurance as a quick shakedown period between development and delivery. They are approaching the software quality life cycle challenge with a holistic approach that considers the dimensions of organizational issues, process, people and technology. These organizations are delivering higher quality solutions to their constituents much faster, and supporting those applications throughout the life cycle with less expense and higher end-user satisfaction.
Lori Wizdo is V.P. of Market Development at Identify. Her 25 year career has included marketing, sales, and strategic planning experience in the information systems and software industries. Immediately prior to joining Identify, Ms Wizdo was V.P. of Strategy and Business Development at Kamoon Inc., a provider of collaborative software solutions, where she was responsible for strategy, market development and field operations. Prior to joining Kamoon, Wizdo held several senior-level strategic marketing roles at Unisys Corp., where she led the successful market launches of the Unisys e-Business strategy and Microsoft business initiatives. A Summa Cum Laude graduate of the Pennsylvania State University, Wizdo has held other senior-level positions with a diverse set of technology firms, from start-ups to global corporations, including Xerox and NCR Corp.
You can contact Ms. Wizdo at [email protected]