1. INTRODUCTION One of the greatest challenges facing software engineers is the management of change control. It has been estimated that the cost of change control can be between 40% and 70% of the life cycle costs . Software engineers have hoped that new languages and new process would greatly reduce these numbers; however this has not been the case. Fundamentally this is because software is still delivered with a significant number of defects. Capers Jones estimates that there are about 5 bugs per Function Point created during Development . Watts Humphrey found “… even experienced software engineers normally inject 100 or more defects per KSLOC . Capers Jones says, “A series of studies the defect density of software ranges from 49.5 to 94.5 errors per thousand lines of code .” The purpose of this article is to first review the fundamentals of software maintenance and to present alternative approaches to estimating software maintenance. A key element to note is that development and management decisions made during the development process can significantly affect the developmental cost and the resulting maintenance costs.
2. SOFTWARE MAINTENANCE Maintenance activities include all work carried out post-delivery and should be distinguished from block modifications which represent significant design and development effort and supersede a previously released software package. These maintenance activities can be quite diverse, and it helps to identify exactly what post-delivery activities are to be included in an estimate of maintenance effort. Maintenance activities, once defined, may be evaluated in a quite different light than when called simply “maintenance”. Software maintenance is different from hardware maintenance because software doesn’t physically wear out, but software often gets less useful with age and it may be delivered with undiscovered flaws. In addition to the undiscovered flaws, it is common that some number of known defects pass from the development organization to the maintenance group. Accurate estimation of the effort required to maintain delivered software is aided by the decomposition of the overall effort into the various activities that make up the whole process.
3. APPROACHING THE MAINTENANCE ISSUE Maintenance is a complicated and structured process. In his textbook, Estimating Software Intensive Systems, Richard Stuzke outlines the typical software maintenance process. It is apparent that the process is more than just writing new code.
The following checklist can be used to explore the realism and accuracy of maintenance requirements.
4. SANITY CHECKS Although sanity checks should be sought on a year-by-year basis, they should not be attempted for overall development. The reason for this is that maintenance activities can be carried on indefinitely, rendering any life-cycle rules useless. As an example, consider Grady (p. 17):
We spend about 2 to 3 times as much effort maintaining and enhancing software as we spend creating new software.
This and similar observations apply at an organizational level and higher, but not for a specific project. Any development group with a history will be embroiled in the long tail ends of their many delivered projects, still needing indefinite attention. Here are a few quick sanity checks:
o One maintainer can handle about 10,000 lines per year.
o Overall life-cycle effort is typically 40% development and 60% maintenance.
o Maintenance costs on average are one-sixth of yearly development costs.
o Successful systems are usually maintained for 10 to 20 years.
Finally, as in development, the amount of code that is new versus modified makes a difference. The effective size, that is, the equivalent effort if all the work were new code, is still the key input for both development and maintenance cost estimation.
5. FIVE ALTERNATIVE APPROACHES All software estimation techniques must be able to model the theory and the likely real world result. The real world scenario is that over time, the overlay of changes upon changes makes software increasingly difficult to maintain and thus less useful. Maintenance effort estimation techniques range from the simplistic level of effort method, through more thoughtful analysis and development practice modifications, to the use of parametric models in order to use historical data to project future needs.
5.1 Level of Effort As is sometimes the case in the development environment, software maintenance can be modeled as a level of effort activity. Given the repair category activities and the great variance that they show, this approach clearly has deficiencies. In this approach, a level of effort to maintain software is based on size and type.
5.2 Level of Effort Plus Stuzke proposed that software maintenance starts with basic level of effort (minimum people needed to have a core competency and then that that basic core staff must be modified by assessing three additional factors; configuration management, quality assurance, and project management. His process addressed some of the additional factors affecting software maintenance.
5.3 Maintenance Change Factor Software Cost Estimation with COCOMO II (Boehm 2000) proposes a deceivingly simple, but also quite useful methodology for determining annual maintenance. Maintenance is one of the menu selections in the menu bar. In COCOMO II Maintenance encompasses the process of modifying existing operational software while leaving its primary functions intact. This process excludes:
o Major re-design and re-development (more than 50% new code) of a new software product performing substantially the same functions.
o Design and development of a sizeable (more than 20% of the source instructions comprising the existing product) interfacing software package which requires relatively little redesigning of the existing product.
o Data processing system operations, data entry, and modification of values in the database.
The maintenance calculations are heavily based upon the Maintenance Change Factor (MCF) and the Maintenance Adjustment Factor (MAF). The MCF is similar to the Annual change Traffic in COCOMO81, except that maintenance periods other than a year can be used. The resulting maintenance effort estimation formula is the same as the COCOMO II Post Architecture development model.
As stated previously, three cost drivers for maintenance differ from development. Those cost drivers are software reliability, modern programming practices, and schedule. COCOMO II assumes that increased investment in software reliability and use of modern programming practices during software development has a strong positive effect upon the maintenance stage.
Annual Maintenance Effort = (Annual Change Traffic) * (Original Software Development Effort)
The quantity Original Software Development Effort refers to the total effort (person-months or other unit of measure) expended throughout development, even if a multi-year project.
The multiplier Annual Change Traffic is the proportion of the overall software to be modified during the year. This is relatively easy to obtain from engineering estimates. Developers often maintain change lists, or have a sense of proportional change to be required even before development is complete.
5.4 Managing Software Maintenance Costs by Developmental Techniques and Management Decisions During Development
When it comes to maintenance, “a penny spent is a pound saved.” Better development practices (even if more expensive) can significantly reduce maintenance effort, and reduce overall life cycle cost. The more effort put into development, the less required in maintenance. As an example, the software development cost and schedule can be significantly impacted (reduced) by letting the number of defects delivered grow. This cost and schedule reduction is more than offset by the increase in maintenance cost. The following discussion is an example of how management decision can significantly affect/reduce software maintenance costs.
Lloyd Huff and George Novak of Lockheed Martin Aeronautics in their paper “Lockheed Martin Aeronautics Performance Based Software Sustainment for the F-35 Lightning II” propose a series of development and management decision designed to impact and reduce software maintenance costs. They propose an eight step process to estimate and control software maintenance . Their proposed steps are:
5.5 A Parametric Assessment of Software Maintenance
Parametric models like SEER for Software allow maintenance to be modeled in either of two ways:
Estimating maintenance as a part of the total lifecycle cost. Choosing the appropriate Maintenance category parameters will include an estimate of maintenance effort with the development estimate for the individual software program. Several reports and charts show breakdowns of development vs. maintenance effort. This method is best used to evaluate life cycle costs for each individual software program.
Estimating maintenance as a separate activity. Using the appropriate maintenance parameters for the software to be maintained you can model the maintenance effort as a separate activity. This method will allow you to fine tune your maintenance estimate by adjusting parameters. Maintenance size should be the same as development size, but should be entered as all pre-existing code. This method can also be useful in breaking out total project maintenance costs from project development costs.
A good parametric estimate for maintenance includes a wide range of information. Critical information for completing a software maintenance estimate is the size or amount of software that will be maintained, the quality of that software, the quality and availability of the documentation, and the type or amount of maintenance that will be done. Many organizations don’t actually estimate maintenance costs; they simply have a budget for software maintenance. In this case, a parametric model should be used to compute how much maintenance can actually be performed with the given budget.
Estimating and planning for maintenance are critical activities if the software is required to function properly throughout its expected life. Even with a limited budget, a plan can be made to use the resources available in the most efficient, productive manner. Looking at the diagram above, you can see that not only are the multiple inputs that impact the maintenance, but there are several key outputs that provide the information necessary to plan a successful maintenance effort.