Setting the stage

Is there enough legacy software in the world to justify investments in software renovation technology? It turns out that we are living on a software volcano: large numbers of new and old software systems control our lives. We admire the sheer bulk of the magnificent volcano, benefit from the fertile grounds surrounding it, yet at the same are suffering from frequent eruptions of lava, steam, and poisonous gas, uncertain what is going on within the volcano, and when the next large eruption will be.

The figures collected by Jones [31] provide insight in the size of the problem. He uses the function point (FP) as unit of measurement for software. It abstracts from specific programming languages and specific presentation styles of programs. The correlation between function points with the measurement in lines of code differs per programming language, and is summarized in Table 1(a). Another point of reference is that the size of Windows 95 is equal to $8.5 \times 10^4$ FP.

The total volume of software is estimated at $7\times10^9$ FP (7 Giga-FP). The distribution of the various programming languages used to implement all these function points is summarized in Table 1(b). Older languages dominate the scene: even today 30% of the 7 Giga-FP is written in COBOL. If we (hypothetically) assume that all software is written in COBOL we get an estimation (via 107 COBOL statements per FP) of $6.4\times10^{11}$ COBOL statements for the total volume of software.

As measure of software quality (or rather, the lack of it), Jones has estimated that on average 5 errors occur per function point. This includes errors in requirements, design, coding, documentation and bad fixes. The result is a frightening figure of $35\times10 ^ 9$ programming errors (35 Giga-bugs) waiting for a chance to burst out sooner or later.

Table 1: (a) Function Points versus Lines of Code; (b) Distribution of languages

Language	Statements/FP
Assembler	320
C	128
Fortran77	107
Cobol85	107
C++	64
Visual Basic	32
Perl	27
Smalltalk	21
SQL	13

Language	Used in % of total
COBOL	30
Assembler	10
C	10
C++	10
500 other languages	40

(a)

(b)

Developing better ways of developing new software will not solve this problem. When an industry approaches 50 years of age--as is the case with computer science-- it takes more workers to perform maintenance than to build new products. Based on current data, Table 2 shows extrapolations for the number of programmers working on new projects, enhancements and repairs. In the current decade, four out of seven programmers are working on enhancement and repair projects. The forecasts predict that by 2020 only one third of all programmers will be working on projects involving the construction of new software.

Table 2: Forecasts for numbers of programmers (worldwide) and distribution of their activities

Year	New projects	Enhancements	Repairs	Total
1950	90	3	7	100
1960	8500	500	1000	10000
1970	65000	15000	20000	100000
1980	1200000	600000	200000	2000000
1990	3000000	3000000	1000000	7000000
2000	4000000	4500000	1500000	10000000
2010	5000000	7000000	2000000	14000000
2020	7000000	11000000	3000000	21000000

Therefore, we must conclude that the importance of maintenance and gradual improvement of software is ever increasing and deserves more and more attention both in computer science education and research.