The 1992 and 2002 Developments Versus The 2002 Industry Trend ...
9
pages
English
Documents
Le téléchargement nécessite un accès à la bibliothèque YouScribe Tout savoir sur nos offres
9
pages
English
Documents
Le téléchargement nécessite un accès à la bibliothèque YouScribe Tout savoir sur nos offres
Publié par
Langue
English
MEASURES FOR EXCELLENCE
Telecom Software Benchmarks
10 Years Apart
1992-2002
Copyright J.W.E Greene
QUANTITATIVE SOFTWARE MANAGEMENT LTD
7 Rue Fenoux 41A Aynhoe Road
75015 Paris Internet: qsm.europe@pobox.com London W14 0QA
Tel: 33-140-431210 CompuServe: 100113,3364 Tel: 44-207-603-9009
Fax: 33-148-286249 www.qsm.com Fax : 44-207-602-6008
Telecom Software Benchmarks: 10 Years Apart 1992-2002
Telecom Background:
What happened in 1992 and 2002?
We benchmarked in 1992; 10 years later we repeat the benchmark. Basically
it’s the same organisation building the same type of software for Telecom
products. High software content determines the time to market and reliability.
10 years is a long time so why the gap? Well objective benchmarking is risky.
The original benchmark revealed local management misconceptions about
productivity. They were betting that software re-use provides significant
process productivity benefits. The 1992 results showed otherwise. Certainly
re-use gave major benefits by reducing the size of the developed software but
this was independent of the local team process productivity. Moreover the
results highlighted that development process productivity and time pressure in
addition to size are major independent factors determining time, effort and
reliability. Management closed its eyes to the evidence. So as the
messengers we ...
Publié par
Langue
English
MEASURES FOR EXCELLENCE Telecom Software Benchmarks 10 Years Apart 1992-2002
7 Rue Fenoux 75015 Paris Tel: 33-140-431210 Fax: 33-148-286249
Copyright J.W.E Greene QUANTITATIVE SOFTWARE MANAGEMENT LTD 41A Aynhoe Road Internet:qsm.europe@pobox.comW14 0QA London CompuServe: 100113,3364 Tel: 44-207-603-9009 www.qsm.com Fax : 44-207-602-6008
Telecom Software Benchmarks: 10 Years Apart 1992-2002
Telecom Background: What happened in 1992 and 2002? We benchmarked in 1992; 10 years later we repeat the benchmark. Basically it’s the same organisation building the same type of software for Telecom products. High software content determines the time to market and reliability. 10 years is a long time so why the gap? Well objective benchmarking is risky. The original benchmark revealed local management misconceptions about productivity. They were betting that software re-use provides significant process productivity benefits. The 1992 results showed otherwise. Certainly re-use gave major benefits by reducing the size of the developed software but this was independent of the local team process productivity. Moreover the results highlighted that development process productivity and time pressure in addition to size are major independent factors determining time, effort and reliability. Management closed its eyes to the evidence. So as the messengers we got shown the door. Time rolls by. Commercial pressures are enormous- its Telecom. The company is committed to move up the CMMI maturity levels. The goal is Level 2 where Measurement and Analysis is a Key Process Area. Management changes take place. Now it’s 2002 and here we go again. Still collecting the same core data as in 1992 and as recommended by the SEI: time, effort, size and defects (Ref. 1). Our measurement techniques are unchanged- measure development process productivity separate from size and time pressure. Measure time pressure. Compare projects against industry reference measures determined from current Telecom projects worldwide. Now, however, the reference measures are based on 2002 values since these are updated about every two years from our industry database. What do we find? Well the process productivity has indeed improved over the 10 years but the time to market pressure has also increased. The time pressure means that today relatively more people are used to develop software more quickly with a significant increase in effort and defects. The commercial results are very interesting as we compare the 10-year differences in process productivity and the consequence of time pressure. Lets look in detail at the benchmarking measures used, the 1992 and 2002 findings and then use the results to show how the bottom line works out. Our final conclusions highlight the practicality and benefits of continuous benchmarking. Keep in mind that benchmarking their suppliers is of real interest to purchasers of external software developments (Ref. 5).
Copyright QSM Ltd. Page 2/9
Telecom Software Benchmarks: 10 Years Apart 1992-2002
The Benchmark Measures: Industry Trend-Lines, Process Productivity and Time Pressure Evidence from thousands of completed developments covering all software application types is shown below in Figure 1. Here we plot the time, effort and defects for each project against the size making the scales logarithmic. Mathematical "least squares best fit" analysis of the data provides "trend lines" that show the time, effort and defects are related to size in a power 2 (exponential) form. (Mathematically we find the correlation coefficient (r ) is significant in each case.. Specific “trend-lines” are derived for each type of software application for instance Telecom, process control or business systems. Further mathematical analysis Development Schedule, Effort, Defects vs. Sizelinks effort, time, size and process productivity in a Schedule vs. Size Effort vs. Size 1000 100000Completesoftware equation. QSM Mixed Application Database QSM Mixed Application Database 10000 1000details of the evidence from 100 10 100 many thousands of projects 10 1 1 and the formulation of the 0.1 0.1 1 10 100 1000 1 10 100 1000 Effective SLOC (thousands) Effective SLOC (thousands) Defects vs. Sizeequation are set out in Ref. 2. QSM Mixed Application Database 100000 10000 1000Note : All 100Fortunately the math is Scales are 10 1capable of being understood Log-Log 0.1 and used at all levels by 1 10 100 1000 Effective SLOC (thousands) All Systems All Systems All Systems expressing the three key Figure 1: Time, Effort and Defects Versus “drivers” in straightforward Software Size(Source Lines of Code SLOC) management terms. Driver 1: Size:Software size quantifies the amount of software function developed. Any concrete measure of software size applies, such as logical input statements, objects and function points. The evidence is that re-use does indeed give benefits. It results in smaller size. However specifying more features may increase size and costs significantly more in terms of time, effort and defects. Driver 2: Process Productivity:Team process productivity is derived mathematically (Ref. 3) and quantified in terms of a process productivity parameter (P). The discrete values of this parameter are expressed using a linear management scale going from 1 to 40 termed the Process Productivity Index or PI. This management scale represents quantified step values. The higher the PI then the higher is the project team process productivity. Here the values apply inversely (namely divide) and improving team process productivity significantly reduces development time and effort.
Copyright QSM Ltd. Page 3/9
Telecom Software Benchmarks: 10 Years Apart 1992-2002
Reference PI values are determined from the industry database for each application type based on recent developments. The values provide a benchmark reference for comparison. Currently for Telecom developments, 2002, a mean PI value of 13 is found. The PI value is a high-level measure that benchmarks the entire development environment. It is separate from the size and time pressure. Measuring the PI continuously reveals process improvement as we show for this development group over the 10 years between 1992 and 2002. Driver 3: Time:Development time is the third driver. What is not self evident is how powerful is the time driver. Development effort, cost and defects are highly sensitive to how long you allow for development. Cutting time to market demands means enormous increase in effort, cost and defects. Conversely planning a little more time, say 4 to 6 weeks, yields large reductions in all three. Time pressure is expressed mathematically as a gradient value in the software equation. A simplified scale from 1 to 10 is used termed the Manpower Build-up Index (MBI) where higher numbers mean more time pressure. Full details are set out in Ref 2. The 1992 and 2002 Developments Versus The 2002 Industry Trend Lines The diagram below, Figure 2, shows the trend lines determined from recent Telecom projects. Each benchmarked project is shown plotted in terms of the time and effort versus the size. The 1992 projects (blue squares) generally take longer times while their effort Trendline Phase 3 Main Build : Time Effortis about average. In these projects the Ma in Build Dura tion (Months) vs. Size evidence is of 100 1992 Proje cts extended Longe r Tim e development time. 10 2002 Proje cts Le s s Tim e 1The recent 2002 110100100010000 project data shows ESLOC (thousands) that less time is being Main Build Effort (Person Months) vs. Size taken while effort 10000 remains around the average for the size. 1000 2002 Proje cts Note that scales are 1992 Proje cts Ave r age Effort Ave rage Effort But Longe r Tim e log-log so the shorter 100 times mean 110100100010000 ESLOC (thousands) significantly increased 1992 Benchmark2002 BenchmarkQSM 2002 TelecomAv g. Line Sty le1 Sigma Line Sty le time pressure. Figure 2: 1992 and 2002 Projects versus the 2002 Telecom Trend Lines: Time: Effort Versus Size
Copyright QSM Ltd. Page 4/9
Telecom Software Benchmarks: 10 Years Apart 1992-2002
1992 and 2002 Process Productivity (PI) and Time Pressure (MBI) Benchmark Findings The core measures are input to the software equation to calculate the team process productivity and the time pressure for each project. The average values determined for the 1992 and 2002 benchmark projects are shown in Figure 3. Average Value of Me tricsFor the process productivity the 1992 Ave rage14.0benchmark shows Proce s s Productivity that the 1992 projects PI = 14 PI achieved an average 2002 Ave rage 16.3 Proce s s Productivity PI of 14. Ten years PI = 16 later the average PI has improved to just 1992 Ave rage Tim e Pre s s ure over 16. In addition 1.8M BI ~ 2 Longe r Tim e the 2002 PI value is M BI above the expected 2002 Ave rage Tim e Pre s s ure 4.4 M BI ~ 4 Telecom PI value, Tim e Re duce d currently 13. 024681012141618 AverageMeanwhile the time 1992 Benchmark2002 Benchmark pressure measure (MBI) has increased Figure 3: 1992 and 2002 Projects: Average from an average of Process Productivity (PI) and Time Pressure (MBI) around 2 in 1992 to 4 in the 2002 projects. This reflects the reduced time to market to deliver the Telecom products. The Benchmark Differences between 1992 and 2002How do we interpret the differences in the benchmark findings over the 10 years and what is their significance? First there is clear evidence of improved process productivity. The average PI value has improved from 14 to 16 over the 10 years. In part this is due to the company following a vigorous implementation of the CMMI Key Process Areas (KPA’s). We find that management factors lead to improved process productivity. The KPA’s focus on essential software management techniques and the quantified results below show action on these bring substantial commercial benefits. Second the benchmark of the current process productivity PI of 16 is above the 2002 Telecom mean value of 13. Hence the development group is performing well above current industry norms. Again the commercial consequences are significant and enable senior managers to demonstrate the value for money their organisation provides.
Copyright QSM Ltd. Page 5/9
Telecom Software Benchmarks: 10 Years Apart 1992-2002
Third time pressure is evident due to getting products to market quicker. This costs substantially more money and results in more defects. Modelling an average project shows the impact of increased time pressure (MBI 4) in 2002 compared to the more protracted schedules back in 1992 (MBI 2). Quantifying the Bottom Line Impact Over the 10 Years For the purpose of calculating the commercial benefits we use the average size of the 2002 developments which is 120,000 statements consisting mainly of C and C++. There are clear benefits due to the improved process productivity. These are shown next. However the cost of reduced development time needs to be understood as we show below. Process Improvement BenefitsIn Figure 4 we illustrate the benefits of the process improvement based on the average size with the time pressure found in 2002 (MBI 4). Quantified Process ImprovementThe benchmark measurement Average Size = 120,000 statements MBI = 4 results over the 10 years show that: Monthly Avg Staff (people) <Benchmark PI 14 MBI 4> 8 10 15 16 17 60 5054 staffPI 1992 Effort 600time is1. Development 40 14 Benchmark30 person months 20reduced by 3 months 2. Total development effort has 10 0Time 15 months 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 JanFebMarAprMayJunJluAugSepOtcNovDecJanFebMarAprMayJun 0' 3 '04 Porejtc:eNDBenefitsfallen from 600 person MonthlyAvgStaff(people) months to 330, a saving of <Benchmark PI 16 MBI 4> 8 10 15 16 17 40 PI 20023036 staff 270 person months Benchmark1620 Effort 330 personstaffing has3. Maximum months dropped from 54 to 36 staff. 10 JanFebMarAprMayJunJulAugSepOctNovDecJanFebMraTime 12 months 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 '03 '04 Proejtc:eNDBenfetis Assuming a labour rate of $100,000 Figure 4: Quantifying the Process per person year this represents a Productivity Improvement saving of over $2,000,000 per 1992 versus 2002. project. Equally interesting is to compare the group’s average development against the 2002 Telecom industry PI value of 13. (Incidentally the Telecom average MBI in 2002 is also 4). Here we find that at a PI of 13 and MBI of 4 to develop 120,000 statements the industry 2002 average gives: Development Time =17 months Development Effort =830 person months Peak staff = 66 people
Copyright QSM Ltd. Page 6/9
Telecom Software Benchmarks: 10 Years Apart 1992-2002
So compared to the 2002 Telecom industry average the benchmark group takes 5 months less and uses 500 person months less effort. This represents a saving of over $4,000,000 and demonstrates the value for money provided by this development organisation. Time Pressure Consequences Figure 5 shows the impact of time pressure. Today’s shorter time for developments is measured by an MBI of 4. This is compared with the 1992-benchmark findings an MBI of 2. 2002 Time To Market ImpactDue to the reduced time we see Average Size = 120,000 statements PI = 16 that the impact in 2002 compared Monthly Avg Staff (people) <Benchmark PI 16 MBI 2> 7 8 10 15 16 17 1 .5 0 1 .2 513 staffto 1992 is that an average Longer Effort 150development: 1 .0 0 Time .7 5 person months 5.0Takes 3 months less time MBI 2 .2 5 Costs 180 person months Time 15 months .0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 JanFebMraArpMayJunJluAugSepOtcNovDecJanFebMar Proje:tceNBDenestifmore effort '030'4 Monthly Avg Staff (people) <Benchmark PI 16 MBI 4> 8 10 15 16 17 40 Sorter36 staff If the commercial pressures did 30 Time Effort 330 personallow the longer time of 3 months 20 MBI 410 months (MBI 2) then a further potential 0 A rp May Jun Jul Aug Sep Oct N ov Dec Jan Feb MarTime 12 months saving of $1,500,000 is practical. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Jan Feb Mar 0' 3 0' 4 Porejtc:eNDBentfeis Figure 5: The Impact of Time Here we see that marketing Pressure: 1992 MBI 2 versus 2002 decisions play a significant role in MBI 4. determining development costs. Conclusions The benchmark results shown here use core data from completed projects: development time, effort and size as well as defects. This basic data is quick to collect and allows benchmarking to be done within 4 to 6 weeks. The results reveal how a development group compares with up to date industry reference measures (Trend Lines, Process Productivity and Time Pressure (MBI)). Compared with the expected 2002 Telecom average values this group is performing exceptionally well and enjoys substantial benefits as a result. In addition it is unique in that the original benchmark performed in 1992 allows an evaluation and quantification of process productivity improvement over the 10 years. Again the results demonstrate the large commercial benefits gained over the last decade. The customers for the products of these development groups are also motivated to benchmark their suppliers in the same way (Ref. 3/5). The increasing costs and risks in the purchasing and the outsourcing of external developments mean that benchmarking leads to informed negotiations with suppliers. Purchasers are able to evaluate supplier value for money as well as checking new proposals are consistent with the supplier capability.
Copyright QSM Ltd. Page 7/9
Telecom Software Benchmarks: 10 Years Apart 1992-2002
The benchmark case study demonstrates that it is practical to continuously collect data and quantify process improvement. Note this is independent of size and time pressure in each development. Companies are able to calculate the return on investment through process improvement initiatives such as CMMI. In the case study we lack the details of the investments over the period and the continuous data collection and measurement on all projects. We find that with initiatives such as CMMI there is increasing pressure to cost justify the investments being made to move up the maturity levels (Ref. 4). There is an equal need to judge if the process improvement changes are really bringing benefits. The results shown here enable the return on investments to be calculated as well as confirm real improvements are made. In addition the data and its use provides the means to satisfy in part the Measurement and Analysis KPA at Level 2 of CMMI. Finally is this benchmarking and ROI calculation capability new? Well no. One of the first instances where we benchmarked over a number of years is set out in detail under Ref. 2. Here the developments are from a large business system group. This work involved evaluating the application development environment and recommending actions for process improvement. Chapter 12 “Managing a Productivity Program” sets out our findings, recommendations and results including full details of the ROI calculation. Over three years the data shows that the ROI is around 77% based on investments of approximately $23,000,000. Our improvement recommendations at that time (1985) are set out that focussed on software management practices. It will be seen that these recommendations are in line with CMMI Key Process Areas. Jim Greene is Managing Director of Quantitative Software Management Europe in Paris, France: telephone 33-140431210; fax 33-148286249. He has over 30 years experience in software engineering, with a particular interest in management methods used by development and purchasing organisations based on the quantification of software development. Ref.1: The SEI Core Measures Anita D. Carleton, Robert E. Park and Wolfart B. Goethert The Journal of the Quality Assurance Institute July 1994 Ref. 2: L.H. Putnam and Ware Myers “Measures For Excellence: Reliable Software, On Time, Within Budget:” Prentice Hall New York 1992 Ref. 3: Geerhard W. Kempff “Managing Software Acquisition” Managing System Development July 1998 Applied Computer Research Inc. P.O. Box 82266, Phoenix, AZ, USA. Ref. 4: Lawrence H. Putnam “The Economic Value of Moving up the SEI ScaleManaging System DevelopmentJuly 1994 Applied Computer Research Inc. P.O. Box 82266, Phoenix, AZ, USA Ref. 5: Lawrence H. Putnam and Ware Myers “Control the Software Beast With Metrics Based Management” CROSSTALK August 2002 The Journal of Defense Software Engineering USAF Hill Air Force Base.
Copyright QSM Ltd. Page 8/9
Telecom Software Benchmarks: 10 Years Apart 1992-2002
For further information on QSM’s practices, refer to Lawrence H. Putnam and Ware Myers, Industrial Strength Software: Effective Management Using Measurement, IEEE Computer Society Press, Los Alamitos, CA, 1997, 309 pp.
Copyright QSM Ltd. Page 9/9
