| VDC data comparing Linux and Windows embedded development |
by Chris Lanfear and Steve Balacco (Aug. 11, 2003)
VDC Whitepaper:
Comparing Linux and Windows embedded development
by Chris Lanfear and Steve Balacco
Venture Development Corporation
August 11, 2003
Background
Venture Development Corporation (VDC) presents a comparison between embedded designs that have used Windows Embedded and the embedded Linux platforms.
As part of VDC's Embedded Software Strategic Market Intelligence Program, VDC has published market research studies regarding Linux in the embedded systems market in May 2001, April 2002, and most recently in June 2003. The scope of these studies covers the worldwide market for Linux employed in the development of embedded systems and includes the Linux operating system, add-on technology, and related services.
VDC is not affiliated with any industry software suppliers or trade journals, and we accept no advertising. Our goal is to provide the embedded systems community with unbiased, fact-driven research and analysis on the embedded hardware and software space through a time-proven and rigorous research methodology.
Included as part of these reports is an analysis of embedded developer responses. For the April 2002 report, an extensive survey was mailed and emailed to a list of developers who were either current subscribers of the Embedded Linux Journal or members of VDC's embedded developer panel. For the June 2003 report, an extensive survey was e-mailed to a list of developers who were current subscribers of such magazines as EDN, Portable Design, Integration Communications Design (ICD), Embedded System Engineering, and VDC's embedded proprietary developer panel. VDC used SPSS, an industry standard statistical software package, for most of the statistical analysis used to produce our findings. Other analyses were performed using proprietary VDC tools and techniques.
For the purposes of this paper, VDC offers the following comments on the end-user data collected by VDC from the April 2002 and June 2003 reports . . .
Average number of software developers per project
Exhibit 1 illustrates the average number of software developers per project. Comments follow.
Exhibit 1: Average Number of Software Developers Per Project
| Linux | Windows Embedded (CE 3x, CE.NET, NTE, XPE) | Commercial OS (Excluding Linux) | |
April 2002 Report | Mean | 6.3 | 4.2 | 5.4 | "N" | 75 | 27 | 114 | June 2003 Report | Mean | 4.8 | 3.7 | 6.6 | "N" | 44 | 36 | 186 |
Comments on Exhibit 1 . . .
In the April 2002 report . . . - The data shows an average of 6.3 software developers per Linux project with an average of 4.2 per Windows Embedded project, and a total average of 5.4 developers per project for projects using a commercial operating system (excluding Linux). This data excludes data points that are 3 standard deviations from the mean which excluded two data points from Linux, no data points from Windows Embedded, and one data point from all using a commercial OS (excluding Linux). For the use of the commercial OS analysis, VDC has excluded Linux due to its availability in both commercial and non-commercial distributions (i.e. Red Hat). As a result, Linux was excluded from this category since the data does not accurately reflect whether the distribution was acquired commercially or downloaded for free.
- For Linux, the calculated mean for the number of software developers per project of 6.3 is accurate to +/- 2.1 software developers at a 95% confidence interval, with a standard deviation of 9.28 for the sample size of 75.
- For Windows Embedded, the calculated mean for the number of software developers per project of 4.2 is accurate to +/- 2.1 software developers at a 95% confidence interval, with a standard deviation of 5.30 for the sample size of 27.
- For respondents citing use of a commercial OS, the calculated mean for the number of software developers per project of 5.4 is accurate to +/- 1.1 software developers at a 95% confidence interval, with a standard deviation of 5.77 for the sample size of 114.
In the June 2003 report . . . - The data shows an average of 4.8 software developers per Linux project with an average of 3.7 per Windows Embedded project, and a total average of 6.6 developers per project for projects using a commercial operating system (excluding Linux). This data excludes data points that are 3 standard deviations from the mean which excluded one data point from Linux, two data points from Windows Embedded, and five data points from all using a commercial OS (excluding Linux). For the use of the commercial OS analysis, VDC has excluded Linux due to its availability in both commercial and non-commercial distributions (i.e. Red Hat). As a result, Linux was excluded from this category since the data does not accurately reflect whether the distribution was acquired commercially or downloaded for free.
- For Linux, the calculated mean for the number of software developers per project of 4.8 is accurate to +/- 1.3 software developers at a 95% confidence interval, with a standard deviation of 4.30 for the sample size of 44.
- For Windows Embedded, the calculated mean for the number of software developers per project of 3.7 is accurate to +/- .8 software developers at a 95% confidence interval, with a standard deviation of 2.50 for the sample size of 36.
- For respondents citing use of a commercial OS, the calculated mean for the number of software developers per project of 6.6 is accurate to +/- 1.7 software developers at a 95% confidence interval, with a standard deviation of 11.98 for the sample size of 186.
The data shows that the number of software developers per project using Windows Embedded are approximately 22.9% to 33.3% smaller than the number of software developers per project using Linux.
Distinction between "new" vs. "enhancement" designs
Exhibit 2 illustrates whether the development project is a "new design" or an "enhancement to a previous design". Comments follow.
Exhibit 2: Type of Project
| Linux | Windows Embedded | |
April 2002 Report | | N = 74 | N = 28 | New Design | 65.1% | 75.0% | Enhancement of a Previous Design | 34.9% | 25.0% | June 2003 Report | | N = 47 | N = 38 | New Design | 78.7% | 65.8% | Enhancement of a Previous Design | 21.3% | 34.2% |
Comments on Exhibit 2 . . . - The type of project can have an effect on the length of time to complete the project where new design projects might have little or no legacy code involved and include a more complex design phase.
- As these operating system platforms continue to evolve and mature, it should be expected that families and iterations of products would be developed as enhancements to previous designs.
- The data shows (from the June 2003 report) that Linux in embedded projects continues to expand and evolve, particularly in new design projects.
Time-to-Market data
Exhibits 3 and 4 illustrate the project "Time-to-Market" or the number of months between the design/specification stages to the time of first shipment. Data collected through the VDC survey for this question is based on ranges, not specific whole numbers; therefore we are unable to offer a total time-to-market in terms of months.
Exhibits 3A and 3B represent data from the April 2002 Report, and Exhibits 4A and 4B represent data from the most recent report, in June 2003. Comments follow.
Exhibit 3A: Time to Market April 2002 Report -- Linux
Exhibit 3B: Time to Market April 2002 Report -- Windows Embedded
Comments on Exhibits 3A and 3B . . .
Exhibits 3A and 3B reflect the time-to-market data from the April 2002 report for Linux and Windows Embedded projects: - The distribution for both platforms for projects completed in the 1 to 6 month range is similar for both environments.
- For projects completed within one year, the data shows 54.8% for Linux and 42.9% for Windows Embedded, which VDC expects is influenced by a higher percentage of projects for Windows Embedded to be "new designs" compared to Linux.
Exhibit 4A: Time to Market June 2003 Report -- Linux
Exhibit 4B: Time to Market June 2003 Report -- Windows Embedded
Comments on Exhibits 4A and 4B . . .
Exhibits 4A and 4B reflect the time-to-market data from the June 2003 report for Linux and Windows Embedded projects: - In the 1 to 6 month range, the data shows a greater percentage of projects completed for Windows Embedded, 28.9%, in comparison to 17.0% for Linux.
- For projects completed within one year, the data shows 63.2% for Windows Embedded and 55.3% for Linux.
- Further analysis of this shift year to year correlates to an increasing number of projects cited as "new designs" for Linux over Windows Embedded where developers cite the increasing use of this platform as an "enhancement to a previous design."
Conclusion
This is a very complex subject and requires due diligence on the part of OEMs to conduct an integrated assessment that includes but is not limited to technology, business models, supportability, terms and conditions, vendor experience, and schedules. The bottom line for OEMs is to produce quality and innovative products, on time and on budget.
There is no typical embedded project, therefore it is often difficult to compare the experiences of two or more design teams from any number of companies.
Regardless of a platform selection, OEMs must get their arms around the design and specification stages of the development cycle to minimize mid-project design changes which can cause project costs and time-to-market to spiral out of control and ROI to turn red. Project schedules must be realistic given the technology and resources available, and continuous management oversight of the entire development cycle is critical overall to minimize risks in determining success or failure, profit or loss.
Copyright © 2003, Venture Development Corp. (VDC). All rights reserved. Reproduced by WindowsForDevices.com with permission.
(Click here for further information)
|
|
|
|
|
|
|
news feed