软件测试估算的最佳方法？

其实下面的文章并没有介绍“最佳”，所有它的标题是打问号的。因为其实也没有最佳。因为这个是根据企业的背景，做的项目属性，人员的水平等诸多因素而定的。就我所用到的，也是很多的结合。比如新的项目来得时候，可以根据Test Work Breakdown Approach来break down，然后根据以往的相似project的经验，应用Metrics-Based Approach来估算小的work item。Iterative Approach其实是经常用的，因为项目的change是很平常的事情。好了，大家还是读一下下面的文章，看看你在用哪些方法？

－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－

What's the best approach to software test estimation?

There is no simple answer for this. The 'best approach' is highly dependent on the particular organization and project and the experience of the personnel involved.

For example, given two software projects of similar complexity and size, the appropriate test effort for one project might be very large if it was for life-critical medical equipment software, but might be much smaller for the other project if it was for a low-cost computer game. A test estimation approach that only considered size and complexity might be appropriate for one project but not for the other.

Following are some approaches to consider.

Implicit Risk Context Approach:
A typical approach to test estimation is for a project manager or QA manager to implicitly use risk context, in combination with past personal experiences in the organization, to choose a level of resources to allocate to testing. In many organizations, the 'risk context' is assumed to be similar from one project to the next, so there is no explicit consideration of risk context. (Risk context might include factors such as the organization's typical software quality levels, the software's intended use, the experience level of developers and testers, etc.) This is essentially an intuitive guess based on experience.

Metrics-Based Approach:
A useful approach is to track past experience of an organization's various projects and the associated test effort that worked well for projects. Once there is a set of data covering characteristics for a reasonable number of projects, then this 'past experience' information can be used for future test project planning. (Determining and collecting useful project metrics over time can be an extremely difficult task.) For each particular new project, the 'expected' required test time can be adjusted based on whatever metrics or other information is available, such as function point count, number of external system interfaces, unit testing done by developers, risk levels of the project, etc. In the end, this is essentially 'judgement based on documented experience', and is not easy to do successfully.

Test Work Breakdown Approach:
Another common approach is to decompose the expected testing tasks into a collection of small tasks for which estimates can, at least in theory, be made with reasonable accuracy. This of course assumes that an accurate and predictable breakdown of testing tasks and their estimated effort is feasible. In many large projects, this is not the case. For example, if a large number of bugs are being found in a project, this will add to the time required for testing, retesting, bug analysis and reporting. It will also add to the time required for development, and if development schedules and efforts do not go as planned, this will further impact testing.

Iterative Approach:
In this approach for large test efforts, an initial rough testing estimate is made. Once testing begins, a more refined estimate is made after a small percentage (eg, 1%) of the first estimate's work is done. At this point testers have obtained additional test project knowledge and a better understanding of issues, general software quality, and risk. Test plans and schedules can be refactored if necessary and a new estimate provided. Then a yet-more-refined estimate is made after a somewhat larger percentage (eg, 2%) of the new work estimate is done. Repeat the cycle as necessary/appropriate.

Percentage-of-Development Approach:
Some organizations utilize a quick estimation method for testing based on the estimated programming effort. For example, if a project is estimated to require 1000 hours of programming effort, and the organization normally finds that a 40% ratio for testing is appropriate, then an estimate of 400 hours for testing would be used. This approach may or may not be useful depending on the project-to-project variations in risk, personnel, types of applications, levels of complexity, etc.

Successful test estimation is a challenge for most organizations, since few can accurately estimate software project development efforts, much less the testing effort of a project. It is also difficult to attempt testing estimates without first having detailed information about a project, including detailed requirements, the organization's experience with similar projects in the past, and an understanding of what should be included in a 'testing' estimation for a project (functional testing? unit testing? reviews? inspections? load testing? security testing?)

With agile software development approaches, test effort estimations may be unnecessary if pure test-driven development is utilized. In general, agile-based projects by their nature will not be heavily dependent on large testing efforts, since they emphasize the construction of releasable software in very short iteration cycles. Therefore test effort estimates may not be as difficult and the impact of inaccurate estimates will be minimized.

For an interesting view of the problem of test estimation, see the comments on Martin Fowler's web site indicating that, for many large systems, "testing and debugging is impossible to schedule".