[ZT] 6 Sigma Case Analysis of Software Development

Source: sawin Author: Ru Haiyan

One side is the announcement of a consulting company in project management training: "6 Sigma is not suitable for a level-2 Enterprise of CMMs. It should wait until the level-4 CMMS is complete and the measurement system is complete ." One side is 2004
At the World Software Engineering Conference, experts from various countries reached a consensus: "CMM/cmme can be combined with 6 SIGMA to promote each other ". What should we do? My previous claim: the dispute is put aside for the moment, so you can hurry and implement it.
Practice and improvement, otherwise the result is likely to be: "We are making progress, but we are more backward than competitors ." Some colleagues accepted my opinion and asked, "Are you using the software ?"
6 Sigma success stories ?" I didn't have it six months ago, but now I have several typical cases. They have their own characteristics. Let's share them here one by one.

1. Can 6 Sigma help solve software technical problems?

The first project was announced at the end of last year at a 6 Sigma excellent Project Conference of a business department. The project name is "XX network management system to improve alert throughput". The problem is that when a large number of alerts are reported,
The alarm processing throughput of Unix servers is only 8 messages/second, And the CPU usage reaches 90%. As a result, operations on other modules are basically not allowed. The user is very dissatisfied with this and asked our company to solve this problem as soon as possible.
Increase the throughput to at least 48 messages per second, but the system cost cannot be greatly increased. How can this problem be solved? One solution is to improve the hardware configuration and processing performance, but this will greatly
This solution is rejected because procurement costs are increased, while performance does not greatly improve. In fact, the scalability of products is reduced. The project team spent a lot of time and energy searching
After a proper solution, I thought of Six Sigma. As you know, the 6 Sigma project is chosen for the "difficult and influential" problems, so the members of this project team set up the project,
6 Sigma is expected to bring dawn in the dark.

Apart from the definition and measurement stages, the analysis of this project is like this: brainstorm a fish bone map to list all possible causes; then, organize the reasons according to the alarm logic processing process
FMEA is used for RPN analysis, and a few factors whose RPN value is greater than 100 are screened as potential key factors. Then, these potential factors are tested and confirmed one by one. The breakthrough of the entire project appears in
In a factor test, as shown in test data 1, the abscissa represents the input alarm traffic, and the ordinate represents the delay of alarm processing. The curve shown in the figure shows a periodic inflection point, and after the inflection point, the alarm traffic increases.
In addition, the processing latency of the server is greatly reduced. This phenomenon cannot be seen without such data if there is no test for this reason. It is difficult for our software engineers to analyze the cause of this phenomenon.
The conclusion is that TCP Parameters are improperly set. After this parameter is modified, perform the same test again. expected data 2 shows that the alarm throughput basically increases linearly with the input traffic, resulting in bottlenecks.
Deleted. This not only confirms that this factor is a key factor, but also verifies the effectiveness of improvement measures. Several other factors are also similar. by testing each suspicious factor, or confirming that this factor is critical
Key factor, or filter the factors that have little influence. Then, find technical solutions for each key factor. The success of this project has created 1.66 million annual earnings for the company.

Looking back at this project, I also came up with the old saying: "99% of the effort to solve the problem is often to find the key reason, and only 1% of the effort is needed to modify it ." 6sigma itself does not provide technical solutions
But its ideas guide us in the right direction, and data is an important basis to ensure our direction is correct. Although this project is a software project, the problem itself y is clear, which is also
One reason for success is that we can adapt to the six sigma feature.

Figure 1 Traffic test data of alarm processing for key factor 1 in a project

Figure 2 alarm handling throughput after protocol parameters are modified for a project

2. Are the results of subjective judgment persuasive?

In this case, the Black Belt Project "reduces exceptionsCodeFailure rate, which is one of the main fault types analyzed by CQ: the failure rate of abnormal code remains high, which reflects the spirit of the person in charge to actively learn and make progress from mistakes, it also reminds many colleagues who still cannot find a suitable project that the CQ library is a convenient project.

This project is a typical measurement system analysis of discrete data for fault classification. During the R & D process, we often encounter situations where we can only talk about what we mean. We all use subjective judgment to make decisions.
Bag ", how is this analysis convincing? Subjective judgment is not equal to shoot your head. This project can be used as a reference. By setting certain standards, you can make subjective judgments consistent and accurate.
Standard. The following is a summary from the project owner.Article:

After the fault classification rules are determined, the fault classification may result in different results for different R & D personnel of the same fault. There may be two reasons for this problem: (1) fault classification criteria
The accuracy is not clear enough. The R & D personnel involved in fault classification have different understandings about faults. (2) R & D personnel involved in fault classification are not able to classify faults according to classification criteria. The solution is to do so before fault classification.
Measure the system analysis to confirm whether the fault classification criteria are clear and whether the R & D personnel involved in fault classification have the fault classification capability.
For the measurement system analysis of fault classification, discrete measurement system analysis can be used. The basic steps for discrete measurement system analysis are as follows:

1. 30 faults are randomly selected from the fault to be analyzed.
2. Multiple development managers jointly determine the classification results of each fault according to the fault classification criteria. (As "True Value"-author of this article)
3. Enable the R & D personnel involved in fault classification to be classified according to the fault classification criteria. (As "measured value"-author of this article)
4. One week later, ask the R & D personnel involved in fault classification to reclassify the fault.
5. Conduct discrete measurement system analysis to determine the accuracy, repeatability and reproducibility of fault classification.

If a measurement system analysis shows that the fault classification is repetitive and the same R & D personnel has different judgment results for the same fault, it is generally a problem of the quality of the R & D personnel, the measures taken are required
Enhance the learning of classification standards. If there is a problem with the reproducibility between different R & D personnel, it is generally because the classification standards are not clear, and the classification standards need to be further clarified. If the repeatability and reproducibility meet the requirements
The standard definition of the fault classification is clear, and the skills of the R & D personnel involved in the fault classification have already met the requirements.
In fact, this method of measurement system analysis is not the first time used. The Green Belt project we studied last year is very similar, as shown in principle 3.

 
Figure 3 schematic diagram of the Measurement System Analysis of a Green Belt Project

This is a typical discrete data MSA case. During the presentation, many developers were surprised: "The original MSA can do this ", or "quantitative data can be obtained in this way ". There are many
Because data quantification is insufficient in the R & D process, it is not suitable for 6 Sigma projects. In fact, not 6 Sigma cannot be used in the R & D field, but we often fail to find the correct method. So
If you think more and do more, you can use it at work as if you taught us from an early age.

3. How to improve execution?

I remember a recent leader said: "Our company never lacks good planning. What we lack is good execution ." Some Problems in the software are execution problems, such as non-strict compliance and unreasonable processes. Someone
Q: "if a solution is a problem of execution, you can choose reasonable suggestions based on its influence or a team project. It is not suitable for 6 Sigma projects ." In fact, the first six sigma item
At the beginning, I didn't know what the key factor was. Second, the execution was not simple. I knew and did the same thing, as a hacker said: "When you hear it, you will forget it. When you see it, you will remember it, you will understand it."

To put it bluntly, this case is another Black Belt Project, "Improving the efficiency of software R & D for function Testers". The R & D efficiency is defined as the labor cost of software development and maintenance per unit. The first feature of this project is very fine
Therefore, each step is completely and clearly described according to the six sigma idea and method. It can be used as a sample guide for beginners. However, for me, its more important feature is its powerful execution. At analysis level
, Three key factors have been identified:

1. Low Degree of modularization;
2. Interface documentation is not standard;
3. The communication mechanisms of the system, software, and hardware departments are not sound;

To solve the first problem, this project proposes the goal of creating 10 modules, but currently it only has three modules. To solve the second problem, we need to create an interface document template, but more importantly, get the user's
The third problem is the approval and operation of the senior executives of the three departments-ministers who personally communicate and coordinate the solution. All of the above are done in this project! How can this problem be achieved? Look at his team members, there is a research institute
Director, General Manager of a product, ministers of three relevant departments, and all related section chiefs, development managers, and some developers, are you still worried that the solution will not be executed?

The case is so simple that every project owner needs to carefully select your team members and strive to make everyone do their best. Various personnel are required in the team: first, customer representatives from all parties, then analysts and experts
Author, Coordinator, process supervisor or organization leader. I remember that there was a project in the past, and the members were basically project managers. This kind of team communication was smooth, but when talking about specific things, everyone
There is no time to do it. How is the project progressing? Only cancel. Think about the number of projects that fall down before they reach the end. Few projects cannot find a solution, but the solutions made cannot be sold to the users.
To the project goal, which is common. Why? Most of the reasons are because there are no representatives of the users in the team, and none of them like what they impose.

Some people may say that when it comes to execution, we need to pull the leaders in, get the leaders done, and let the leaders come forward to promote execution and tracking. That's all right. This is not correct, because we cannot solve the problem of the leadership, but it will be
Leaders can solve this problem: when selecting a project, they must grasp the issues that the leaders are most concerned about. Even if they are not most concerned about, they should also be at the forefront of their problem list. If you want to solve this problem
Ask him to join the project team and ask him to follow up. However, if the selected project is not in the forefront of the leadership issue list, he will be distracted, at the same time, it consumes Resources
He naturally refused to do things. This is why we emphasize top-down projects. At present, our company is still a human governance society to a certain extent. We acknowledge this reality and take the initiative to adjust our practices to adapt to the present situation.
Is a wise way to do things.

The above is a typical and successful software 6 Sigma project I recently collected. However, I have to admit that among the thousands of projects that the company has completed, software projects still account for a very small proportion. If it is because
If there are no successful cases that affect your confidence, or you do not know how to start, I hope this article will provide you with some reference. The combination and mutual promotion between CMM/cmme and 6 Sigma are
The field is a new topic and there is still much room for expansion. At present, our company also has many black bands in EPG, and the selected project is the new topic. One year later, let's review it again, hoping to make a bigger breakthrough,
Let us also leave a glorious stroke in the history of 6 Sigma.

References:
Making failure a successful mother, Xu jinchang, 2004-3