关于分工的思考 (Thoughts on Division of Labor)
Did you ever have the feeling that adding people doesn't help in software development? Did you ever think about the reason? And do you have any idea to make a change?
Traditional software engineering emphasizes on division of work by modules, values cooperation between developers. This looks reasonable at first glance, because it is what we are used to do in the production line of traditional industry. But software development is not a production process, this is why we call it software development rather than software production. The closer analogy for software development is the new production development in traditional industry.
The major challenge of software development is in design, which is about idea but labor. The quality of design can't be improved by adding hands. It even hurts sometimes, because more people means diffusion of responsibility and higher communication cost, which is considered one of the major problem in software development. We usually see that to implement a feature, modification can be made in module A or module B, the owners of the modules argue with each other to reject the modification in their own module. Division of work causes people think from their own local point of view, and protect their own interests, however the whole system hurts. Remember that modules can be loosely-coupled, but people have to work closely to make the whole system work. Traditional software engineering emphasizes on improving cooperation, but I think the otherwise, the better way is reducing unnecessary cooperation, in other words, reducing unnecessary division of work.
Instead of adding hands, what really makes sense is finding the intelligent brains. Quality comes from intelligent brains. As we know, most of the master pieces don't come from cooperation, and so is the good design and implementation in software. For a highly-cohesive complex system like Android, it really doesn't need dozens of people and division of work to have it come true, actually only 2 or 3 intelligent people is enough. In which case dozens of people are really needed? It must be many unrelated systems, such as, Bing, Office and MSN in Microsoft. For any highly-cohesive complex system, I didn't see any reason that 2 or 3 excellent developers are not enough. If 3 people can't make it, 100 neither. The point is not in hands, it's in brain.
Besides finding the intelligent brains, provided currently available people configuration, what can we do better? In my development practice, I value the following practices:
1) Collective Ownership. That means we don't assign a module to a specific developer, instead every developer can write code for any module. The ownership of code is equally shared within the team. Everyone Know All is the philosophy behind. I really appreciate this philosophy in software development. Some say it's too hard for every developer to know every module, this saying is not true. Average IQ is enough to handle every module of complex system like Android, I believe it without any doubt. Do you think your system is more complex than Android? The key is your determination.
2) Feature/Issue Ownership. Code ownership is shared, but every feature or issue should has an owner, and we need to track the status regularly. The owner should be able to work on many modules to finish the task or resolve the issue. Compared with module ownership, feature/issue ownership reduces unnecessary communication cost, avoids many argues between developers of different modules. It also makes developers focus on features that really make sense to user, and think from the system level.
3) Pair Programming/Peer Review. Pair Programming/Peer Review is totally different from traditional division of work. It's not division, it's working together. Pair Programming and Peer Review give inspiration to design, and give feedback to implementation, so they really improve the quality a lot. Even more, people grow much faster in this environment.
To sum up, the idea of having many people and division of work by modules is not a good fit for software development. Remember the 2 golden rules for software development: 1) We need brains but hands. 2) Don't divide, integrate!
关于分工的思考 (Thoughts on Division of Labor)的更多相关文章
- terminal(终端),shell,tty,console(控制台)区别
原文地址 stackexchange:What is the exact difference between a 'terminal', a 'shell', a 'tty' and a 'con ...
- agnentX学习存在疑问?
在RFC2741中这样定义: 4.3中有如下段落: A general architectural division of labor between master agent and subage ...
- 第5章 Hyperledger Fabric功能
Hyperledger Fabric is a unique implementation of distributed ledger technology (DLT) that delivers e ...
- 福大软工 · 第八次作业(课堂实战)- 项目UML设计(团队)
Team information 队名: 彳艮彳亍团队 各成员短学号.名: 学号: 姓名: 本次博客链接: 041602209 黄毓明(临时队长) https://www.cnblogs.com/m ...
- 智课雅思词汇---十三、前缀ab-是什么意思
智课雅思词汇---十三.前缀ab-是什么意思 一.总结 一句话总结:分离,脱离;相反;加强意义 前缀:ab- [词根含义]:分离,脱离;相反;加强意义 [词根来源]:来源于拉丁语前缀ab-. [同源单 ...
- OpenMP for Fortran
OpenMP for Fortran OpenMP Directive Syntax of OpenMP compiler directive for Fortran: !$OMP Directive ...
- IOS 7 Study - UIViewController
Presenting and Managing Views with UIViewController ProblemYou want to switch among different views ...
- [HIve - LanguageManual] Join Optimization (不懂)
Join Optimization Join Optimization Improvements to the Hive Optimizer Star Join Optimization Star S ...
- 对PostgreSQL的prepared statement的深入理解
看官方文档: http://www.postgresql.org/docs/current/static/sql-prepare.html PREPARE creates a prepared sta ...
随机推荐
- 【MySQL】 查询某个数据库有多少张数据表
SELECT COUNT(*) TABLES, table_schema FROM information_schema.TABLES WHERE table_schema = 'performanc ...
- EditText获取和失去焦点,软键盘的关闭,和软键盘的显示和隐藏的监听
软键盘显示和隐藏的监听: 注: mReplayRelativeLayout是EditText的父布局 //监听软键盘是否显示或隐藏 mReplayRelativeLayout.getViewTreeO ...
- Vue - class与style绑定
1.通过v-bind绑定一个class A:直接绑定 B:通过v-bind绑定一个对象 C:绑定一个返回对象的计算属性 D:绑定一个数组对象 2.绑定内联样式 A:直接子啊属性名上面绑定 B:绑定到一 ...
- 如何用hypermesh生成包含interface的流体网格
在计算气动声学的时候,有些情况是需要我们提取流体计算的结果作为声学分析的边界条件,但是,有些流体网格因为物理模型的问题需要我们设定interface,恰恰你是机械,对流体了解一点,又不想花费太多时间来 ...
- 【转】oracle中rowid的用法 (全面)
ROWID是数据的详细地址,通过rowid,oracle可以快速的定位某行具体的数据的位置. ROWID可以分为物理rowid和逻辑rowid两种.普通的堆表中的rowid是物理rowid,索引组织表 ...
- php备份数据库
php备份数据库原理和方法 原理 查找所有表 查找所有字段,列出所有字段名 字段类型等信息 查找所有数据 读取后注意特殊符号转换addslashes() 生成sql 把数据库格式化生成对应sql 相关 ...
- 【转】bash 参数展开(Parameter Expansion)
转自:http://www.360doc.com/content/13/0513/20/9437165_285204629.shtml ${parameter} 取parameter的值 ${para ...
- linux 目录权限的特殊之处
目录的读权限,不能进入目录.仅允许我们读目录,获得在该目录中所有文件名的列表,但无法查看目录中文件的内容. 目录的执行权限,可以进入目录,但不允许我们读取目录的文件列表,但可以查看目录中文件的内容.当 ...
- 【WPF】WPF中的List<T>和ObservableCollection<T>
在WPF中 控件绑定数据源时,数据源建议采用 ObservableCollection<T>集合 ObservableCollection<T> 类:表示一个动态数据集合,在添 ...
- Hibernate criteria 增加排序项