|
Many process models I came across so far have their background in a certain engineering setup which I call "the happy engineering world".
I called it this way, because in this "world" you design systems where requirements can be set up from the beginning, sub-components
may be even available to a large extend and very often resources are not strictly limited. As an example may serve an airport traffic control
system. They do not need to design their radar systems and mainframe computers along side the actual airport control system project. They can rely on existing
components and concentrate on the collection and analysis of the requirements for the actual control software, then go to the design phase, then
to the coding. The tests will be strictly a test of the initial requirements. This all sounds pretty much like following on of the process models e.g. the V-model I showed below. However, there is also a quite different world and this is mostly forgotten by the preachers of process models. Let me give you as an example one of my past projects. I was responsible for the software development for a new continuous variable transmision for a vehicle. When I say new I mean new. The transmission itself was a brandnew development using brandnew technologies and we simply did not know how it would behave. Of course a good number of requirements were known to us. We knew how the output currents had to look like to control the transmission, and a few other things. Looking back I could say it was roughly 60% of the requirements we knew for the first sample. After that the mechanical department changed and imprroved the transmission several times. Many requirements remained the same, some where changed, others were added and we had to change the software several times. We had four or five different samples plus a good number of intermediate changes on existing samples. You would call it a typical prototyping? No, it wasn't because the samples were not thrown away and after that the series development started. We simply went into production with the last sample, its mechanics and its software. This kind of development is quite common in many industries. The answer of the process engineers was so far to cascade V-Cycles to express the different sample phases, but this does not give the true picture either. Of course we did a kind of requirements engineering and we wrote down the most important design decisions in the early sample phases, but only as much as was necessary to communicate among the team members. This was far from satifying the ideas of process assessors. But doing more would have been an absolute waste of time, because everybody knew that it was most likely superseeded by the next sample. So what is the solution to it? As I said the process model we used was a cascaded V-cycle model with the possibility to tailor the completeness and quality of the work products and process steps of the earlier V-cycles, requiring a complete and good last V-cycle. The spiral model would have been a solution too, however there are these nasty risk analysis phases which we did not want to do and which made not much sense in our situation. Quite frankly there is no ideal process model for some develoment work. What ever you use, it will always need some adaptation and tailoring. Therefore I just want to give a short overview on the most common process models. Which ever you may choose for your situation, be aware that it will never match your situation completely. The Waterfall Model The waterfall model is believed to have been the first process model which was introduced and widely followed in software engineering. The innovation was that the first time software engineering was devided into separate phases. When I did my first programs in PL/1 and RPG in the early 1970's there was no awareness of splitting up software development into different phases. Programs were very small, the requirements only a few and after punching a pile of cards the program was done and could be tested by inserting it into the card reader and observing what it did. As programs became bigger the need for a better requirements phase, some more thoughts on the design, etc. were needed. Programmers found it more and more difficult to keep an abstract of the program in their mind and transfer it into code. Also the thought of having a separate testing phase performed by dedicated testers evolved. The different phases of software engineering were identified and simply cascaded in each other, allowing for loops in case it was found in a subsequent phase that the previous phase was not done properly.  The phases of "The Waterfall Model" are: Requirement Analysis & Definition: All requirements of the system which has to be developed are collected in this step. Like in other process models requirements are split up in functional requiremetns and constraints which the system has to fulfill. Requirements have to be collected by analyzing the needs of the end user(s) and checking them for validity and the possibility to implement them. The aim is to generate a Requirements Specification Document which is used as an input for the next phase of the model. System Design: The system has to be properly designed before any implementation is started. This involves an architectural design which defines and describes the main blocks and components of the system, their interfaces and interactions. By this the needed hardware is defined and the software is split up in its components. E.g. this involves the definition or selection of a computer platform, an operating system, other peripheral hardware, etc. The software components have to be defined to meet the end user requirements and to meet the need of possible scalability of the system. The aim of this phase is to generate a System Architecture Document this serves as an input for the software design phase of the development, but also as an input for hardware design or selection activities. Usually in this phase various documents are generated, one for each discipline, so that the software usually will receive a software architecture document. Software Design: Based on the system architecture which defines the main software blocks the software design will break them further down into code modules. The interfaces and interactions of the modules are described, as well as their functional contents. All necessary system states like startup, shutdown, error conditions and diagnostic modes have to be considered and the activity and behaviour of the software has to be defined. The output of this phase is a Software Design Document which is the base of the following implemetnation work. Coding: Based on the software design document the work is aiming to set up the defined modules or units and actual coding is started. The system is first developed in smaller portions called units. They are able to stand alone from an functional aspect and are integrated later on to form the complete software package. Software Integration & Verification: Each unit is developed independently and can be tested for its functionality. This is the so called Unit Testing. It simply verifies if the modules or units to check if they meet their specifications. This involves functional tests at the interfaces of the modules, but also more detailed tests which consider the inner structure of the software modules. During integration the units which are developed and tested for their functionalities are brought together. The modules are integrated into a complete system and tested to check if all modules cooperate as expected. System Verification: After successfully integration including the related tests the complete system has to be tested against its initial requirements. This will include the orignal hardware and environment, whereas the previous integration and testing phase may still be performed in a different environment or on a test bench. Operation & Maintenance: The system is handed over to the customer and will be used the first time by him. Naturally the customer will check if his requirements were implemetned as expected but he will also validate if the correct requirements have been set up in the beginning. In case there are changes necessary it has to be fixed to make the system usable or to make it comply to the customer wishes. In most of the "Waterfall Model" descriptions this phase is extended to a never ending phase of "Operations & Maintenance". All the problems which did not arise during the previous phases will be solved in this last phase. The weekness of the Waterfall Model is at hand:
The V Model A further development of the waterfall model issued into the so called "V-Model". If you look at it closely the individual steps of the process are almost the same as in the waterfall model. Therefore I will not describe the individual steps again, because the description of the waterfall steps may substitute this. However, there is on big difference. Instead of going down the waterfall in a linear way the process steps are bent upwards at the coding phase, to form the typical V shape. The reason for this is that for each of the design phases it was found that there is a counterpart in the testing phases which correlate to each other.  The time in which the V-model evolved was also the time in which software testing techniques were defined and various kinds of testing were clearly separated from each other. This new empasis on software testing (of course along with improvements and new techniques in requirements engineering and design) led to the evolution of the waterfall model into the V-model. The tests are derived directly from their design or requirements counterparts. This made it possible to verify each of the design steps individually due to this correlation. Another idea evolved which was the traceability down the left side of the V. This means that the requirements have to be traced into the design of the system, thus verifying that they are implemented completely and correctly. Another feature can be observed when you compare the waterfall model to the V-model. The "Operation & Maintenance" phase was replaced in later versions of the V-model with the validation of requirements. This means that not only the correct implementation of requirements has to be checked but also if the requirements are correct. In case there is the need of an update of the requirements and subsequently the design and coding, etc. there are two options. Either this has to be treated like in the waterfall model in a never ending maintenance phase, or in going over to another V-cycle. The earlier versions of V-models used the first option. For later versions a series of subsequent V-cycles was defined, as shown in the below diagram:  This idea also correlated with the established sample phases for products as it is present in many industries. One of the cascaded V-cycles became the V-cycle of a sample phase. In addition to this the V-cycles were tailored. This means that in earlier sample phases not all the intermediate work products and process step were established in their full beauty but it was simply reduced to what makes sense. By these measures the V-model became a usable process model. It does not consider every detail and possibility but evaluated over a multitude of projects in various industries it proved its usability. The Spiral Model The process begins at the center position. From there it moves clockwise in traversals. Each traversal of the spiral usually results in a deliverable. It is not clearly defined what this deliverable is. This changes from traversal to traversal. For example, the first traversals may result in a requirement specification. The second will result in a prototype, and the next one will result in another prototype or sample of a product, until the last traversal leads to a product which is suitable to be sold. Consequently the related activities and their documentation will also mature towards the outer traversals. E.g. a formal design and testing session would be placed into the last traversal.  There are different instances of this model so far I saw it with 3 to 6 task regions. The above picture shows 4 task regions. These regions are:
The advantages of the spiral model are that it reflects the development approach in many industries much better than the other process models do. It uses a stepwise approach which e.g. goes hand in hand with the habit of maintaining a number of hardware sample phases in cases where the product to be produced is not only software for a given environment, but also contains the development of hardware. This way the developers and the customer can understand and react much better to risks in the evolutionary process. By having an iterative process which reduces formalisms and omittable activities in the earlier phases the use of resources is optimized. Further, any risks should be detected much earlier than in other process models and measures can be taken to handle them. The disadvantages of the spiral model are that the risk assessment is rigidliy anchored in the process. First of all it demands risk-assessment expertise to perform this task and secondly in some cases the risk assessment may not be necessary in this detail. For completely new products the risk assessment makes sense. But I dare to say that the risks for programming yet another book keeping package are well known and do not need a big assessment phase. Also if you think of the multitude of carry over projects in many industries i.e. applying an already developed product to the needs of a new customer by small changes, the risks are not a subject generating big headaches. Generally speaking the spiral model is not much esteemed and not much used, although it has many advantaged and could have even more if the risk assessment phases would be tailored down to the necessary amount. |
|
|
