Late Maintenance of Highly Complex Areas

After signing the contract, the supplier has to deliver, as discussed earlier. It may be the case that glue codes were developed by the supplier long ago, hence the supplier can extend the system, and the customer will use that system. This is theoretical. However, often it is seen that supplier is unable to fulfill its promise.

Common difficulty areas are performance (response time) and integration with other products. For example, the supplier's current installations have about 30 customers and work well with them. But when faced with a planned 2000 customers, the response time skyrockets. Or, the connection to System X seemed easy, but in practice it did not work as expected. Most customers did not care about this before. What is the problem, they ask? If the supplier does not deliver as promised, he does not get his money; he also has to pay a fine. These customers think the supplier is just lazy and the money will cure him. Any problems may not be solved by the supplier. Such projects can drag on for years. The customer never gets the appropriate system and the supplier loses money.

The main issue faced by parties is the high chances of being late. The supplier is inclined to deliver the easy part first while delaying more difficult one. Ideally, it is the customer’s responsibility is to demand a guarantee for complex areas before selecting a component among those available. However, it is very costly and not realistic.

Proponents can do so during the proposal.

One can wonder now - what are the factors that satisfy the requirements for proper integration? The answer is -

for harmonious CBS proper integration is needed. The system does not work correctly or is incomplete without the right components.

What are the necessary major factors for maintenance of the systems implemented outside?

Software management is known as one of the most expensive resources in the software development process. Very soon, the development phase of software will be replaced by integration in CBD.

There is a need for maintenance at every phase of software. Hence, software developers need support for every sustainability perspective, and for that support, they need to understand sustainability and its impact on CBS. In particular, software consistency with COTS is different from legacy software management, with the possibility of managing operations other than by the developer. Whether this component is an “internally” implemented one or is purely outsourced.

The importance of discussing COTS selection and integration show up when considering that COTS products are developed to be generic, however they can be integrated into a system and are used in a specific context with certain dependencies. COTS components may be heterogeneous in nature, for example, they may have different interfaces, support different business protocols, and use different data formats and semantics. Hence, these mismatches should be eliminated before integrating COTS components into a CBS.

Differences between COTS products and system integration exist due to their architectural differences and constraints. These mismatches should be removed at the time of integration. Thus, a classification of mismatches or incompatibilities can be useful for COTS integration.

There is a significant amount of effort that is invested in developing a system by integrating these products. There are so many existing approaches for calculating the efforts invested in integrating the components.

Nowadays there are many studies using neural networks for predicting software quality matrices. Heiat (2002) compares prediction performance of neural networks and radial basis function neural networks through regression analysis. This method shows that neural network methods improve mean absolute percentage errors as compared to regression analysis.

Reddy et al. (2008) combined multilayer perceptron networks with COCOMO II for software cost estimation. The proposed method is evaluated on 63 COCOMO II projects. The main aim of this proposed method is to train multilayer perceptron in better way. In this proposed model, in intermediate layer efforts factors and scale factors are detailed and these factors are evaluated by COCOMO II. The Sigmoid Function was used as the activation function for the intermediate layer. An algorithm is used for teaching backpropagation. The data training phase is 80% and testing phases is 60%. The results suggest that the standard error of relative error in the composition model is less than the COCOMO II model.

We proposed another neural network-based approach. In this approach we define three factors which affect the Component Integration Efforts to a great extent. These factors are:

  • (1) Interaction Complexity.
  • (2) Understanding.
  • (3) Component Quality.
< Prev   CONTENTS   Source   Next >