MADM - MultiAttribute Dependability Model

The MADM was developed by the University of Maryland and the Fraunhofer Center Maryland.
Model Characteristic			Value
Dependability attribute			Dependability - Unified model merging multiple attributes
Object			Product
Purpose			Measure dependability of a product. Predict dependability of a modified product
Dependability perspective			Any stakeholder with an interest in the dependability of a certain software system or product being developed. (May also be applicable to hardware products)
Assessment technique			Model-based analytical
Measurement model
		Type	Quantitative
		Inputs	Need a set of attributes and a quantitative method for computing the value of each attribute present or required to be present in a system or product.
			Need the relative importance of each attribute compared to the other attributes (weighting factors)
			Need the utility function that converts the computed value of an attribute into  a per cent usability number (from 0% to 100%) to normalize all the attributes
		Outputs	Relative dependability of a system or of the requirements for a system.
		Model description	The basis of the model is to provide a negotiation strategy for a set of stakeholders to arrive at a consensus:
			The attribute values are converted to utilities and the weighted sum of the attribute values is computed to produce a dependability utility.
			The required utility for each stakeholder is compared to the computed utility of the system
			If adequate, the system is deemed dependable. If not adequate, a negotiation strategy is employed to either modify the required dependability needs of the stakeholders or of the need to increase system dependability. The process is then repeated after modifying the system to compute a new system dependability.
Activity and artifact			During specifications to determine which development strategy is most likely to succeed and during testing to see if system has achieved the required dependability.
Maturity			Many models for each attribute have been around for many years. The union of multiple attributes into a unified model of dependability is new. Several small laboratory experimental uses of the model have been made. Needs to be tested on a large-scale industrial project for validation and usefulness.
Accuracy			Unknown. Need to calibrate the model and test it for accuracy.
Cost			Unknown, but should be relatively low during specifications phase. Costs can range from low to very high depending upon what data is already being captured by the system as it is developed and tested.
Known limitations, Problems			Probably more accurate in a product line environment or in a maintenance upgrade than in a new system. Very hard to compute the required utility of a new system when there is no system to base comparisons on.
			For effectiveness, need to merge MADM with an ROI process so that the cost of making dependability improvements can be factored into the decision process. Currently working on an economic process. May also look at models like iDAVE from USC as a mechanism to compute costs of dependability improvement.