Common use of Modelling Clause in Contracts

Modelling. As already noted in this document, monitoring both the service and customer behaviour are very important in making sure both current service level agreements are kept to and understanding how additional customer demands can be met. Modelling goes hand in hand with monitoring. Monitoring data can be used to validate a system model, train a model and as input to a model that predicts future requirements. Modelling itself varies from a domain expert making estimates based on experience to pilot studies and prototypes. Models can be used to predict the resources required for an SLA and to predict the affect of a change in the system (hardware or software). There are a huge variety of modelling techniques available. The simplest models may just use trend analysis: taking the historical usage and extrapolating into the future. A domain expert can use this type of information and predict resource requirements for SLAs. Work along these lines was carried out in the SIMDAT project.22 Analytical models can be built to represent system behaviour using mathematical techniques such as queuing theory. Such models can be used to predict response time for instance. Data on expected customer and resource performance can be used to train models such as Bayesian belief networks and artificial neural networks. Stochastic models can then be built. The IRMOS project23 is using models such as these as well as finite state machines to predict resource requirements for SLAs. Finally, simulation modelling may be used to understand the effect of different customer workloads on a real or prototype system. Software can be used to simulate user behaviour (service requests etc), perhaps simulating high workloads not normally reached in day to day operation. In this way the behaviour of a system to workload can be accurately assessed.