This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: CS4 Modelling and Simulation LN-16 16 Parameterisation and Workload Characterisation So far in this course we have been concentrating on constructing a good representation of the system, appropriate for the investigation we wish to carry out. However it is not just the physical system which is represented within the model but also the effect of the environment upon the system. The most obvious inﬂuence of the environment upon the system is to produce work for the system to do. Therefore it is just as important to have a good representation of the workload of the system as to have a good representation of the system itself. Indeed performance measures are not, in general, absolute for a given system: performance is predicted on the basis of a given workload. This was reﬂected in lecture note 14 when validation of the input parameters was given equal importance with validation of the assumptions used within the behaviour of the model. Finding suitable and/or realistic values for input parameters is often termed workload characterisation . There have been well-documented problems arising from badly parameterised models. The Hubble Space Telescope was rigorously simulated prior to its launch, as a cheaper alternative to testing. However, once positioned it was found that it could not produce a sharp image because the main mirror was badly ﬂawed. The error was traced to the erroneous input data used in the simulations. Some corrective optics have now been added to the telescope in space (at a cost much higher than ground-based testing of the mirror), but the telescope will never perform as well as planned. 16.1 Parameterisation Parameterisation is the process of assigning values to variables within a model in order to ensure that it is as accurate as feasible or appropriate. In a simulation model as well as the actual values to be used, there will be some consideration of the appropriate distribution to be used for generating the values. It is important that we consider the availability of data to assist in parameterising our models from the outset of model design and construction. For example, the level of abstraction at which work is represented must correspond to the level of abstraction at which the system is represented; i.e. since values must eventually be assigned to all input parameters, it would be a waste of effort to develop a very detailed model if there is not data available from which to construct a similarly detailed workload characterisation. Conversely, there is no need to invest a lot of effort in sophisticated statistical analysis of workload data to parameterise a crude model which is only intended to give rough estimates. The first step is to choose which parameters to include in the model. As usual this will be inﬂuenced by the objectives of the study, and by consideration of what is likely to affect performance. If we examine the workload of a system it is likely to have many different characteristics, such as inter-arrival time of jobs, type of job, resource required...
View Full Document
- Spring '10