In This Section
4C Partners
'Why Cost Models are Risky' by Sean Barker, 4C Project Advisory Board
On the final day of the 4C Project, Advisory Board member Sean Barker reflects on some of the work from the last two years, particularly that on Cost Drivers, and how this relates to a practical application in different working environments...
Reading through the 4C deliverables from the past two years, I was struck by the comment that institutions find it easier to create their own cost models than to reuse existing ones. Then, listening to the discussions in our Edinburgh meeting, I noted one concern that seemed important enough to be given its own name: “Indirect economic determinants”. Taking the discussion to pieces on the train back, I realised that these indirect economic determinants1 where what, in Systems Engineering, would be called non-functional requirements – requirements defining the quality of service of the system.
The most obvious non-functional requirement is not to lose the data in the case of some disaster. Here, normal computer room practice is to make a backup copy in case the computer breaks down, and a second copy in a remote location in case the computer room catches fire. However, there remains the possibility that both computer rooms could be destroyed, so for vital records, to reduce the probability of loss to an acceptably low level, a third or even a fourth separate location might be needed. But duplicating facilities costs money, so most organizations will calculate a risk assessment—the probability of an event multiplied by the costs incurred if it happens—and compare the liability of the risk to the costs of mitigating it.
For example, the Aerospace sector has, over the last ten years or so, transitioned from designs based on drawing to designs based on computer models. These models are not only the basis of manufacture, but must also be available to the regulatory authorities for airworthiness certification, and may also be called on by the courts should an aircraft crash. Since the loss of the airworthiness certificate could ground scores of aircraft costing hundreds of millions of pounds, the risks of losing the data are substantial. Moreover, to be usable in a court, the data must also have “evidential weight” provided through detailed procedures and audit trails. Although the business function required—keeping the data for seventy years—remains the same, the non functional business requirements are for a very low chance of losing data and for evidential weight. The naive model of retention costs—volume of data times the period of retention—does not account for the costs of providing access to the authorities or audit trails for lawyers.
The hypothesis here is that meeting non-functional requirements changes the cost model needed because it must include additional cost factors, such as repository replication or building audit trails. This is illustrated in the figure as a family of different cost models. As the importance of non-functional requirements rises, it is not simply that the cost of retention increases, but rather the predictions of cost need to jump to the curve for a different model, one which includes the required risk mitigations.
The reason it has proved difficult to reuse cost models is less to do with measuring the various cost factors that the model makes explicit – although accountancy practices also vary - and more to do with the non-functional requirements for risk reduction. When developing a cost model, a memory institution will focus on what it identified as the functional cost drivers and may fail to notice that these make assumptions about the risk reduction measures which are written into the DNA of the institution.
Moreover, memory institutions such as libraries are typically vertically integrated, providing all the functions needed for retention. The RASSC project developed stack model for for the services needed for retention. For example, physical data storage can be proved by a specialist IT company – there are already several in the market – where the cost model is based on volumes and integrity of data retention. But with a CAD model, the software used to read it can make mistakes, so CAD models need information level validation, such as checking the volume of the model is unchanged. These require additional specialist services on top of physical storage such as developing validation test. Some of these services will be used episodically, e.g. when a new CAD system is introduced or the file format standards are updated. Consequently, for CAD models, the information level services will have a quite different cost model to physical storage.
It seems likely that the service sack approach could be used to build up a set of cost model components covering both the functional elements already studied and at least the common risk mitigations needed to meet non-functional requirements. Obviously, the 4C results will provide the foundations for elaborating this more complex approach, although it seems there is more to do to develop service stacks and their cost models.
For a more detailed development of the argument, see Cost Models for Non-Functional Qualities.
Sean Barker, 4C Advisory Board
Currently working for BAE SYSTEMS Advanced Technology Centre, UK, Sean represents 'industry' on the 4C Project Advisory Board. He is currently working on a range of projects including: Product Lifecycle Support (PLCS) ISO and OASIS-open; Concurrent Engineering - Data Clouds - Product data repositories - Long term model retention. His publications cover - Computational geometry, using topology based on quantization of fuzzy logic; Building STEP models from legacy systems; Design Definition Management; Dimensions of Integration; Editor for various ISO 10303 parts, including Information Rights, Issue, ... ; Long term/through life data retention/Data repositories.
1 See D4.1—A prioritised assessment of the indirect economic determinants of digital curation
