Services and resource profiles as metrics for the allocation of IT infrastructure costs

This paper addresses the general problem of allocating costs for shared and distributed IT infrastructures to business entities, like cost centres or processes. In contrast to earlier times, when monolithic mainframes were the central resources in data centres, client/server architectures with distributed heterogeneous components now dominate. In such an environment, resource consumption (e.g. CPU time) usually cannot be fully apportioned to users or other business entities. We therefore propose an approach which is based on predetermined resource profiles as estimates for the mean resource consumption of business services. Instead of metering every component during operation and consolidating the log data afterwards, only service invocations need to be logged. For the determination of the resource profiles we developed a measurement methodology and a software toolkit. The approach was successfully tested and evaluated for OLTP systems at the BMW Group. Furthermore, by combining resource profiles with Queuing Network Theory, we could demonstrate their appropriateness for capacity planning.