The model proposed by Herodotou et al [7], described fully in [8], is to our knowledge the most usable and detailed model currently available. It accurately models the data-flow
between tasks and task phases, but its cost model does not distinguish between CPU and I/O costs.