The cost of timing deltas can be measured with a better view of the changes made and the explanations behind these changes. There are costs associated with communication, storage and computation.
The size of the original replay_op structure is increased by a little under 6%. This is a conservative estimate, since repay_op contains pointers. Memories may be associated with them. This is permanent memory increase, and would also come about in the trace files. To rsd_node the increase in memory size of the basic node structure would be less than 1%. This additional memory is however only runtime memory. Increases to communication and storage costs seem more or less negligible. The communication of these extra fields is bundled with other communication, so the cost for communication does not increase dramatically. Computation overhead is negligible when compared to the compression algorithm, and is called exactly once for every compression performed.
Even when we consider internode compression, the communication, computation and storage costs are negligible. So the overhead of timing deltas feature to MPICTrace is forgivable.