Then we shall wait until the technology catches up with reality.

Obviously this one computer is messed up as my NRW WU are coming in with an estimate of 17+ hours. Just have to wait until the system sorts itself out.

yes BOINC dosnt estimate well

again the CEP units seem to run a tad long sometimes

and yes this definatly shifts my computer to high priority

As ever, well observed, Rickjb.

Once started, the progress is very predictable. I don't have all the details about CEP work unit generation, but I would expect each batch to contain similar sized molecules, created combinatorially. It's not like the medical projects where they use a library of drug-like compounds. The batches must be as homogeneous as possible, otherwise the WCG size estimation algorithm has no hope of success.

As I said earlier, the false positives in validation are skewing the initial estimates.

One solution that springs to mind, given the small size of checkpoint files, is to create a portmanteau work unit from a dozen parts of other work units. Send the checkpoint files as input files, and return checkpoint files as results. Since each started molecule would have a reliable time estimate, it would be possible to construct work units with predictable sizes. It would even be possible to construct differently sized work units on the fly.

The downside? This is a very technical, very complicated solution. WCG don't have time to implement it, and there may be catches I haven't considered, such as the increased download sizes. Also, generating work based on recently returned results has hazards of its own.