It ought to be possible to improve the "time to completion" estimate. Currently, it seems to be over optimistic at the start, and then _increase_ consistently as work is done.

After about an hour of processing, the CPU time increases by one second per second as expected, but but the time to completion increases by two seconds for every elapsed second - I'll never be done!! ;-)

I've seen the same thing. The reason for this over-optimism is the fact that the first ~50% of every work unit goes by very quickly--less than 10 minutes on my 1 GHz Pentium 3. But around then, as you said, the "To Completion" counter begins to tick upward, rather than down as it should. The last ~50% of the work unit takes the majority of the processing time, all totaling about 8-9 hours/WU. This must be due to the way the WU is processed, but this fact renders the progress bar and "to completion" counter useless. Is there no other way by which BOINC/SETI can report its progress to more accurately reflect the total amount of raw processing it must perform on each work unit? Or perhaps, in the case of the "to completion" counter, a running average of the total time taken to complete all prior workunits could be kept, and used? In any case, there must be a better way to give these two features some accuracy.

In order to improve the information provided by the client, I have tried a very simple function. It provides an acceptable estimate of the total time, from wich it is easy to derive the progress percetage and time to completion. I have tried it on my PC, it seems to work. I have writed it in MATLAB. I'm not able to integrate this code into the client or to compile an executable reading the information from the state file, I whish someone can do it. Write to me if you can, I will send you the simple code. Mario