I had problems with computation errors on MCM WUs described in a separate thread. This gave me cause to watch what was happening a little more pro-actively. I noticed on another machine I'm running that I got an error message on another MCM WU saying it had not marked the WU as 'finished'.

I decided to leave the project and updated my project preferences on WCG. Despite leaving the project days after leaving, WCG kept sending my machines MCM WUs with short deadlines, effectively blocking that machine from doing any computations other than MCM (or at the very least prioritising MCM WUs). This meant that my GPU processing capacity was not being fully utilised because the GPUGrid WUs were requiring just under one of the CPUs to run the GPU WU, but BOINC was giving all resources to MCM WUs because of their tight deadlines. It felt like my machine had been hijacked by MCM WUs. I presume WCG continued sending me MCM WUs despite leaving the project because they had short deadlines. This, if true, if not in accordance with what I want. Anyway, I aborted all but the running MCM WUs and told WCG that I would not accept WUs from any projects except those I indicated. That seems to have stopped MCM WUs being sent to me.

I am more than a little disappointed to have made this discovery. It seems to me that by a project giving tight deadlines both WCG and BOINC will prioritise that project at the expense of others and circumvent the expressed desires of the volunteers. It looks like processing priority may be had by a project taking advantage of a weakness in WCG and BOINC scheduling algorithms. Giving tight deadlines whether conscious or not is not in the same spirit as my volunteering.