Thanks for the input.

I confess I have mixed feelings about continuing to run OET. I cannot say that I have read all relevant material regarding performance, efficiency and points; I have found the most relevant information on WCG is scattered across sometimes hundreds of forum posts. Part of why I enjoy crunching is that it allows me to accomplish something useful to humanity while living a schedule that leaves little time for extra research (and sometimes even sleep). So it is impossible to put in the necessary time to get a full understanding.

However, I've come to use the points given to tasks for any particular project (alongside SNURK's calculation of my Gflops) as an estimation of the amount of work that is being accomplished compared to power input. With OET1, even within a series like 333, the points being granted per hour of CPU time vary wildly. I am accustomed to the variability in points granted -- but usually within 10% or so, with occasional outliers. On some of the OET tasks, I've gotten really good points for the given machine -- often better than average for other projects. On some tasks on the same modern machine, the points per hour are less than on my oldest 2005-vintage machine. A variation of 70% or more in points boggles. And Snurk's estimation of my Gflops has dropped almost 10 points since I started crunching OET.

Examining the results for one of my machines, points granted for one ~11 hour CPU time task were half what they were for another ~11 hour task run the day before. On that same machine, a current task has run just over 12.5 hours of CPU time, and is just shy of 58% complete. As a comparison of VINA as a platform on that machine, FAHV jobs showing estimated computation size of 9159 GFLOPs consistently take just shy of an hour to run. For all four of the currently running OET1's from series 335, the estimation is 22312 GFLOPs, including the one that looks like it will run many more hours.

Compounding the problem is that, when a job runs that long, there are now hours between checkpoints that I understand are taken at specific percentage thresholds. Two of the machines are scheduled for a good de-dusting/cleaning this weekend, but I hate to take them down because, at the moment, on just one of them, it would represent 11 hours of lost work.