Almost all the BOINC projects i've run have been CPU and/or GPU intensive. I've run into a couple new (to me) projects that appear (vmstat) to be heavily I/O intensive. I have three questions.

1. Does WUProp only count CPU time? (I assume so). So these jobs that take ten hours of wall clock might get 6 hours of credit? I mean, that's fine, i just want to know for sure what is being measured. If it were wall-clock time, that would be different than CPU time. If i could somehow get both, then i'd know how much time i'm waiting on disk, and it wouldn't matter if that was magnets or SSD or how fast the interface is or how many devices. I might be able to convince BOINC of my idea on how to maximize throughput.

2. I'm seeing idle time in the 20-60% range. "top" says it's not swapping. Could i tell BOINC to run 2 cores of CPU intensive apps along with 4 that are more disk intensive? That's 6 apps for a 4 core system. I mean, it might be done with XML edits of some kind. I'll have to find where the Linux install decided to put the BOINC files. 400k files on the root partition. Can't be that hard.

3. This system has 7 GB free, under normal conditions. Could i set up a 6 GB RAM disk and get BOINC to use that? Then everything would count as CPU usage. A startup script would copy the entire BOINC tree to RAM. There wouldn't be any check pointing for power failures, unless i get a UPS system (budget looks OK), and script power down of the mains to copy BOINC data back to spinning magnets before system shut down. Easy, right? Of course UPS systems' aren't advertised with energy listed in watt hours, and instead use VA - which continues to not be a unit of energy, so if you know that your system consumes 260 watts all the time (because you measured it), and you want to know how long you have to shut down, well, there's a black box calculator that doesn't let you plug in 260 watts, so it gives you answers with zero confidence, even if the answer looks like three minutes. Well, the write back to disk should take 2 minutes, maybe. That could be benchmarked. I understand car batteries, maybe i should build my own UPS system from scratch, with like a pi zero, some custom sensors, and relays, and and and what? vacuum tubes? Jeez. Maybe UPS makers have PDFs about the products with real numbers somewhere on the interwebby thing. It would be so much easier if disk intensive apps could be configured to use RAM directly. So far, another 16 GB or 32 GB RAM looks cheaper than a UPS system, as long as the app is checkpointing anyway. This is a question if anyone has done something this crazy. But maybe it's hardly even a comments-accepted thingy.