ARP checkpoints every 12.5% of progress.
If you have/want to reboot without much loss suspend ARP-tasks not started yet and the running ones at 12.5%, 25%, 37.5%, 50%, 62.5%, 75% or 87.5% of progress or wait until they have finished.

Bottom line: I lose the 2 hours. Maybe the Berkley people should give the users more control over everything. Not much I can do about that.

Crystal Pellet: what you describe sounds like the default percentages... and where could I change these? I would have to guess when a checkpoint occurs or finished so as to not lose too much.

lavaflow: Don't use BOINCTasks and you lost me with LIAM. Looked at BOINCTasks. Not for me. I only have 4 IPs in the basement. Takes 30 seconds to check.

Read about the alert rule thread. It presumes you are actually watching all this or the machine you are on is crunching during idle time. The machine I am on does not crunch. It only does BOINCmanager when and if I start it to "look in" on the crunchers in the basement.

In my case I may lose more since I don't check on all this all the time. It actually happened when I tried to micro-manage to get an ARP to start before other tasks by suspending them. Forgot about it. This is why I got into the RPC thing.

But thanks for the replies and ideas.

Yes I am watching 1 machine (Xubuntu) to do some debugging which is how this thread started. It hangs at times so I want to go to 20.04 LTS and latest boinc then see. Every time I check it is ARPing.

My only Win10 is a lowly Duo Core Centrino and I disabled auto booting after updates and no ARP. Hmmm just came to me. Change that Xubuntu to the same profile.

So I have my answer.

Can't make checkpoint intervals shorter, but you can make them longer with the Write to Disk control. I've set it to 600 seconds at most. Would you run 16-32 and more threaded machines, the writing with 60 seconds default and short checkpoint interval projects would be kind of perpetual. Not want to know what that would do to an SSD or USB drive, but a shortened lifespan might be in the cards.

I had a closer look at that a while ago. Sorry I don't recall all details now. An ARP task writes between 4 and 5 gigabytes to disk and that is because it re-writes the whole data at every checkpoint, mostly unmodified. That's 500 megabytes every time and even more at the end.
What you should be aware of is that OPN is even worse if you let it checkpoint at will. Recent observation showed it writing about 1 gigabyte per task (+/- 300 megabytes), and that's only one hour of work on my Ryzen 3600. That one could do 150 tasks every day. Many people seem to run OPN exclusively these days and at default settings it could be a killer.
I also tried reducing the checkpoint frequency through BOINC but found that this doesn't seem to be a hard limit. Sometimes tasks still write checkpoints when I don't expect them so I don't feel comfortable relying on that.