>>/1610/
 >>/1611/
 >>/1612/
Yeah, the write is most likely temp storage. I've heard that in some situations, the actual amount written to disk is less since some temp-file magic means some of the shortest-lived data is purged from the write cache before it can actually be committed, but I don't have good numbers on the topic.

The PTR does a lot of database work. The no_db_temp_files parameter will reduce it by a good amount. Not sure how much, but it would reduce it.

Increasing db_transaction_commit_period may also reduce it, I think, since it will reduce the commit frequency (at the cost of increasing per-transaction size, which will strain your temp folder and slow processing time).

A user also wrote this document with a lot of related info, if it helps: https://hydrusnetwork.github.io/hydrus/Fixing_Hydrus_Random_Crashes_Under_Linux.html

I'm not a Linux expert myself, so I can't talk super confidently here. Let me know how you get on!