Done. I've tinkered with settings for sidekiq workers, web threads and database connections during the day. You noticed many 5 minute outtimes and several longer delays for incoming or outgoing toots. In the moment the setting allows for shorter queues than number of worker threads, with full memory utilization (considering disk cache as important) and one core having lots of 100% bursts while not being always at 100%.
A Mastodon instance aimed at (but not limited to) the cycling community.
Now memory and disk IO are close to the limit what's possible on this machine. So the move to the new server really is important.