Just a follow up, after this was fixed and the machine caught up with the backlog, it is now running with a load of about five, and this is with a friendica, hubzilla, mastodon, yacy, and roundcube web mail. So on average now 13 cores are completely idle. There are other things I will be adding to this but it just goes to show how important identifying and tuning bottlenecks is. I thought perhaps it might be a good idea to illicit how I found this. I noticed with a ps aux that many processes were sitting in state Sl which means it's a multi-threaded process in interruptible sleep. I wanted to find out WHAT it was waiting on, so I did strace -p pid to trace one of the sleeping processes, I found it sleeping on trying to open a socket that belonged to mariadb. I knew from previous testing that mariadb was capable of around 13,000 transactions per second and it was only doing slightly over 1000, so this lead me to this kernel variable being the issue.