Posted on October 7, 2022

We will be upgrading the remaining systems including the physical hosts to the 6.0.0 kernel release. We’ve got it presently installed on about a dozen servers and so far only one expedited RCU CPU Stall and that occurred during boot-up on the physical server with the most guest machines. This can happen even with a good kernel because systemd’s parallelization results in large numbers of processes in run state simultaneously and this in turn can cause the RCU process not to get CPU in the 30ms allocated, but in the week we got only that one and no others where previous 6.0-rc5 and rc6 kernels generated 20-30 in the first five minutes and were basically unusable.

No single service should be out for more than about ten minutes EXCEPT for yacy which rebuilds it’s indexes every time it restarts and this process takes about half an hour.

The 6.0 kernel re-wrote the scheduler in a way that benefits multi-core systems. Although the changes were largely aimed at AMD’s Ryzen CPU’s, we say substantial improvements on our Intel based servers as well, particularly in the area of scheduling latency. It reduced the time it took our web server to load a WordPress page from about 280ms to 40ms, a seven fold improvement and the largest improvement I’ve seen from any Linux kernel upgrade since kernel .98.

This affects all Eskimo North services including shared web hosting, private virtual servers, shell servers, e-mail, and our free federated services friendica.eskimo.com/, hubzilla.eskimo.com/, nextcloud.eskimo.com/, and yacy.eskimo.com/. If you wish to support our free services, please consider signing up for a Linux shell or web hosting account at eskimo.com/services/free-trial…