21 Apr
2014
21 Apr
'14
8:50 p.m.
Am 21.04.2014 18:56, schrieb Leonid Isaev:
On Thu, 17 Apr 2014 21:31:07 +0200 AFAIU, there are 2 real issues here: 1. We hook to the boot process a bunch of disk-intensive operations which did not belong there in the 1st place. 2. Even if a boot delay for timers is implemented or the behavior of Type=idle units is "fixed" somehow in systemd, still all "cron" timers will be started in parallel which may result in a slow/unresponsive system. Note, that under anacron they were serialized by run-parts.
The Linux scheduler and I/O scheduler are supposed to handle such workloads gracefully. The units are configured to be in the proper io scheduling class and with proper nice values.