search for: unp_l

....00% kernel{thread taskq} ---- Most of the userland processes get stalled and if the sitation is not resolved, their network services stop responding alltogether after some time. This affects relayd and dovecot for sure, but not sshd. When processes are slow, they often block in states like '*unp_l' or 'pipewr' according to top, and overall cpu time spent in kernel grows by factors (and system load too). See [0], [1], [2] for top screenshots when things work badly. Under normal circumstances load is below 5 and systime below 5%. Load during daytime makes this situation more likely...