flush_zone service impact

[Daisuke HIGASHI]

Hi,

Rauno Tuul via Unbound-users <unbound-users at nlnetlabs.nl>:

>  msg-cache-slabs: 4
>  rrset-cache-slabs: 4
>  infra-cache-slabs: 4
>  key-cache-slabs: 4

Try large *-slabs value (e.g. 1024).

My guess is that long outage comes from lock contention between cache
flushing thread (thread0) and working threads (other than thread0). If
a slab is very big due to very large cache, thread0 holds spin lock of
the slab for long time during cache flush.
Large number of (small) slabs will reduce each slabs’ lock time by
thread0 and also reduces probability of lock contention. It may
introduce some memory overhead but would not be problematic since you
have much memory.

Regards,
-- 
 Daisuke HIgashi