bind dies randomly

[Greg]

I have 4 Redhat 8 Linux servers performing DNS (Bind 9.2.1-9 on all
4).  Two of them are outside (1 being the master), and the other two
are inside (all inside clients use one of the internal DNS servers as
their primary).  This works quite well except for one thing...  At
random intervals the one or both of the internal DNS servers (slaves)
will stop answering DNS requests altogether (this can occur after a
few hours or up to a week or two).  Simply restarting the DNS service
on these machines will clear up the issue.  There are no errors
whatsoever in the logs (both are logging DNS at "debug" and the named
log has no errors.  Any ideas on what causes this?