Bind max socket/query per IP

[Mark Andrews]

You really need to read up on queuing theory.  The fairest way to queue is to
have a single queue and to process off the end of that.  Unfortunately interfaces
don’t form a single queue, they form multiple queues.  This sort of behaviour
is expected with multiple queues.  The main address is the long queue and the
alias is the short queue.  Different clients get different response rates (unfair)
but the overall rate is the same.  This is the reason banks often have a single
queue that feeds multiple tellers.  That way every client gets fair treatment
w.r.t. the amount of time they wait.

Additionally there been bugs where the current client has not been replaced to
read the next request when named has had to fetch/forward packets which just
stalls the queue processing for those queues.  If you are not up to date you
may be running a version with those bugs and have a query pattern that triggers
them. There are no known instances of this class of bug in the current code.



> On 22 May 2019, at 6:31 am, Ict Security <ict.security.job at gmail.com> wrote:
> 
> Hi guys,
> 
> I am experiencing a very strange problem.
> 
> Under heavy load, Bind becomes extremely load above a certain number of Qps but, if i query an alias IP address (where normally queries don't arrive), Bind answers immediately.
> 
> I was wondering if there is a kind of limitation on a single IP address due to socket or something elss.
> 
> I have increased and tuned kernel parameters, but problem still persist.
> 
> Thank you, cheers!
> FC
> _______________________________________________
> Please visit https://lists.isc.org/mailman/listinfo/bind-users to unsubscribe from this list
> 
> bind-users mailing list
> bind-users at lists.isc.org
> https://lists.isc.org/mailman/listinfo/bind-users

-- 
Mark Andrews, ISC
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742              INTERNET: marka at isc.org