[ntp:questions] NTP as HA-cluster service

Todd Glassey CISM CIFI tglassey at earthlink.net
Fri Jun 5 15:03:19 UTC 2009


David Woolley wrote:
> James wrote:
>   
>> I have NTP setup as a heartbeat service in a two-node cluster.  If I
>> was to offer this service to the public pool, would there be an issue
>> with hosts using it when the service failed over to the other node?  I
>>     
>
> NTP should not be failed over.  It should be provided on addresses that 
> correspond to specific physical machines.  Failover for ntpd should be 
> done by having the client have both machines as servers.
>   
More than that - load balancing NTP through all traditional methods 
involves rewriting the headers which means that the chain of  custody is 
altered and very possibly destroyed since the latency and asymmetric 
loading of the LB solution adds to the non-deterministic slop ntp 
already has to live with.

So the answer is dont apply TCP availability solutions to UDP real time 
based processes if you want them to work in a proper manner.

Todd Glassey
>   
>> ask because currently I don't have a DRBD or network filesystem for
>> NTP, I have it configured identically on each node and the service
>> just runs on one of the nodes at at time (with a virtual IP). The node
>> that isn't the server just ntpdate's itself against the virtual IP
>> every so often.
>>     
>
> _______________________________________________
> questions mailing list
> questions at lists.ntp.org
> https://lists.ntp.org/mailman/listinfo/questions
> ------------------------------------------------------------------------
>
>
> No virus found in this incoming message.
> Checked by AVG - www.avg.com 
> Version: 8.5.339 / Virus Database: 270.12.53/2154 - Release Date: 06/04/09 05:53:00
>
>   




More information about the questions mailing list