XNTPD - Broadcast vs. Polling?

[Rick Tweedy]

Hi All,

I was wondering if anyone had information or could point me to sources where I might find details on the difference between a polling configuration and a bradcast situation?

From what I have read here in various messages it seems as though broadcasting is preferred although I am not sure why? My concern is that in our Data Warehouse environment every bit on the pipe is accountable for. I do not want to be adding any additional strain on the net caused by broadcasting time.

Thanks in advance for anything you have to add.

Rick Tweedy

[Rick Garland]

The overhead for broadcasting is not much. More info can be obtained from
http://www.eecis.udel.edu/~ntp/ntpfaq.

[Kofi ARTHIABAH · Accepted solution]

Hi Rick:

hmmm.. I would go with polling because you have better control as to when the client has to sync up ( so for example, you can configure your client sync up to raise alerts if it has been unable to contact any of the xntpd's in a specified period (you might want to know about this) ) ... plus, like you rightly mentioned, there would be some extra network traffic (not very significant) but its there.

configure your clients to run ntpdate from the cron at regular intervals - you can make it fancy by wrapping the ntpdate call around a script that does checking for success or failure of the ntpdate and send out an alert if required.

cheers.

[Anthony deRito]

Rick, NTP Broadcasting is actually a way to REDUCE network traffic. When one server is configured to send out regular time broadcasts (instead of having all the NTP clients polling the server for its time) the network traffic is reduced. Read the SAM help doco for references to this.

You would use this in the case when many clients require NTP info. If you only have a few, it is best to use polling because you have more control by using peer polling techniques.

Tony

[Rick Tweedy]

Hi All,

Thanx for all the info! It seems to me that broadcasting would be the best solution. It is obvious(now hahaha) that the traffic would be must small than polling from multiple machines.

It do not really understand why the concern over control leads to polling being preferencial? The only thing that I could think of is that you may not want the clock change while a oracle(for example) is up. In theory the time change could corrupt the DB but I would think that this sort of thing has been accounted for. I know that if there is too much difference in the clocks the update wont happen and I assume that if the update does happen that the time diff is small and things like oracle will componsate?

Thanks again for the help. Let me know if any of my assumptions are wrong above. Off to hand out sum points!

Cheers