On Mon, Jun 25, 2012 at 6:29 PM, Dave Hart <[email protected]> wrote: > Yes. iburst only engages when a server reach goes from 0 to nonzero. > burst applies every poll to which iburst doesn't. burst in a default > config would be evil. iburst just makes sense IMO.
As others have mentioned (and the docs agree), iburst causes a six-packet burst after a server comes back "online" after being unreachable, and not just when the client starts. This could be a very bad thing. Basically, you're talking about a 6x transient packet load and bandwidth increase on a server that may have just come back up after maintenance and may still be busy doing lots of other startup tasks. For a busy pool server with thousands of clients that could be ugly. I suppose you could just keep a busy server down for >2048 seconds every time you do maintenance in hopes that all clients would back off to randomized max-poll intervals. But it seems like an operational hardship. -- RPM _______________________________________________ pool mailing list [email protected] http://lists.ntp.org/listinfo/pool
