[pkg-ntp-maintainers] Bug#556465: Bug#556465: ntp: default config for pool servers should not use iburst

[Kurt Roeckx]

On Sun, Dec 27, 2009 at 11:34:58AM +0100, Attila Kinali wrote:
> On Sat, 26 Dec 2009 14:09:42 +0100
> Kurt Roeckx <kurt at roeckx.be> wrote:
> 
> > > Using iburst causes ntp to send a voley of packets to the ntp servers.
> > > While this is not a problem in a local network or when using an ntp
> > > server that doesn't serve a large amount of clients, it might swamp
> > > public servers from the ntp pool or debian pool (think about switching
> > > on of a lot of machines monday morning at 8).
> > > 
> > > Hence i suggest to stop using iburst in the default config for any
> > > pool server.
> > 
> > The documentation says:
> >    iburst
> >           When the server is unreachable, send a burst of eight packets
> >           instead of the usual one.
> [...]
> > So I see no reason to remove this.

You skipped the "is a recommended option" part.

> I think i've written above why this is a problem, but maybe it
> was not clear enough.
> 
> Normally, the number of ntp servers from a pool is considerably
> smaller than the number of clients using them. Probably in the
> range of one server to 10'000 clients or more.

I do know how the pool operates, I run a pool server myself.

> Now consider that most of these clients get switched on around
> the same time (when people start to work or when they get home).
> Traffic analysis shows that this is quite consistent per region
> and ususaly within 15min-30min for "got to work" switch-on and
> around 30min-60min for "got home" switch-on

I didn't analyze any of the traffic, but I think the peak from
people running something like ntpdate from cron is alot worse
than this.

Anyway, did someone every bring this up on the timekeepers list?
The pages do not mention the use of iburst.


Kurt