Hi Henk,

Henk P. Penning wrote:
On Wed, 14 Sep 2005, Heiko Gerstung wrote:

  There is also 'burst' (does a burst (as iburst) on every query) ;
  that works very well because the first packets clear the way
  for the rest through caches, switches, routers etc ;
  we use it to peer a couple stratum-2's in our org.
I knew that someone would come back with burst :-)

  I would love to see simple, strict rate controls, but there
  are legitimate reasons for a client to send a lot of packets.

  It seems you can only set simple, strict limits if there
  is also a mechanism to exclude certain hosts (friends)
  or situations (say, the first 100 packets). Yet another
  feature.

We would have to find an algorithm which allows bursts, iburst and maybe exclude specific clients (yet another "restrict" parameter), e.g. monitor the rate of requests in a given time frame (x requests in y seconds are OK).

I have to take a look into the statistical data collection routines of ntpd to find out if at least the necessary data can be achieved easily.

Another point would be to check how to "drop" a request as fast and efficient as possible, but I guess that would be the easy part.

The main point is whether you/we think that dropping requests of "hammering" clients would be a desirable way of getting rid of them or not.

Implementing this into a NTP server application is AFAICS technically possible, the question is if it will help or not.

Kind regards,
Heiko


--
------------------------------------------------------------------------

*MEINBERG Funkuhren*
Auf der Landwehr 22
D-31812 Bad Pyrmont, Germany
Tel.: ++49 (0)5281 9309-25
Fax: ++49 (0)5281 9309-30
eMail: [EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>
Internet: www.meinberg.de <http://www.meinberg.de/>

------------------------------------------------------------------------

Meinberg radio clocks: 25 years of accurate time worldwide

_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers

Reply via email to