[ntp:questions] Re: NTP does not sync when using pool.ntp.org

Brad Knowles brad.knowles at skynet.be
Fri Oct 10 01:10:12 UTC 2003


At 12:20 AM +0000 2003/10/10, Dale Worley wrote:

>  It's not as bad as it looks -- if we list the same 100 servers for all
>  three lists, and can get the lists to be handed out at random, there
>  is a 97% chance of the three names resolving differently.

	Therein lies the rub -- "handed out at random".  There are many 
studies that can clearly show that many resolvers are so badly 
screwed up that it's impossible to have any kind of reasonable 
guarantee that this will happen.  Indeed, one resolver bug alone is 
enough to result in a.root-servers.net getting twice as much traffic 
as any other root nameserver (at the time), simply because that name 
occurs first in the list.

>  If we use *four* aliases, there is a 94% chance that they all return
>  different addresses, and a 99.93% chance that they will return at
>  least three different addresses.

	If you can guarantee that each alias will return the addresses in 
a different order, then having a larger number of smaller sets of 
addresses should be good.  However, that depends a great deal on 
Adrian's implementation.

>  The remaining problem is that if some intermediate cache hands out the
>  same addresses for all four names.  We can reduce that by putting the
>  addresses for each alias in different orders.  Then there is only a
>  problem if the intermediate cache sorts the addresses, which is
>  possible.  One possible way to avoid that problem would be to leave
>  one address off the second name, two off the third name, and three off
>  the fourth name, so that if the server is maintaining four counters to
>  cycle through the four (sorted) lists, the pointers can't stay
>  synchronized with each other for more than one cycle through the
>  lists.

	The problem is more complex than this.

	I've got alternative solutions I'm working on that I hope will be 
a better solution.  I hope.

-- 
Brad Knowles, <brad.knowles at skynet.be>

"They that can give up essential liberty to obtain a little temporary
safety deserve neither liberty nor safety."
     -Benjamin Franklin, Historical Review of Pennsylvania.

GCS/IT d+(-) s:+(++)>: a C++(+++)$ UMBSHI++++$ P+>++ L+ !E-(---) W+++(--) N+
!w--- O- M++ V PS++(+++) PE- Y+(++) PGP>+++ t+(+++) 5++(+++) X++(+++) R+(+++)
tv+(+++) b+(++++) DI+(++++) D+(++) G+(++++) e++>++++ h--- r---(+++)* z(+++)



More information about the questions mailing list