Re: [Ntp] Leap smearing

[Hal Murray <[e-mail]>]

martin.burnicki=[e-mail] said:
> However, if you have "normal" clients and you just want to prevent your
> applications from hickups when the system time is stepped back, it's easier
> to let the server smear the time, because you only have to configure this
> once, and all clients behave the same. 

That seems appropriate when you have an existing installed base using an 
existing protocol.

But we are talking about a new protocol so there is no installed base.  It 
seems reasonable to me to move the smear decision to the client side.  But 
that forces the client to decide which means one more parameter for the admin 
to keep track of and possibly screw up.

If we decide to allow/encourage doing the smearing at the server, I think we 
should be sure to document that the reason is administrative convenience.  The 
client software should probably have a 3 way option: smeared, non-smeared, 
follow-servers.  The follow-servers option will have all the current problems 
if the selected servers don't agree.


-- 
These are my opinions.  I hate spam.